JPH10198703A - Data base system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform a flexible retrieval with a high degree of freedom according to the intension of each user. SOLUTION: A node assembly composed of plural nodes A-K is defined, and a positive link showing positive relation or a negative link showing negative relation between the respective nodes is defined. This node/link assembly is placed at the front end of data base so as to provide correspondent data when any arbitrary node is designated. A keyword is defined for each node. When an operator applies a positive retrieving instruction together with the prescribed keyword, the node connected through the positive link to the node corresponding to the applied keyword is retrieved and when a negative retrieving instruction is applied, the node connected through the negative link is retrieved. Any one of retrieved nodes is selected and correspondent data are provided. The node/ link assembly changes by learning.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database system. The present invention relates to a database system having a function of providing a database.

[0002]

2. Description of the Related Art In the modern information society, the role played by database systems is becoming increasingly important.
Database systems have been built for information in all fields. In order to construct a high-quality database system, it is important to record high-quality data, but it is even more important to provide a high-quality search environment. No matter how high-quality the database is, if the data required by the user cannot be found, no purpose can be achieved.

[0003] Search operations in a database system are usually performed using keywords. If the administrator of the database system associates a certain keyword with each data in advance, when the user inputs a specific keyword, the corresponding data can be presented.

[0004] Recently, as seen in the spread of the Internet, it has become relatively easy to access a computer installed in a remote place. On a global scale. For this reason, it has become possible to construct a global database system by connecting local databases constructed in individual regions to one another via communication lines. In such a distributed database system, a search method using a node aggregate has been proposed in order to provide a high-quality search environment. That is, a node aggregate including a plurality of nodes is defined, and predetermined data is associated with each node. Then, when a specific node of interest is designated from this node aggregate, data associated with this node of interest is provided.
If a keyword is given to each individual node, basically, a data search based on the keyword is performed. However, if each node is associated with a predetermined link aggregate, another keyword (node) associated with the link aggregate can be found when a certain keyword (node) is specified. Will extend beyond the local database to the global database associated with the link aggregate.

[0005]

The search processing in the conventional database system can be basically called a processing of inputting a predetermined keyword and finding out a matter related to the keyword. For example, a search process in a system using a node aggregate can be said to be a process of finding another related node by using a predefined link between related nodes. However,
A search process that finds "items related to a certain keyword" (here, called a positive search process) cannot perform a flexible search, and conversely, a search process that finds "items not related to a certain keyword" In some cases, it is more efficient to execute (here, called a negative search process).

In a system using a node aggregate, an administrator sets an environment in advance, and a general user performs a search operation or the like under this environment. The environment setting by the administrator is usually made in consideration of the convenience of the standard user, and thus does not always meet the needs of the individual users. For example, what kind of keyword to associate with certain data is determined by the administrator who sets up the environment, and it is not necessarily the case that an appropriate keyword is associated with a specific user. Not exclusively.

Accordingly, an object of the present invention is to provide a database system capable of performing a flexible search with a high degree of freedom according to the intention of each user.

[0008]

[Means for Solving the Problems]

(1) The first aspect of the present invention defines a node aggregate including a plurality of nodes, associates each node with predetermined data, and, when a specific node is designated, corresponds to this node. In a database system having a function of providing attached data, data storage means for storing data to be provided in association with a node,
Link storage means for storing a link aggregate consisting of a set of links indicating relationships between nodes; target node setting means for setting a specific target node based on an instruction from an operator; Search means for extracting, as a negative related set, a complement set of a set of related nodes related to a target node under a predetermined condition; and selecting all or a part of the nodes belonging to the negative related set as candidate nodes. From the candidate presenting means for presenting to the operator, the candidate adopting means for allowing the operator to adopt a specific candidate node as the adopted node from among the candidate nodes presented by the candidate presenting means, and the data storing means corresponding to the adopted node. Data providing means for extracting attached data and providing the extracted data to an operator.

(2) The second aspect of the present invention is the above-mentioned first aspect.
In the database system according to the aspect, further provided is a population defining means for defining a population indicating a specific node aggregation, and the candidate presenting means includes only nodes belonging to a logical intersection of the negative association set and the population. Is presented as a candidate node.

(3) A third aspect of the present invention is the above-mentioned first aspect.
In the database system according to the aspect, further provided is a learning unit that executes a learning process for defining a negative association between the target node and the adopted node, and the search unit determines a negative related set for the specific target node. At the time of extraction, if there is a node in which a negative association with the node of interest is defined by the past learning process, the node is extracted as a member of a negative association set, and the candidate presentation means
The order of priority when presenting candidate nodes is determined in consideration of negative relevance.

(4) According to a fourth aspect of the present invention, a node aggregate including a plurality of nodes is defined, predetermined data is associated with each node, and when a specific node is specified, In a database system having a function of providing data associated with nodes, data storage means for storing data to be provided in association with the nodes, a positive link indicating a positive relationship between the nodes, and a Link storage means for storing a link aggregate consisting of a set of negative links indicating a negative association with the target node setting means for setting a specific target node based on an instruction from an operator; Then, a positive related node that has a positive relationship with the target node under a predetermined condition is searched, and a set of the positive related nodes is extracted as a positive related set. Using a search process and a link aggregate, a negative related node having a negative association with a target node is searched under a predetermined condition, and the set of negative related nodes is set as a negative related set. Negative search processing to be extracted, search means for performing, a candidate presenting means for presenting to the operator all or a part of nodes belonging to the positive related set or the negative related set as candidate nodes, and presenting by the candidate presenting means Candidate selecting means for allowing the operator to select a specific candidate node as the selected node from the candidate nodes which have been selected, and data for extracting data associated with the selected node from the data storage means and providing the extracted data to the operator. Providing means.

(5) The fifth aspect of the present invention relates to the above-described fourth aspect.
In the database system according to the aspect, an updating unit for updating the setting of the target node setting unit is further provided so that the selected node becomes a new target node. Thus, either positive search processing or negative search processing can be specified.

(6) The sixth aspect of the present invention is the above-mentioned fifth aspect.
In the database system according to the aspect, further provided is a set defining means for defining a set indicating a specific node set, and the candidate presenting means selects only nodes belonging to a logical product set of the related set and the set. As a set of nodes.
It is redefined as a new population.

(7) The seventh aspect of the present invention is directed to the above-mentioned fourth aspect.
In the database system according to the aspect, a positive signal transmission coefficient is defined for the positive link, a negative signal transmission coefficient is defined for the negative link, and when the search means performs a positive search process, A signal having a positive signal value is transmitted from the node of interest to another node along the link, and the value obtained by multiplying the signal value before transmission by the signal transmission coefficient defined in the transmission path is defined as the signal value after transmission. Then, a process for extracting a node from which a signal having a signal value equal to or higher than a predetermined threshold value is obtained as a positive related node is performed. The signal transmitted from the node of interest to another node along the link is defined by multiplying the signal value before transmission by the signal transmission coefficient defined in the transmission path as the signal value after transmission. If a signal with a signal value Nodes is obtained to perform the process of extracting as negatively associated node.

(8) The eighth aspect of the present invention is the above-mentioned seventh aspect.
In the database system according to the aspect, the predetermined threshold value when performing the positive search processing is set to a positive value, and the predetermined threshold value when performing the negative search processing is set to zero. Things.

(9) The ninth aspect of the present invention is the above-described seventh aspect.
In the database system according to the above aspect, based on the adopted node adopted by the operator, a learning means for correcting the signal transmission coefficient of the link or adding a new link to the link aggregate is further provided. is there.

(10) In a tenth aspect of the present invention, in the database system according to the ninth aspect described above, the learning means selects all paths from the target node to the positive related node or the negative related node as learning targets. If there is a path from the node of interest to the adopted node, the absolute value of the signal transmission coefficient for the link on this path is defined with respect to the signal transmission coefficient of another link on the learning target path. The correction is made to increase relatively.

(11) According to an eleventh aspect of the present invention, in the database system according to the ninth aspect described above, when a node that has not transmitted a signal due to a negative search process is adopted, the learning means sets the node of interest to A new negative link is newly defined between the link node and the adopted node, and processing for adding the negative link to the link aggregate in the link storage means is performed.

(12) According to a twelfth aspect of the present invention, in the database system according to the ninth aspect described above, a frequency coefficient storage means for storing a frequency coefficient indicating an adoption frequency for each node is further provided, and a learning means is provided. Performs a correction to increase the frequency coefficient for the selected node relatively to the frequency coefficient for the other nodes, and the candidate presenting means outputs the signal value of the signal transmitted to each candidate node, The priority order is defined for each candidate node according to both the frequency coefficient of the candidate node and the candidate node is presented based on the priority order.

(13) In a thirteenth aspect of the present invention, in the database system according to the twelfth aspect, the frequency coefficient of the specific node is equal to or less than a predetermined criterion. A first link is defined between the first node and another second node, and a second link is defined between the node and another second node. When the absolute value of the signal transmission coefficient of the link is equal to or greater than a predetermined criterion, the first link and the second link are deleted, and a first link is provided between the first node and the second node. And a link reconfiguration unit having a function of performing a link reconfiguration for adding a new link having a code corresponding to the product of the code of the second link and the code of the second link.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below based on an embodiment shown in the drawings.

[0022]§1. The database system according to the present invention
Basic concept of system  First, in order to explain the basic concept of the present invention, FIG.
For a basic database system as shown
Describes how to access data. The data shown in this Figure 1
The base system is a database containing a large amount of data.
Node 1 and a set of nodes and links to access it
And the body 2. Node link aggregate
2 is composed of a number of nodes and links.
In the illustrated example, nine nodes N1 to N9 and seven link
L1 to L7 are defined. To each node N1 to N9
Defines predetermined keywords K1 to K9, respectively.
Each link between nodes uses individual keywords
Plays the function of linking. In this sense,
Link group 2 is a "keyword network"
Can also be called.

As shown, a link is not defined between all nodes, but only between nodes that are related to each other (ie, between nodes for which keywords related to each other are defined). ing. For example, in the illustrated example, the link L1 defined between the node N1 and the node N2 is a keyword K1 and a keyword K1.
2 has some relevance. Similarly, a link L3 defined between the node N2 and the node N4 indicates that the keyword K2 and the keyword K4 have some relationship, and a link defined between the node N4 and the node N5. L4 is the keyword K4
And the keyword K5 have some relevance. The node N1 and the node N5 are indirectly connected via such a link, and by providing the keyword K1, the keyword K5 can be obtained as a related keyword. Each link is weighted, and the weight indicates the degree of association between nodes.

What keywords are defined for which nodes and what links are defined between the nodes are as follows.
Primarily, this is a task assigned to the administrator of this database system, and when a general user uses this system, some node / link aggregate 2 has already been constructed. . However, as will be described later, the node / link aggregate 2 has a function of learning based on a user's use action, and the weight of each link is modified as the user uses it. A new link may be defined between nodes for which no link has been defined. Therefore, even if the administrator does not define any links, the node / link aggregate 2 is gradually formed as the user uses them.

When the operator 3 wants to use specific data in the database 1, the operator 3 first accesses the node / link aggregate 2. From the viewpoint of the operator 3, the node / link aggregate 2 functions as a front-end processor for accessing the database 1. The operator 3 first designates a specific node in the node / link aggregate 2 as a target node. The node can be specified by inputting a keyword. For example, operator 3
Gives a specific keyword K1 to the node / link aggregate 2, the node N1 corresponding to the keyword K1 is designated as the target node. In the drawing of the present application, for convenience of description, the target node at that time is displayed by drawing a circle around the node point of the black circle. In FIG. 1, a circle drawn around the node N1 indicates that the node N1 is the current node of interest.

Each node in the node / link aggregate 2 is associated with specific data in the database 1. In other words, if one node is specified, data associated with this node can be extracted from the database 1. Therefore, if the operator 3 gives an instruction to view the data on the current node N1 at the present time, the data associated with the node N1 is extracted from the database 1 and provided to the operator 3. Node and database 1
For example, the correspondence with each data in the above may be performed by giving each node specific address information in the database 1. When a data browsing instruction regarding the node of interest N1 is given, the database 1 is accessed to read out predetermined data based on the address information of the node N1, and this is provided to the operator 3.

As described above, if the operator 3 gives the keyword K1 to the node / link aggregate 2 and gives an instruction to browse the data, the data corresponding to the node N1 is read from the database 1 and 3 will be provided. However, such a search process is only a search process in a conventional general database system in which when a predetermined keyword K1 is input, data previously associated with the keyword K1 is presented. Of course, the database system according to the present invention can perform such a conventional general search process. However, the main point of the present invention is that when an operator gives one keyword to the system, this keyword is used. Another object of the present invention is to make it possible to search for another keyword related to, and to realize a more flexible search operation.

For example, when the operator gives a predetermined keyword K1 to the node / link aggregate 2, the node N1 is designated as the node of interest as shown in FIG. As described above, if necessary, the operator can give an instruction to perform “browsing of data associated with the node of interest N1”. Here, it is assumed that the operator is not directly interested in the data associated with the node N1, but is searching for other data related to the keyword K1. In this case, an instruction to perform “search processing using node N1 as a target node” may be given to node / link aggregate 2. When such a search instruction is given, the node / link aggregate 2 executes a process of searching for another node related to the target node N1 while referring to the defined link aggregate. For example, the node of interest N1
, When all nodes directly or indirectly linked by a link are searched for as related nodes, in the case of the example of FIG. , N4, and N5 are searched as related nodes. However, in practice, it is preferable to extract only nodes having a certain degree of relevance as related nodes in consideration of the weight of each link. For example, in the example of FIG. 1, when the weight of the link L4 is small, it is determined that the relationship between the target node N1 and the node N5 is low, and the node N5 is removed from the related node. Further, as will be described in detail in §4.3 below, the signal transmission between nodes connected by one link is defined as the number of hops H = 1, and only nodes connected with a predetermined number of hops or less are defined. It is preferable to impose the condition of being an associated node. For example, when the condition that the number of hops is H = 2 or less is imposed, in the example of FIG. 1, the number of hops H = 3 for the target node N1 regardless of the weight of each link.
Is not extracted as a related node.

Here, by performing a search process using the node N1 as a target node, three nodes N2, N3,
Assume that only N4 has been extracted. FIG. 2 is a diagram showing the result of such a search process. Three nodes N
The reason why 2, N3 and N4 are drawn as outlined node points is to indicate that these three nodes are nodes extracted as related nodes. Hereinafter, the related nodes extracted by the search will be indicated by outline node points. The node N5 indicated as a black node point in the figure is not extracted as a related node in the current search.

When the related nodes N2, N3 and N4 have been searched in this way, these "related nodes" are presented to the operator as "candidate nodes". Specifically, the candidate nodes may be presented by displaying each keyword K2, K3, K4 defined for each of the candidate nodes N2, N3, N4 on a display or the like. Here, the “related node” is referred to as a “candidate node” because the operator selects a new node of interest from these nodes. As shown in FIG. 2, the current node of interest is node N1, but the operator
One of 2, N3, and N4 is designated as a new node of interest. For example, if the operator
Assuming that the candidate node N4 is adopted as a new node of interest, as shown in FIG. 3, the node N4 becomes a new node of interest, and all other nodes return to normal nodes. The candidate nodes to be adopted are specified, for example, by keywords K2, K3, and K3 presented on the display screen.
It can be performed by an operation of selecting the keyword K4 among K4.

In consideration of the convenience of the description in “AND search” described later in §7, in the specification of the present application, a search process for a predetermined target node has a certain degree of association with the target node. The node extracted and determined as "related node" is referred to as a "related node".
Will be referred to as a “candidate node”. In the “AND search” described in §7 described below, only some of the extracted “related nodes” are “candidate nodes”.
However, in the normal search described here, all of the extracted “related nodes” are presented as “candidate nodes” as they are, so for the time being, “related nodes” = “ It can be considered as "candidate node". Therefore, in the description up to §6, if there is no particular problem, “related node” and “candidate node” will be treated as synonyms.

Viewing the above search processing as an operation on the operator side, it is as follows. First, the operator inputs a predetermined keyword K1 and gives a search instruction. Then, a search process is performed with reference to the link aggregate, and other keywords K2, K3, and K4 having a certain degree of relevance to the keyword K1 are displayed on the display screen (the candidate nodes N2, N3, and N4 are displayed). Presentation). The operator selects a keyword that seems to be related to the data he or she is looking for from among these new keywords (adopts one of the candidate nodes as a new node of interest). As a result, the node N4 becomes a new node of interest.

As described above, by giving a search instruction, the operator can walk one after another to related nodes and move in the keyword network. Moreover, if necessary, the data associated with the current node of interest can be browsed at any time. For example, as shown in FIG.
If the user wants to browse the data associated with this node N4 when becomes the new node of interest, a browsing instruction may be given at that time. Then, the data corresponding to the node N4 is read from the database 1 and presented to the operator. Also, if an instruction to perform “search processing using node N4 as a target node” is given, a related node having some degree of relevance to node N4 is extracted as a candidate node. For example, in the previous “search processing using the node N1 as the target node”, the relevance is low, and the node N5 leaked from the candidate node is searched as a candidate node in the current search processing.

In general, when searching for necessary data from a database system, an appropriate keyword can be immediately imagined, but the optimum keyword is not always imagined. In such a case, the system according to the present invention enables flexible search with a very high degree of freedom. That is, if the user enters the keyword K1 that comes to mind for the time being and gives a search instruction, the keywords K2, K3, and K4 related thereto are automatically presented as candidates on the display, and the operator becomes a search target. More suitable keywords can be adopted to access the data. Here, K4 (node N4) is used as a more appropriate keyword.
Is adopted and a search instruction is given again, for example, a new keyword K5 is presented. Here, assuming that the keyword K5 is the optimum keyword for accessing the data to be searched, the keyword K5 (node N5) is adopted. From the target data.

§2. Learning node / link aggregates One of the features of the database system according to the present invention is as follows.
The point is that the node / link aggregate 2 has a learning function. As described above, the node / link aggregate 2 shown in FIG. 1 is primarily constructed by the administrator of this database system, but then gradually forms in the course of use by the user. It will change. Therefore, it is sufficient to prepare one common database for all users, but it is preferable to prepare the node / link aggregate 2 separately and independently for each user. As described above, if a separate node / link aggregate 2 is prepared for each user, even if all the node / link aggregates 2 are initially a common one constructed by the system administrator, As each user uses this system, the node / link aggregate 2 for each user can change to a form that is easy for each user to use.

Thus, the node / link aggregate 2 is
According to the present invention, learning is performed in accordance with the following basic policy in order to change to an easy-to-use form. That is, when a plurality of candidate nodes are extracted by searching for a certain node of interest, and when the operator selects a desired candidate node from among these candidate nodes, the weighting of the link on the path from the node of interest to the selected node is performed. Increase. For example, in the example above,
As shown in FIG. 2, three candidate nodes N2, N3, and N4 are searched for the node of interest N1, and the operator selects the node N4 from among them. As a result, as shown in FIG. Has become a new node of interest. In this case, the weights of the links L1 and L3 on the path from the target node N1 to the adopted node N4 are increased. In FIG. 3, the links L1 and L3 are indicated by thick lines, which indicates that learning for increasing the weight has been performed.

It is also possible to reduce the link weight by learning. For example, in the case of the above example, three candidate nodes N2 and N
Even though 3, N4 was presented, the operator adopted node N4, and node N3 was omitted from the adoption. In other words, the link L2 did not participate in the selection of the node. In such a case, the link L2
When the correction for reducing the weight of is performed, learning in a form according to the usage form of the user becomes possible. For example, when this user performs a search with “node N1 as a target node,
The event "the node N4 among the candidate nodes has been adopted" has been performed five times. In this case, in any of the five learnings, the correction is performed such that the weights of the links L1 and L3 are increased and the weights of the link L2 are reduced. As a result, the weights of the links L1 and L3 become very large, and conversely, the weight of the link L2 becomes very small. Therefore, for example, at the time when “the sixth search using the node N1 as the target node” is performed,
The relationship indicated by the link L2 (the relationship between the nodes N2 and N3) is considerably reduced, and the node N3 is no longer extracted as a candidate node.

As described above, if learning to decrease weighting is performed together with learning to increase weighting, a node that is unlikely to be adopted in the future as far as the past history is viewed, At the time of retrieval, it is possible not to be extracted as a candidate node, so that the candidate nodes can be narrowed down. In an actual database system in which the total number of nodes becomes enormous, it is important to narrow down the number of candidate nodes presented to the operator to some extent in order to improve usability.

In order to perform both the learning to increase and the learning to decrease the link weights, the learning may be performed based on the following criteria. That is, several candidate nodes are extracted by a search for a predetermined target node, and if one node is selected from these candidate nodes, all paths from the target node to the individual candidate nodes are to be learned. It is a path.
Then, among the links on the learning target path, the weights of the links on the path from the target node to the adopted node are increased, and the weights of the other links are reduced. Such weight increase / decrease correction is, in effect, a "weight of a link on a path from a target node to an adopted node,
It can be said that "the correction is relatively increased with respect to the weight of other links."

As described above, when learning is performed by increasing or decreasing the weights, in the above-described example, the following weight corrections are performed. That is, as shown in FIG.
Assuming that three candidate nodes N2, N3, and N4 are extracted by searching for the target node N1, all links on the path from the target node N1 to each of the candidate nodes N2, N3, and N4 become links on the learning target path. Become. Specifically, the links L1, L2, L3 are to be learned. If the operator adopts node N4, the node of interest N1
A modification is made to increase the weights of the links L1 and L3 on the path from to the adopted node N4, and to decrease the weights of the links L2 on the other learning target paths. At this time, the link L4 does not become a learning target because the node N5 is not a candidate node, and the weight remains unchanged.

In short, the concept of the link weight learning described above is to reduce the weight of the link to the node which has been presented to the operator as a candidate node but not adopted. ,
On the contrary, the object is to increase the weight of the link to the adopted node. The important point is that all learning is performed based on the action of the operator (user) of "selecting one from a plurality of candidate nodes". Therefore, when the database system is used by a large number of users, learning proceeds in a different manner for each individual user. As described above, since the node / link aggregate 2 is prepared separately and independently for each user, the more the user is used, the more the node / link aggregate 2 for each user becomes. Learning according to usability will proceed.

Although only the link weights have been described as learning targets here, in the present invention, weights are also defined for nodes, and node weights are also subject to learning. The handling of this node weight will be described later.

§3. Generation of New Link In the above §2, it has been described that learning is performed on link weighting. However, it is difficult to realize a flexible search process according to the usability of each user only by modifying the weight of the existing link. For example, in the node / link aggregate 2 illustrated in FIG. 1, the nodes N2 and N
3, N4 and N5 are linked by a link. Therefore, if the search process is repeatedly performed, the node N1
It is possible to reach the nodes N2, N3, N4, N5 from. In fact, in the case of the above-described example, the candidate nodes N2 and N
3, N4 has been realized, and then the node N
4 and the second search using the selected node N4 as a new target node, it is possible to reach the candidate node N5.

However, no link is established between the nodes N1 to N5 and the nodes N6 to N9. , Nodes N6 to N9 cannot be searched.

As described above, the node link aggregate 2 is constructed primarily by the administrator of this system. Therefore, nodes N1 to N5 and node N6
If no link is defined between the keywords K1 to K9 and the keywords K6 to K5
That is, it has been determined by the administrator that there is no relevance between the system and 9. However, such a judgment of the administrator is not universal, and for a specific user, for example, it may be recognized that the keyword K4 and the keyword K7 are closely related. Absent. In such a case, for example, as shown in FIG. 4, a temporary link L8 as shown by a broken line is generated between the node N4 and the node N7, and the temporary link L1 and L7 are temporarily generated. By performing a search using both the link L8 and the link L8, a search with a higher degree of freedom can be performed. That is, in the state where the temporary link L8 is added, if a search is performed using the node N4 as the target node, for example, the nodes N1, N2, N3, N5, N6 and N6 shown as outlined node points in FIG. N7, N
8 can be extracted as a candidate node.

In the present specification, the existing link described above is called a "static link", and a link temporarily generated at the time of retrieval is called a "dynamic link" to distinguish between the two. In the example shown in FIG. 4, the links L1 to L7 indicated by solid lines are static links, and the link L8 indicated by broken lines is a dynamic link. If a dynamic link is generated at the time of search, the link to the node of interest is not completely connected by the existing static link, and even if there is a node that has an unconnected part, it is temporarily added to this unconnected part. By defining dynamic links in the search, search becomes possible,
Candidates can be extended to such disconnected nodes.

Now, as shown in FIG. 4, in the search using the node N4 as the target node, the dynamic link L8
Are temporarily defined (the method of defining the dynamic link will be described later), so that the nodes N1, N2, N
Suppose that 3, N5, N6, N7, and N8 are extracted as candidate nodes. At this time, for the operator,
Keywords K1, K2, K3, K5, K6, K7, K8
Are presented as information indicating individual candidate nodes. Then, it is assumed that the operator has adopted the node N6 as a new target node from these candidate nodes. As described above, the operator can browse data associated with the new node of interest N6, and can perform a new search using the node N6 as the node of interest. As described above, the definition of the dynamic link considerably increases the degree of freedom of the search range.

As described above, when the candidate node N6 is extracted by the search using the temporarily defined dynamic link L8, and the candidate node N6 is adopted (in other words, the dynamic link L8 becomes the target node N4 If the link is on the path from to the adopted node N6),
The process of adding this dynamic link L8 as a new static link L8 is performed. That is, the temporary dynamic link has been promoted to a permanent static link. FIG. 5 shows a state in which the adopted node N6 is set as a new target node and the dynamic link L8 is promoted to the static link L8. At this time, as shown by the thick line in the figure, the weights of the links L8 and L5 on the path from the target node N4 to the adopted node N6 are increased (the original weights before the increase of the link L8 are dynamic. Link L
8 when defining it). On the other hand, other links to be learned (links L1, L2, L3, L4 on the path from the target node N4 to each candidate node)
With respect to the weighting of L6), learning to decrease is performed.

Eventually, in this example, the temporarily defined dynamic link L8 is promoted to the static link L8, and is added as a new member of the node / link aggregate 2. However, a temporarily defined dynamic link may disappear without being promoted to a static link. For example, as shown in FIG. 4, in a state where some candidate nodes are presented, instead of the operator adopting the node N6,
When the node N5 is adopted, the dynamic link L8 does not become a link on the path from the target node to the adopted node, and thus disappears without being promoted to a static link. In short, a temporarily defined dynamic link will remain as a static link if it participates in the operator's adoption as a path, but will disappear as it is if it is not involved in the adoption. become.

As described above, the dynamic link required by the user is promoted to the static link and the node
If the process of adding to the link aggregate 2 is performed, the links not provided by the administrator of the database system will gradually increase, and a link aggregate that is easy to use for the user will be formed. become. Also, if such a link addition method is used, even if the system administrator does not define any links at first (that is, even if no static links exist at first), the system administrator can use the link addition method. As a person uses this system, a static link is gradually formed. Therefore, the method of adding a new link described here is a very effective method.

In the above description, a method of defining a dynamic link is not mentioned at all. However, in order to temporarily define a dynamic link L8 as shown by a broken line in FIG. It is necessary to keep. In the illustrated example, the dynamic link L8 is defined between the nodes N4 and N7.
There is room to define a dynamic link between N6, between nodes N4 and N8, and between nodes N4 and N9. Also, there is room for defining a dynamic link between the nodes N4 and N3 and between the nodes N4 and N1, and there is room for defining a dynamic link between any nodes that are not directly connected by a static link. Is there.
However, it is not preferable to define a dynamic link between nodes that do not have any relation, because the “link” indicates some relation between the two nodes.

Therefore, according to the present invention, at the time of search, between nodes that are not directly connected by a static link, the relevance of the keywords defined for both nodes is specifically evaluated, and when the evaluation result satisfies a predetermined condition. First, a dynamic link is defined between both nodes. For example, in the case of the example shown in FIG. 4, the relevance between the keyword K4 defined for the node N4 and the keyword K7 defined for the node N7 is evaluated, and the evaluation result satisfies a predetermined condition. The dynamic link L8 is defined between -N7.

As an example of a method of evaluating the relevance of two keywords, there is a method of quantitatively evaluating the degree of coincidence of character strings constituting both keywords. For example, since the keyword "high blood pressure" and the keyword "blood pressure value" match by two characters of "blood pressure" out of three characters, a quantitative evaluation such as a matching degree of "2/3" is obtained. It is possible.
Alternatively, if a certain thesaurus is prepared and the similarity is quantitatively determined in this thesaurus, it is possible to quantitatively evaluate the relevance of the two keywords. For example, "high blood pressure" and "high pressur
Using a thesaurus with a degree of similarity to “e” of 100, a keyword “hypertension” and a “high p
The degree of matching with the keyword “ressure” can be quantitatively evaluated. When such an evaluation value is equal to or more than a certain reference, a dynamic link may be defined between both nodes. Further, the evaluation value can be used as it is as the weight of the dynamic link.

In order to evaluate the relevance between nodes as rationally as possible, it is preferable to define a plurality of equivalent keywords for one node. For example,
In the example illustrated in FIG. 4, it has been described that the keyword K4 is defined for the node N4, and the keyword K7 is defined for the node N7. In this case, as shown in FIG. 6, the keyword K4 is composed of one representative keyword K40 and a plurality of equivalent keywords K41 to K45, and the keyword K7 is one representative keyword K70 and a plurality of equivalent keywords K
71 to K75, and a dynamic link may be defined between both nodes when the evaluation result of any one of the keywords is equal to or more than a certain standard. In the example shown in FIG.
Since the evaluation result of the relevance between 42 and the equivalent keyword K74 is equal to or higher than the reference, a dynamic link is defined between the nodes N4 and N7.

In this case, each of the equivalent keywords is a keyword equivalent to the representative keyword and can be used in place of the representative keyword. For example, if an equivalent keyword such as “high blood pressure”, “abnormal blood pressure”, and “hypertension” is defined for the representative keyword “high blood pressure”, the relevance of any of the equivalent keywords is evaluated. I just need
More rational evaluation results can be obtained. That is,
Although the nodes are originally related nodes, it is possible to eliminate the absurdity that the evaluation result of "no relation" is output because the expression form of the keyword character is different.

§4. To distributed database systems
Application Examples The basic concept of the database system according to the present invention, the method of learning a node / link aggregate, and the method of generating a new link have been described with reference to simple examples. Here, an embodiment in which the present invention is applied to a distributed database system will be described more specifically.

<4.1: Definition of Class Link in Distributed System> In recent years, an environment in which a plurality of computers are interconnected via a network has been generalized, and a database constructed for each computer has been used. ,
Distributed database systems that can be used from other computers have become widespread. In such a distributed database system, each local database is handled by the concept of "class". Here, for convenience, as shown in FIG. 7, three classes A, B,
The following description will be made taking a very simple distributed database system made of C as an example.

In FIG. 7, a circle is drawn for each class, and nodes N1 to N7 are shown on the circle. Where each circle represents a unit of an individual class,
Nodes on each circumference indicate nodes belonging to the specific class. For example, nodes N1 and N2 are nodes belonging to class B, nodes N3 and N4 are nodes belonging to class A, and nodes N5, N6, N
Reference numeral 7 denotes a node belonging to the class C. Normally, the databases for each class are provided at spatially separated locations, and are connected to each other by a communication line or the like. Also in the illustrated example, it is assumed that the classes A, B, and C are spatially separated from each other. It should be noted that the illustrated circumference is for indicating the affiliation of each node, and is not for indicating the link between the nodes. Therefore, in the state shown in FIG. 7, no link is defined between the nodes N1 to N7. Also, in the illustrated example, only a total of seven nodes are shown, but actually, there are many nodes in each class.

In such a distributed database system, in order to define a link between each node, in the present invention, association between individual classes is defined in advance. Here, the association between the classes is referred to as “class link”. Up to now, §1 to §3
The links (static link and dynamic link) described in (1) are links between nodes (links generally called instance links) indicating the association between nodes, but the class link defined here is: A link between classes indicating the association between classes.

FIG. 8 is a diagram showing an example of a class link definition. In the figure, a straight line or a circle indicated by a bold line indicates a class link. Specifically, a class link AB is defined between classes A and B, and a class link A is defined between classes AC.
C is defined. On the other hand, a class link AA is defined between the classes A and A, and a class link BB is defined between the classes BB. The class links AB and AC shown by straight lines are class links that indicate the degree of association between different classes, and here are “remote links”.
I will call it. On the other hand, class links AA and BB indicated by circles are class links indicating the degree of association between oneself and oneself, and will be referred to as “local links” here.

Here, in order to avoid confusion, terms relating to “link” used in this specification are arranged as follows.

(1) Instance link (link indicating the relation between nodes: a general term for the following static link and dynamic link) (1): Static link (permanent link constructed as a link aggregate) : Weight learning is performed as the user uses the system.In this specification, when there is no particular problem, the link may be simply referred to as “link”. Temporary link defined at the time of search when the keyword defined for each node is relevant: a dynamic link is promoted to a static link if used for node adoption,
(2) Class link (Link indicating the relationship between classes and classes: The following general term for remote links and local links :)
In the case of the embodiment described here, weights are defined for each link, but learning on weights is not performed.) (2): Remote link (link indicating relevance between different classes: (2) Local link (link indicating the relationship between oneself and oneself for the same class: indicated by a thick circle in FIG. 8).

After all, in the example shown in FIG.
Class links are defined between the classes A and C, between the classes A and A, and between the classes B and B, however, the class links are defined between the classes B and C and between the classes C and C. It has not been. What class links to define,
Then, what weight is defined for each class link is determined at the discretion of the administrator of this database system. However, in practice, when defining a class link, it is not necessary to simply consider the convenience of search, but rather the conditions of use of the database corresponding to each class, the contents of the use contract, the use fee, and the access time. And so on. Therefore, there is a case where a class link is deliberately not defined or a class link having a very small weight is defined due to a restriction on management of the database system. In particular, a database of medical cases may define only a very limited number of class links from the viewpoint of protecting the privacy of patients.

In the example of this embodiment, a thesaurus dictionary is prepared for each class link. For example, in the example shown in FIG. 8, a thesaurus dictionary Tab is prepared for the remote link AB, a thesaurus dictionary Tac is prepared for the remote link AC, and a thesaurus dictionary Taa is prepared for the local link AA. , Local link B
A thesaurus dictionary Tbb is prepared for B. These thesaurus dictionaries are used when defining a dynamic link, and how to use them will be described later. However, it is very significant to prepare a unique thesaurus dictionary for each class link. For example, if class A is a Japanese database, class B is an English database, and class C is a French database, the English / Japanese / English thesaurus is used as the thesaurus dictionary Tab, and the thesaurus dictionary is used. It is reasonable to use "French-Japanese / Japanese-French thesaurus" as Tac.

<4.2: Definition of Static Link in Distributed System> The administrator of the database system defines a class link as shown in FIG.
Define static links between individual nodes. That is, referring to the keywords associated with the individual nodes, a task of establishing a static link with a predetermined weight is performed between the nodes associated with the mutually relevant keywords. When defining this static link, the condition of the class link described above is to be followed. That is, a static link can be defined between classes for which class links are defined, but a static link cannot be defined for classes between which no class link is defined in principle. FIG. 9 is a diagram showing a specific example of a static link defined for each node shown in FIG. For example, the static link L1 is connected to the node N1
Link between -N2, which is a permitted link because the local link BB is defined for class B. Similarly, the static link L2 is a link permitted by the remote link AB, the static link L3 is a link permitted by the local link AA, and the static link L4 is a link permitted by the remote link AC. . On the other hand, since no remote link is defined between the classes B and C, for example, no static link is defined between the nodes N1 and N7. In addition, since no local link is defined for class C, a static link cannot be originally defined between nodes belonging to class C. However, here, exceptionally, node N6 is determined by the administrator's intention. A static link L5 is defined between -N7. Thus, in the present embodiment,
In principle, it is preferable to define a static link based on the conditions indicated by the class link. However, if the administrator determines that exceptional measures are particularly necessary, the static link may be appropriately set contrary to the principle. Can be defined.

<4.3: Search Processing in Distributed System> Now, as shown in FIG. 9, in a distributed database system including three classes A, B, and C, seven nodes N1 to N7 are used. In the case where the five static links L1 to L5 are defined, how the search processing and the learning processing are performed will be specifically described.

First, suppose that the node N1 is selected as the first node of interest as shown in FIG. The designation of the first node of interest is performed by the operator inputting the keyword K1 corresponding to the node N1.
After the node of interest has been determined in this way, next, a search process for this node of interest is executed. In this embodiment,
A search process for a specific node of interest is performed by a process of transmitting a signal having a predetermined signal value from the node of interest to another node along a static link (or a dynamic link as described later). Therefore, for each static link (and dynamic link), a signal transfer coefficient for indicating a weight is defined. Here, it is assumed that a signal transmission coefficient is defined for each link as shown in FIG.
The signal transfer coefficients in this embodiment are all shown as percentage values, and in the illustrated example, the link L1: 2
The coefficients are defined as 5%, link L2: 50%, link L3: 30%, link L4: 60%, and link L5: 80%. The value of the signal transfer coefficient of each static link is primarily defined by the system administrator. However, as will be described later, learning is performed as the user uses the system. The numbers will be modified.

The search process using the node N1 as a target node is a process of extracting another node having a certain degree of relevance to the node N1 as a candidate node. Here, in order to extract such a candidate node, a signal having an initial signal value of 100 is transmitted from the node of interest N1 along each link. Then, each time the signal passes through each link, the signal transmission coefficient defined for the link is multiplied by the original signal value. For example,
In FIG. 10, in the signal transmission from the node N1 to the node N2, a multiplication with a signal value of 100 × 25% is performed, and the signal value of the signal reaching the node N2 is attenuated to 25. Similarly, in the signal transmission from the node N1 to the node N3, a multiplication with a signal value of 100 × 50% is performed, and the signal value of the signal reaching the node N3 is attenuated to 50. The signal reaching the node N3 is further transmitted to the node N4. In this signal transmission, the signal value is 50 × 30%
Is performed, the signal value of the signal reaching the node N4 is attenuated to 15. Furthermore, this signal value 1
5 is transmitted from the node N4 to the node N5, a multiplication with a signal value of 15 × 60% is performed, and the node N5
Will be attenuated to 9.

The table shown in the lower column of FIG. 10 shows the signal values of the signals transmitted to each node when a signal having a signal value of 100 is given to the target node N1. Such signal transmission is similar to the flow of current through electronic circuits connected by resistance elements. That is, if each link is considered as a resistance element having a predetermined resistance value (a link having a smaller signal transmission coefficient has a larger resistance value), and if the signal value is considered as a voltage value, the signal attenuation will be a voltage drop due to the resistance element. Will be equivalent.

Now, the larger the weight of a link (the two nodes connected by this link will have a greater relationship), the larger the signal transmission coefficient will be defined. , The signal attenuation is reduced, and the signal value at the node where the signal arrives becomes a large value. Therefore, it can be said that a node having a larger signal value has a higher relevance to the target node. Therefore, in this embodiment, the priority is defined for each node based on the signal value of the signal that has arrived. The priorities in the chart shown in the lower column of FIG. 10 are the priorities defined in this manner. Although the node N5 is a node having a priority, in this example, the node N5 is handled as “below the condition”, and the priority is not particularly defined.

In the example shown here, the lower limit condition of the effective signal value is set to 10, and the signal value of the transmitted signal is 10
The following nodes are excluded from consideration as "below conditions". As described above, the node whose signal value is equal to or less than the condition is treated in the same manner as when there is no signal transmission. Therefore, in the case of the example of FIG. 10, the node N5 is treated as if there was no signal transmission even though the signal having the signal value 9 was transmitted. Even if another connected node exists (FIG. 10)
In the example of (1), such a node does not exist), but the signal transmission processing downstream from the node N5 is no longer performed. Eventually, in this example, the node N5 is treated in the same manner as the nodes N6 and N7 that did not transmit any signal.

Thus, in the search processing using the node N1 as the target node, only the nodes N2, N3, and N4 for which a valid signal value has been obtained are extracted as related nodes, and these three nodes are extracted as candidate nodes. Will be presented to the operator. In FIG. 10, the nodes indicated by white node points are these candidate nodes. Although the node N5 has a slight relevance, it is not extracted as a candidate because the relevance is below the condition.

The operator selects the presented candidate node N
A new node of interest is selected from 2, 2, and N4. At this time, each candidate node is presented to the operator based on the priority. For example, in the case of the example in FIG. 10, the presentation is performed in the order of the candidate nodes N3, N2, and N4 in accordance with the priority order. Displayed on the display.
If all the keywords cannot be displayed on one screen of the display, they are displayed while switching the screens in priority order.) As described above, when the candidate nodes are presented based on the priority order, it is possible to consider the degree of relevance when determining the adopted node (new target node). That is, the operator can recognize that the candidate node displayed with higher priority has a higher degree of association, and can be preferentially adopted.

In the above example, whether or not to extract as a related node (candidate node) is determined based on whether or not the signal value of the transmitted signal satisfies the condition as a valid signal value. However, in addition to the method of setting conditions based on such signal values, it is also possible to set conditions based on the hop count H. That is, the signal transmission between nodes connected by one link is defined as the hop number H = 1, and when the hop number H exceeds a predetermined upper limit, the signal transmission processing is stopped. is there. For example, FIG.
In the case of the example shown in FIG. 0, the signal transmission to the nodes N2 and N3 directly connected to the node of interest N1 by a link is the signal transmission corresponding to the hop number H = 1, but the node N4
The signal transmission to the node N5 is a signal transmission corresponding to the hop number H = 3, and the signal transmission to the node N5 is a signal transmission corresponding to the hop number H = 3. Therefore, for example, if the upper limit of the hop number H is set to H = 2, the signal transmission in which the hop number H becomes 3 or more is stopped. In the case of the example shown in FIG. 10, the signal transmission to the node N4 is performed, but the signal transmission to the node N5 downstream therefrom is not performed.

In practice, it is preferable to set an AND condition between the condition based on the signal value and the condition based on the number of hops. That is, it is only necessary to extract, as a candidate node, only a node that can transmit a signal with a hop number within a predetermined condition from a node of interest and whose signal value of a transmitted signal is equal to or higher than a predetermined condition. Specifically, first, a search range (a range in which signal transmission calculation is performed) is limited by a condition based on the number of hops, and a signal transmission calculation process is performed only on nodes connected with a predetermined number of hops or less, Only a node that finally obtains a signal having a signal value equal to or more than a predetermined value may be set as a candidate node. In this way, setting conditions for candidate node extraction and setting only nodes having a certain degree of relevance as candidate nodes has the advantage of shortening the search time in terms of system performance, and the usability of the search function. It is important to improve. In an actual database system, there are a huge number of nodes, so if nodes with only low relevance are presented as candidate nodes,
The search waiting time becomes long, the number of candidates becomes too large, and the usability is reduced.

<4.4: Search Processing Considering Class Link Weight> The above-described search is a search considering the static link weight (signal transmission coefficient). It is also possible to perform a search process in consideration of the above. FIG. 11 is a diagram illustrating an example of a search process that considers both the weight of a static link and the weight of a class link. Here, the percentage values shown below the respective static links L1 to L5 are the signal transmission coefficients of the respective static links, and are exactly the same as the values shown in FIG. On the other hand, each class link AA, B
The percentage values shown below B, AB, and AC are the signaling coefficients of each class link.

Here, when performing signal transmission starting from the node of interest, the product of the signal transmission coefficient for the static link and the signal transmission coefficient for the class link is used. For example, similar to the example shown in FIG.
A signal having an initial signal value of 100 from the node of interest N1 is
Consider the case of transmitting along each link. Then, in FIG. 11, in the signal transmission from the node N1 to the node N2, the signal value is 100 × 25% (link L1) × 20%
The multiplication (link BB) is performed, and the signal value of the signal reaching the node N2 is attenuated to 5. Similarly, in signal transmission from node N1 to node N3, signal value 10
The multiplication of 0 × 50% (link L2) × 80% (link AB) is performed, and the signal value of the signal reaching the node N3 is 4
It will attenuate to zero. The signal arriving at the node N3 is further transmitted to the node N4. In this signal transmission, a signal value of 40 × 30% (link L3) × 90% (link AA) is multiplied, and the signal arriving at the node N4 is transmitted. Will attenuate to 10.8. Furthermore,
This signal having a signal value of 10.8 is transferred from node N4 to node N5.
Signal value 10.8 × 60% (link L
4) × 10% (link AC) multiplication is performed, and the signal value of the signal reaching the node N5 is attenuated to 0.648.

The table shown in the lower section of FIG. 11 shows the signal values of the signals transmitted to each node when a signal having a signal value of 100 is given to the target node N1. In this example,
The lower limit condition of the effective signal value is set to 10, and a node whose signal value of the transmitted signal is 10 or less is excluded from consideration as “below the condition”.
As a result, only the nodes N3 and N4 are extracted as the candidate nodes, and the priority is defined in this order.

As described above, if the search is performed in consideration of both the weight of the static link and the weight of the class link, the system administrator supervises the tendency of each search process. It becomes possible to control it. For example, if the communication line for accessing a particular class tends to be very congested, nodes that belong to this class can be extracted as candidates by making modifications to reduce the weight of the class link for this class. Can be suppressed.

As described above, the weighting element of the static link is to be learned, and a link aggregate having a different weight is formed for each user. On the other hand, if the weight of the class link can be set only by the system administrator, the search processing of the entire system is supervised and managed by the administrator while respecting the learning content of each user. It becomes possible to do.

<4.5: Learning Process in Distributed System> Next, a specific learning process in the distributed system will be described. As described above, since the weight of the class link is not a learning target, a simple case in which the class link is uniformly weighted at 100% is considered here. That is, instead of the example shown in FIG.
Consider the learning process for the example shown in FIG.

Now, as shown in FIG. 10, as a result of performing a search using the node N1 as the node of interest, three nodes N
2, N3 and N4 are presented as candidate nodes, and the operator adopts node N4 of the three candidate nodes. As a result, the node N4 becomes a new node of interest. Then, learning is performed by the adopting action of the node N4. Learning is performed by modifying the weight of the static link to be learned. Specifically, a correction for increasing or decreasing the signal transmission coefficient of the static link is performed.

The basic principle of the learning process is as follows.
First, all paths from the target node to each candidate node are extracted as learning target paths. Then, the signal transmission coefficient of the link on the path from the target node to the adopted node is corrected to be relatively increased with respect to the signal transmission coefficients of the other links. In particular, in the embodiment described here, the signal transmission coefficient for the link on the path from the target node to the adopted node is increased, and the signal transmission coefficient for the other links on the learning target path is reduced. I am trying to fix it. Eventually, of all the static links, of the learning target paths, the links on the path from the node of interest to the adopted node are corrected to increase the signal transmission coefficient, and for the links other than the learning target paths, Correction to reduce the signal transfer coefficient is performed, and no processing is performed for links that are not on the learning target path.

In this example, the signal transmission coefficient percentage value is increased by 20 for the above link, and the signal transmission coefficient percentage value is increased by 20 for the above link.
I am trying to make a correction that only decreases. The learning target path in this example is, as shown in FIG.
From each of the candidate nodes N2, N
3, N4, and specifically, the static links L1, L2, and L3 are to be learned. Among them, for the links L2 and L3 on the path from the target node N1 to the adopted node N4, the signal transmission coefficient value is set to 2
A correction is made to increase by 0, and links L2 and L3
Are 50% + 20% =
70% and 30% + 20% = 50%. On the other hand, with respect to the remaining link L1 to be learned, the signal transmission coefficient value is corrected to be reduced by 20, and the signal transmission coefficient after learning is 25% −20% = 5%. The learning is not performed on the links L4 and L5 that have not become the learning target paths, and the value of the signal transfer coefficient remains unchanged. FIG. 12 is a diagram showing a state after a specific learning process according to such a basic policy is performed.

Now, let us see what kind of change occurs in the search processing due to such learning. FIG. 13 is a diagram showing a search result when the node N1 is again designated as the target node and a search is performed in a state where a new signal transfer coefficient is defined by the learning as described above.
The search act itself shown in FIG. 13 is exactly the same as the search shown in FIG. 10, but a different search result is obtained because the signal transmission coefficient of each link is modified. That is,
Similarly, when a signal having the initial signal value 100 is transmitted from the node of interest N1 along each link, in the signal transmission from the node N1 to the node N2, a multiplication with a signal value of 100 × 5% is performed, and the multiplication to the node N2 is performed. The signal value of the arrived signal is 5
(Under the condition). On the other hand, node N1
Signal from the node N3 to the node N3 has a signal value of 100 × 70
%, And a signal having a signal value of 70 is obtained at the node N3. Further, in transmission from the node N3 to the node N4, a multiplication with a signal value of 70 × 50% is performed, and a signal with a signal value of 35 is obtained at the node N4. Further, when the signal having the signal value 35 is transmitted from the node N4 to the node N5, a multiplication of the signal value 35 × 60% is performed, and
5, a signal having a signal value of 21 is obtained.

The table shown in the lower section of FIG. 13 shows the signal values of the signals transmitted to each node when a signal having a signal value of 100 is given to the target node N1. This chart is
Comparing with the table shown in the lower column of FIG. 10, it can be seen that the combinations and the priorities of the nodes extracted as the candidate nodes have changed. That is, in the search shown in FIG. 10 before learning, the node N2 is a candidate node, whereas in the search shown in FIG. 13 after learning, the node N5 is a candidate node instead of the node N2. this is,
This means that the degree of association between the nodes N1 and N4 has increased by learning.

If the operator again adopts node N4 after the search shown in FIG.
3 further increases, and conversely, the links L1, L2
4 will be reduced. In the previous search, the link L4 was not targeted for learning. However, in this search, the node N5 has become a candidate node, so the link L4 is also targeted for learning. However, since the link L4 is not a link on the path from the node of interest N1 to the adopted node N4, learning to reduce the signal transfer coefficient is performed.

As described above, each time the user performs a search for a predetermined node of interest and selects a new node of interest from among the candidate nodes presented by the search, the learning of the link on the learning target path is performed. Is performed. That is, the link used as the path from the node of interest to the adopted node is corrected by increasing the weight,
For the other links on the learning target path, the weight is reduced and corrected. In this way, if learning is performed to increase the weight of a frequently used path, a link aggregate that is easy to use for individual users will be constructed.

In the present embodiment, predetermined upper and lower limits are set for the signal transmission coefficient, and an increase correction exceeding the upper limit and a decrease correction exceeding the lower limit are not performed. I have to. For example, if the upper limit of the signal transfer coefficient is set to 150% and the lower limit is set to 1%, the increase correction is performed only up to 150% and the decrease correction is performed only up to 1%. However, if the signal transfer coefficient is 1% or less, it may be determined that the static link is to be eliminated.

Further, in this embodiment, no directionality is defined for link weighting. For example, the signal transmission coefficient indicating the weight of the link L2 is different from the case where a signal is transmitted from the node N1 to the node N3,
3 and a case where a signal is transmitted to the node N1. Normally, when the relevance of the node N3 from the node N1 is high, on the contrary, the relevance of the node N1 from the node N3 is also high. Therefore,
There is no problem even if the direction of the link weight is not particularly defined. However, if the relevance of the second node as viewed from the first node is not always the same as the relevance of the first node as viewed from the second node, the link weights are given a direction. You can also. In this case, for example, as a weight of link L2, a signal transmission coefficient in a direction from node N1 to node N3 and a signal transmission coefficient in a direction from node N3 to node N1 are separately defined. Good.

<4.6: Presentation of Candidate Nodes in Consideration of Node Weight> The learning processing for correcting the weight of the static link has been described above. If the candidate nodes are to be subjected to learning processing, individual candidate nodes can be presented to the operator in a priority order that takes into account the weight of each candidate node.

This will be shown in a concrete example. Here, it is assumed that the frequency coefficients indicating the weights are defined for all the nodes, and the frequency coefficients of all the nodes are set to 100% in the initial state in which the learning is not performed. The frequency coefficient defined for each node has no effect on the signal transmission process, but when that node is extracted as a candidate node, it is used as a parameter for determining the priority. When such a frequency coefficient is defined, the priority order of the candidate nodes is determined by the method shown in the table of FIG. 14 during the search processing shown in FIG. That is, the priority of a specific node is determined based on the product of the signal value of the signal transmitted to the node and the frequency coefficient of the node. In the example shown in FIG. 14, since the learning has not been performed yet, the frequency coefficients of all the nodes are 100%, and the product obtained by multiplying the frequency coefficients has the same value as the original signal value.

Now, let us consider a case where three candidates N3, N2 and N4 are presented to the operator in this order by the search shown in FIG. 10, and the operator adopts the node N4. In this case, as described above, learning for correcting the signal transmission coefficient of each link is performed. However, when weighting for a node is defined, learning for correcting the weighting of the node is performed. To do. Specifically, for the selected node, a correction to increase the frequency coefficient is performed, and for nodes other than the nodes on the learning target path, corrections to reduce the frequency coefficient are performed. For, a learning process of not performing any correction may be performed. As a result, the frequency coefficient indicating the weight of a node is a parameter indicating the frequency of adoption of the node in the past.

Here, for the adopted node, a correction is made to increase the frequency coefficient to × 1.5, and for the other nodes on the learning target path, a correction is made to reduce the frequency coefficient to × 0.7. Consider an example. Then, Figure 1
When the candidate node N4 is adopted in the search shown in FIG. 0, the frequency coefficient of the adopted node N4 is 100
% × 1.5 = 150%. on the other hand,
The frequency coefficients of the other nodes N2 and N3 on the learning target path are corrected to 100% × 0.7 = 70%. Since the nodes N5, N6, and N7 are not on the learning target path, the frequency coefficient is not corrected.

By the adoption action of the candidate node N4, learning for the node is performed, and learning for the link is also performed as described above.
3 as shown in the upper column. Considering the search result when the search is performed again with the node N1 as the target node after such learning is performed, the frequency coefficient defined for each node has no effect on the signal transmission process. , The signal values of the signals obtained at the nodes N1 to N5 are as shown in the lower column of FIG. However, the priority when presenting the candidate nodes N3, N4, N5 to the operator is determined based on the product of the signal value and the frequency coefficient. This product is as shown in the chart of FIG. 15, and after all, at the time of the search processing shown in FIG. 13, the priority order of the candidate nodes is determined by the method shown in the chart of FIG.

When the priority shown in FIG. 15 is compared with the priority shown in the lower column of FIG. 13, it can be seen that the priority is replaced with the priority. That is, if the priorities are determined in consideration of the weights of the nodes, the priority of the node N4, which is the adopted node, becomes higher than the priority of the node N3 used as a mere passing node, and the usability of the user is increased. It can be seen that is further improved.

<4.7: Search Processing Using Dynamic Link> In the distributed database system, the search example described so far has been a search using an existing static link. Here, a search example using the dynamic link described in §3 will be described.

For example, in the example shown in FIG. 10, suppose that static link L2 between nodes N1-N3 has not been defined. In the search using only the static link, in this state, when the search is performed with the node N1 as the target node, the node N1 is set as the candidate node.
Only 2 will be searched. As described above, a static link is primarily a link set by a system administrator, and is not always associated with nodes that are useful for individual users. Therefore, a temporary dynamic link is defined between specific nodes having relevance between keywords by using the following method, and a search is performed.

The basic method for defining a dynamic link is as described in §3. Here, as shown in FIG. 16, a specific condition for generating a dynamic link L2 between nodes N1 and N3 forming an unconnected portion because there is no existing link will be described. The determination as to whether or not the nodes N1-N3 of the unconnected portion should be connected by a dynamic link is made based on the evaluation result of the relevance of the keywords K1 and K3 defined for both nodes N1 and N3. That is, if the evaluation result satisfies a predetermined condition, the unconnected portion is temporarily connected by the dynamic link L2. Specifically, for example, each keyword K
If some equivalent keywords are defined as 1 and K3, respectively, the condition is satisfied if any equivalent keyword constituting the keyword K1 matches any equivalent keyword constituting the keyword K3. It may be determined that the evaluation result to perform is obtained. Of course, it is not necessary to make an exact match of the keyword an indispensable condition. For example, two-thirds of a character string such as
If they match, it may be determined that the condition is satisfied.

In the present embodiment, a thesaurus dictionary prepared with class links is used to evaluate the relevance of two keywords. As shown in FIG. 8, a thesaurus dictionary Tab is prepared for the remote link AB. Therefore, nodes N1-N
When evaluating the relevance between the three, the thesaurus dictionary T
The evaluation is performed with reference to ab. For example, in the thesaurus dictionary Tab, "high blood pressure", "high pressure"
"," Blood pressure abnormalities ", and" arteriosclerosis "are all defined as synonyms, and in evaluating the association between the nodes in class A and the nodes in class B, these synonyms are It can be treated as the same keyword.

After all, in order for the dynamic link L2 to be defined between the nodes N1 and N3 in the disconnected part shown in FIG. 16, the relevance between the keyword K1 of the node N1 and the keyword K3 of the node N3 must be determined. , Thesaurus dictionary T
Evaluation is performed with reference to ab, and it is necessary that the evaluation result satisfies a predetermined condition.

In FIG. 16, the unconnected portion related to the node N1 is not limited to between the nodes N1 and N3. Since the only node connected to the node N1 by the existing static link is the node N2, the other nodes, that is, between the nodes N1 and N4, the node N1
The portions between 1 and N5, between the nodes N1 and N6, and between the nodes N1 and N7 are also unconnected portions related to the node N1. However, in the present embodiment, the occurrence of a dynamic link is restricted by imposing a condition that “a dynamic link can be defined only between classes for which a class link is defined”. Therefore, in the case of the above-described example, it is allowed to define a dynamic link between the nodes N1 and N3 and between the nodes N1 and N4 among the disconnected portions related to the node N1 if the evaluation result of the keyword satisfies the condition. (Because the class link AB exists between the classes AB), for example, the node N
As shown in FIG. 17, it is not allowed to define a dynamic link between 1 and N7 (because there is no class link between classes BC). For the same reason, between nodes N1-N5, between nodes N1-N6, N2-
Definition of a dynamic link is not allowed between N5, between nodes N2 and N6, between nodes N2 and N7, between nodes N5 and N6, and between nodes N5 and N7.

A predetermined weight, that is, a signal transmission coefficient is given to the dynamic link temporarily defined at the time of the search. Therefore, a certain signal transmission coefficient is also given to the dynamic link L2 shown in FIG.
As for the signal transmission function, it performs exactly the same function as other static links. The signal transmission coefficient given to the defined dynamic link is, for example, “uniform 5
0% ", but a coefficient corresponding to the evaluation value of the relevancy of the keyword may be given. For example, the signal transmission coefficient of the dynamic link L2 shown in FIG. 66%. " When the equivalent keyword is used, the weight may be determined for each equivalent keyword, and the value of the signal transmission coefficient may be determined according to the weight of the matched equivalent keyword.

The handling at the time of learning the dynamic link temporarily generated in this way is as described in §3. That is, the dynamic link
When used as a path from the node of interest to the adopted node, a modification is made to promote the link to a static link and increase the signal transfer coefficient. On the other hand, if it is not used as a path from the focused node to the adopted node, it is deleted as it is. For example, as shown in FIG. 16, at the time of a search using the node N1 as a target node,
Consider the case where the dynamic link L2 is defined. In this case, for example, when the node N4 which is a candidate node is adopted, the dynamic link L2 remains as the static link L2, and the signal transmission coefficient is also increased and corrected. However, another candidate node, node N
If 2 is adopted, it will disappear as it is.

As a result, by temporarily defining a dynamic link having the above-described characteristics at the time of a search, the range of nodes to be searched can be expanded. Also, by leaving a dynamic link involved in a user's selection action as a static link, it is possible to newly add a link that is useful for each user.

§5. The database system according to the present invention
Next , the operation procedure of the database system according to the present invention will be described based on a flowchart.

<5.1: Procedure for Using Database System According to the Present Invention> FIG. 18 is a flowchart showing the procedure for using the database system, mainly focusing on user operations. First, in step S1, the user determines an initial target node. Any method may be used to determine the initial focused node. For example, when a keyword is input, a node in which a keyword that is the same as or related to the input keyword is defined is displayed as a candidate (actually, a keyword is displayed). From the displayed keywords, If a specific node is adopted by the operator, an initial target node can be determined. By adopting a method in which a class is specified in advance and an initial target node is selected from the nodes in the class, it is possible to determine the initial target node by narrowing down the candidates.

In the following step S2, it is determined whether or not to use the data associated with the node of interest.
If the operator wants to use (browse) data for the current node of interest, an instruction to that effect may be given. When an instruction to use data is given from the operator, the process proceeds from step S2 to step S3, and data corresponding to the node of interest is provided. That is, in the system shown in FIG. 1, based on the keyword defined for the target node in the node / link aggregate 2,
The corresponding data is extracted from the database 1 and presented to the operator 3. If necessary, data can be used repeatedly.

In the following step S4, it is determined whether or not to search for this node of interest. If the operator has not performed the search, the use procedure of this system is temporarily ended. If the operator accesses this database system again to check for another content, the procedure from step S1 may be started again. If the operator wants to search the current node of interest, the operator is instructed to that effect. When a search instruction is given from the operator, the process proceeds from step S4 to step S5, and a search process is performed. The detailed procedure of the search processing in step S5 will be described later with reference to the flowchart of FIG. By this search processing, related nodes determined to have a certain degree of relevance to the node of interest are extracted, and these related nodes are presented as candidate nodes.

In practice, the candidate nodes are presented by displaying keywords defined for each candidate node on a display. The operator selects the keyword that seems to be most relevant to the data being sought while viewing the display of the keyword. That is, the candidate node selection process in step S6 is performed.
Subsequently, based on the operator's selection action, step S
At 7, a learning process is performed. The detailed procedure of the learning process in step S7 will be described later with reference to the flowchart of FIG.

When the learning process is completed in this way, in step S8, an update process is performed with the adopted node as a new target node, and the procedure returns to step S2.

After all, by repeating the procedure of the flow chart shown in FIG. 18, the operator can repeatedly execute the search and change the node of interest from one to another. If an instruction to use (view, print, etc.) the data is given in step S2 as needed, the data corresponding to the current node of interest can be used each time. The feature of such a search method is that the search is not for data directly but for a related keyword. As described above, the process of presenting a specific node to an operator is actually a process of displaying a keyword defined for the specific node on a display. Therefore, the operation of changing the node of interest from one operation to another is an operation of changing the keyword from the viewpoint of the operator. Such a search method is effective when it is unclear what keyword should be used to finally search for the data that is sought. In other words, the operator gives a keyword that is considered to be relevant for the time being as a keyword for determining the initial node of interest in step S1, thereafter executes the search processing in step S4, and presents it in step S6. The task of selecting the most relevant keyword from among the new keywords (selection of a node) may be repeatedly executed.

Specifically, for example, consider a case in which past case data similar to a case of a specific patient is searched. Now, it is assumed that the operator designates a case database of a particular hospital as a class, inputs a keyword of “high blood pressure”, and performs a search. As a result, suppose that “abnormal blood pressure”, “juvenile hypertension”, “senile hypertension”, “hypertensive retinopathy”, and the like are presented as candidates as keywords indicating candidate nodes (processing inside the system is , The node where the keyword “hypertension” is defined becomes the node of interest, and the node where the keywords such as “abnormal blood pressure”, “juvenile hypertension”, “senile hypertension”... Are presented as candidate nodes Become). Here, for example, since the patient is an elderly person, it is assumed that the keyword “senile hypertension” is adopted (as a process inside the system, a candidate node in which the keyword “senile hypertension” is defined is adopted, It becomes a new node of interest). If necessary at this point, the operator can browse data corresponding to the keyword “senile hypertension” (providing data corresponding to the node of interest).

Here, as a result of further searching based on the keyword “senile hypertension”, as keywords indicating candidate nodes, “arteriosclerosis”, “fundus bleeding”, “renal dysfunction”, “heart failure” , ..., etc. are presented as candidates. At this time, if an abnormality is found in the renal function of the patient, the keyword “renal dysfunction” can be adopted, and if necessary, data corresponding to this keyword (for example, The patient's case data can be viewed.

As described above, in the database system according to the present invention, if a search is performed by giving a certain keyword for the time being, another keyword related to the given keyword is presented as a candidate from the system side. Even in a case where an appropriate keyword does not come to mind on the operator side, a more flexible search can be performed. In addition, since the system itself has a learning function, the more likely the keyword is used, the more likely a keyword that is to be adopted is presented preferentially, so that usability is further improved.

It should be noted that not only a character string but also an image can be used as a keyword. For example, if the data associated with a certain node is represented by a simple icon, and this icon is defined as a keyword of that node, when displaying search results on the display, the icon as a keyword is displayed. Can be arranged and presented to the operator. in this case,
The operator's adoption can be performed by a simple operation of clicking an icon with a mouse pointer or the like.

In the flow chart shown in FIG. 18, after the node is selected in step S6, the learning process in step S7 is performed unconditionally. When the learning process is performed in, unintended learning may be performed. For example, the operator has adopted the node in step S6, but since the adoption was unwilling,
Let us consider a case where neither the data use nor the new search processing is performed for this adopted node. In such a case, the adopting act itself is contrary to the intention of the operator, and it is not preferable that the learning process of step S7 is performed by such an unwilling adopting act. In order to deal with such an adverse effect, after the adoption action in step S6, the learning process is not immediately performed, and the data use (step S3) or the search process (step S3) using the selected node as a new target node is performed. The learning process may be performed only when S5) is performed.

<5.2: Procedure of Search Process> FIG. 19 is a flowchart showing a detailed procedure of the search process in step S5 in the flowchart shown in FIG. Here, the procedure of this search processing will be described for a specific example in which a static link as shown in FIG. 16 is defined. First, in step S11, the hop number H is set to an initial value 1. Subsequently, in step S12, the starting node is set to 1
Extract one. At first, the target node N1 is extracted as a starting node. In the next step S13, one target node for this originating node N1 is extracted. Theoretically, in this step S13, all nodes other than the starting node are sequentially extracted as target nodes. In the example shown in FIG. 16, all the nodes N2 to N7 other than the originating node N1 are sequentially extracted as target nodes. In step S13, only one target node is extracted. Therefore, it is assumed here that the node N2 is extracted as a target node in numerical order.

After the originating node N1 and the target node N2 have been extracted, various determinations are made between these nodes in steps S14 to S17. First, in step S14, it is determined whether there is a class link between both nodes. If there is no class link between the nodes, the process proceeds to step S19. In the example shown in FIG. 16, a class link (local link) BB is provided between the nodes N1 and N2 (to be precise, between the class B to which the node N1 belongs and the class B to which the node N2 belongs). Exists, so step S1
It will advance to 5. In step S15, it is determined whether there is a static link between both nodes. In the example shown in FIG. 16, since the static link L1 exists between the nodes N1 and N2, the process proceeds from step S15 to step S19.

Subsequently, from step S19 to step S1
3, the target node N3 is extracted, and various determinations are made with respect to the start node N1 and the target node N3. First, in step S14, a class link (remote link) AB is set between the nodes N1 and N3 (more precisely, between the class B to which the node N1 belongs and the class A to which the node N3 belongs). Since it exists, the process proceeds to step S15. However, since there is no static link between the nodes N1 and N3, the process further proceeds to step S16. Since there is no dynamic link between both nodes yet, the process further proceeds to step S17, where the relevance of the keywords of both nodes is determined. Is evaluated. Here, assuming that the evaluation result of the relevance between the keyword K1 defined for the node N1 and the keyword K3 defined for the node N3 satisfies a predetermined condition, the process proceeds to step S18, where the nodes N1-N3 In the meantime, the dynamic link L indicated by a broken line in FIG.
2 will be defined.

After all, the determination processing in steps S14 to S17 is processing for determining whether a dynamic link should be defined between both nodes. One of the conditions for defining a dynamic link is, as described above, "there must be a class link between the classes to which both nodes belong". If this condition is not satisfied, the process proceeds from step S14 to step S14. It will jump to S19. Also, since a dynamic link is defined for a node at a disconnected portion where no existing static link exists, if a static link already exists between both nodes, the process jumps from step S15 to step S19. Will do. Similarly, if a dynamic link has already been defined (for example,
In a later procedure, the node N3 is a starting node, a node N
In the case of processing with 1 as the target node, a dynamic link L2 has already been defined between the two nodes), and the process jumps from step S16 to step S19. Thus, in step S17, if the keywords of both nodes satisfy the condition of relevance,
Since neither a static link nor a dynamic link has been defined between the two nodes, a dynamic link is defined in step S18.

In the flowchart shown in FIG. 19, the target nodes are extracted one by one in step S13, and it is determined in step S14 whether or not a class link exists for the extracted target nodes. However, in performing the actual arithmetic processing, first, one target class is extracted, and it is first determined whether a class link exists for the extracted target class, and the class link exists. In this case, the target nodes are extracted one by one from the target class, and the process proceeds to step S15 and subsequent steps.
If there is no class link, efficient processing can be performed by not extracting the target node from the class at all.

When the process of extracting all target nodes N2 to N7 is completed for the first origin node N1, the process proceeds from step S19 to step S20, and it is determined whether or not all origin nodes have been extracted. The “origin node” means a node that is the origin of signal transmission, and the initial origin node is only the node N1 that is the target node.
In step S20, it is determined that all the originating nodes have been extracted, and the process proceeds to step S21.

In step S21, a signal is transmitted from the originating node for one hop, and a node whose signal value is equal to or higher than a predetermined level is set as a new originating node. In the example shown in FIG. 16, if signal transmission for one hop is performed from the originating node N1, signal transmission from the node N1 to the node N2 via the existing static link L1 and signal transmission from the node N2 via the dynamic link L2 just defined are performed. Signal transmission from the node N1 to the node N3 is performed. As described above, the signal value of the signal at the originating node is attenuated by the signal transmission coefficient of the link. Therefore, among the nodes to which the signal is transmitted, the signal value is equal to or higher than a predetermined level (in the example described in §4,
Only the node having a signal value of 10 or more) is set as a new starting node. Also, if there is a node whose signal value is less than 10, it is handled as if there was no signal transmission to that node.

In the following step S22, it is determined whether or not the hop number H has reached the upper limit value Hmax. For example, H
Assuming that max = 2 has been set, H = 1
Then, the process proceeds from step S22 to step S23, in which the number of hops H is increased by one, and step S1 is performed.
The processing from Step 2 is repeated.

In the case of the example shown in FIG. 16, nodes N2 and N which have received a signal of one hop from the original originating node N1.
3 is the second generation origin node. In step S12, it is assumed that the start node N2 is extracted from the two start nodes. In the following step S13, one target node is determined for the originating node N2,
In the determination processing of 14 to S17, it is determined whether a dynamic link can be defined. Here, it is assumed that no dynamic link is defined for any of the target nodes between the origin node N2 (in the illustrated example, the presence of the remote link AB causes the node N2-
A dynamic link may be defined between N4, but here, it is assumed that there is no sufficient relationship between the keywords K2 and K4.) Then, the process returns from step S20 to step S12, in which the origin node N3 is extracted and the possibility of defining a dynamic link is similarly determined, but it is assumed that a new dynamic link has not been defined.

Thus, since the extraction of all the originating nodes N2 and N3 is completed, the process proceeds from step S20 to step S21. In step S21, a signal for one hop is transmitted from the two origin nodes N2 and N3. (Since no signal is transmitted again to the link already used for signal transmission, it is returned from the nodes N2 and N3 to the node N1. No signal transmission is performed), and a process of setting a node whose signal value is equal to or higher than a predetermined level as a new starting node is performed. FIG.
In the example shown in FIG. 6, signal transmission is no longer performed from the originating node N2. On the other hand, if signal transmission for one hop is performed from another start node N3, signal transmission from the node N3 to the node N4 is performed via the existing static link L3. Here, if the signal value of the signal arriving at the node N4 is equal to or higher than a predetermined level, the node N4 becomes a new starting node.

In the following step S22, it is determined whether or not the hop number H has reached the upper limit value Hmax. Here, H
Since the setting of max = 2 has been made, the hop number H has reached the upper limit, and the process proceeds to step S24. In this step S24, all nodes on the path (on the path through which the signal flows) are extracted as related nodes. In the case of the example shown in FIG. 16, signals are transmitted to the nodes N2, N3, and N4 using the links L1, L2, and L3 as paths, and these nodes N2, N3, and N4 are extracted as related nodes. Become. As described above, the related node is presented to the operator as a candidate node. If the signal value is less than the predetermined level even if the signal is transmitted, the signal is treated as if the signal was not transmitted to the node, and the node is not extracted as a related node. .

<5.3: Procedure of Learning Process> FIG. 20 is a flowchart showing a detailed procedure of the learning process in step S7 in the flowchart shown in FIG. First, in step S31, all paths through which a signal has flowed by the search processing are extracted as learning target paths. (Even if there is signal transmission, if the signal value is less than a predetermined level, signal transmission to that node Is treated as if it did not exist, so the path to that node is not a learning target path.) In the example illustrated in FIG. 16 described above, when the candidate nodes N2, N3, and N4 are searched by the search using the node N1 as the target node, the links L1, L2, and L3 are extracted as the learning target paths.

Subsequently, in step S32, among the learning target paths, for the link on the path from the node of interest to the adopted node, the signal transmission coefficient is increased in the case of the static link, and is increased in the case of the dynamic link. , The signal transmission coefficient is increased, and this dynamic link is promoted to a static link. In the example shown in FIG. 16 described above, when the candidate node N4 is adopted, among the links on the path from the target node N1 to the adopted node N4,
For the dynamic link L2, learning to increase the signal transmission coefficient and to elevate it to a static link is performed, and for the static link L3, learning to increase the signal transmission coefficient is performed.

On the other hand, in step S33, among the learning target paths, for links on paths other than the above, the signal transmission coefficient is reduced in the case of a static link, and the link itself is extinguished in the case of a dynamic link. Is performed. In the case of the example shown in FIG. 16 described above, learning for reducing the signal transfer coefficient is performed for the static link L1.

Finally, in step S34, learning relating to the frequency coefficient of the node is performed. That is, of all the nodes on the learning target path, the frequency coefficient is increased for the adopted node, and the frequency coefficient is decreased for the other nodes. In the case of the example shown in FIG. 16 described above, learning for increasing the frequency coefficient for the adopted node N4 is performed, and learning for decreasing the frequency coefficient for the other candidate nodes N2 and N3 is performed.

§6. The database system according to the present invention
Specific Configuration of System <6.1: Basic System> FIG. 21 is a block diagram showing a specific configuration of a database system according to an embodiment of the present invention. Actually, the essence of the database system according to the present invention is realized by software. However, here, for convenience, this system will be described as a set of a plurality of functional elements. The individual blocks shown in FIG. 21 indicate individual functional elements constituting the system, and are actually constructed by software. Therefore, the individual blocks shown in FIG. 21 are not directly associated with specific hardware components constituting the database system. For example, hardware (storage device) for storing a keyword defined for each node
Further, a display device for presenting various information to the operator, an input device for inputting instructions from the operator, and the like are not shown as independent functional elements, but as hardware components, these are: Naturally, they are included in the present system.

In this database system, a node aggregate including a plurality of nodes is defined, and data is stored in association with each node. Data storage means 10
Is storage means for storing data to be provided in association with individual nodes. The data providing unit 20 has a function of extracting data associated with a predetermined node of interest from the data storage unit 10 and providing the extracted data to the operator 100. The target node is a target node setting unit 3
Set by 0. That is, the operator 100
Gives an instruction to set the specific node as the target node to the target node setting means 30, the specified node is set as the target node, and the operator 100 determines which node is the target node. (Actually, a keyword defined for the node of interest). When the operator 100 wants to use the data associated with the current node of interest, the data providing request may be made to the data providing unit 20. The data providing unit 20 extracts data associated with the current node of interest set in the node of interest setting unit 30 from the data storage unit 10 in response to the provision request, and provides the extracted data to the operator 100. Provide as data (eg, display on display).

In response to a search request from the operator, the search means 40 stores candidate nodes related to the current node of interest set in the target node setting means 30 in the link stored in the link storage means 50. It has a function to search by referring to the aggregate. In the link storage means 50,
A node / link aggregate including a set of nodes and a set of links (static links) indicating the degree of association between nodes is stored. When the operator 100 issues a search request for the current node of interest, the search means 4
0 performs a process of using the link aggregate in the link storage unit 50 to search for a related node under a predetermined condition with respect to the target node set in the target node setting unit 30; A set of related nodes is extracted as a related set. The specific method of this search processing is described in §4.
3 or §5.2.
In addition, the search unit 40 has a function of generating a dynamic link separately from the static link in the link storage unit 50, and has a function of evaluating the relevance between nodes with reference to a thesaurus dictionary. (See §4.7).

When a set of related nodes is extracted in the search processing by the search means 40, the candidate presenting means 60 presents all or a part of the nodes belonging to the related set to the operator 100 as candidate nodes. In the examples described so far, all of the related nodes belonging to the related set have been presented as they are to the operator 100 as the candidate nodes. However, if necessary, the candidate presenting means 60 performs screening to determine the specific conditions. Only some related nodes that are satisfied may be presented to the operator 100 as candidate nodes. A specific example of such sieving is
This will be described in §7 “AND search”. The candidate presentation to the operator 100 by the candidate presentation means 60
For example, it can be performed by displaying a keyword defined for each candidate node on a display.

The candidate selecting means 70 has a function of causing the operator 100 to select a specific candidate node from the candidate nodes presented by the candidate presenting means 60.
When the operator 100 gives an instruction to adopt a specific candidate node, the candidate selecting unit 70 transmits the designated candidate node to the updating unit 80 and the learning unit 90 as the selected node. The updating unit 80 performs a process of updating the setting of the target node setting unit 30 so that the adopted node becomes a new target node. In the embodiment described in this specification, the number of adopted nodes is limited to one.
It is also possible to allow adoption of a plurality of candidate nodes (in this case, each of the plurality of adopted nodes becomes a new node of interest). On the other hand, the learning means 90 corrects the degree of association for the link on the path from the target node to each related node. That is, a process of correcting the link weighting is performed on the link aggregate in the link storage unit 50.

Thus, when using this system, the operator 100 first gives an instruction for setting the initial target node to the target node setting means 30 to set the initial target node, and thereafter, if necessary, A data provision request to the data provision means 20 or a search means 40
When a candidate node is presented by the candidate presenting means 60, a selection instruction is given to the candidate selecting means 70 to determine a new node of interest from the presented candidates. Processing will be performed. Each time the operator 100 selects a new node of interest, the link aggregate stored in the link storage unit 50 is corrected, and the weight of the link on the path from the old node of interest to the new node of interest is increased. You.

In the link storage means 50, a separate link set is prepared for each operator (user), and the learning means 90 sets a link set corresponding to the operator who has adopted the candidate node. Will be corrected. Therefore, learning is performed separately and independently for each operator, and the more this system is used, the more the usability for each operator is improved.

<6.2: Definition and Correction of Signal Transfer Coefficient> As described above, each link (static link) constituting the link aggregate stored in the link storage means 50 is given a weight. Predetermined signal transmission coefficients are defined as the indicated values. Search means 40
Performs a process of transmitting a signal having a predetermined signal value from a target node to another node along a link when searching for a related node for the target node. In this signal transmission process, the transmitted signal value is increased or decreased based on the signal transmission coefficient defined for each link, and only the node to which the signal having the signal value equal to or higher than the predetermined level is transmitted is extracted as the related node. Will be. Nodes whose signal values have fallen below a predetermined level are treated as if signal transmission was not performed, and no signal transmission processing is performed further downstream. Signal transmission between nodes connected by one link is defined as the number of hops 1, and also when the number of hops exceeds a predetermined upper limit, the signal transmission processing is stopped. Therefore, even if the signal value is equal to or higher than the predetermined level, the signal transmission processing is not performed on the node whose hop count from the target node exceeds the predetermined upper limit.

As described above, by the signal transmission processing from the node of interest, only the nodes that have transmitted the signal having the signal value equal to or higher than the predetermined level are extracted as the related nodes by the search means 40, and the relevant nodes are extracted. The candidate presenting means 60 presents them to the operator 100 as candidate nodes. When the candidate presenting means 60 presents the candidate nodes, the presentation based on the priority order is performed. That is, the candidate presenting means 60 has a function of defining a priority order for each candidate node according to the signal value of a signal transmitted to each candidate node, and presenting the candidate node based on the priority order. Specifically, a keyword defined for a candidate node having a higher priority order may be preferentially displayed on the display (for example, displayed on the upper side of the list). In the end, such a priority display is used to preferentially display a candidate node that is likely to have been adopted in the past in a search for the current node of interest, and a candidate having a high possibility of being adopted is selected again. An easy presentation will be performed.

The learning means 90 corrects the signal transmission coefficient of each node based on the selection action of the operator 100.
In other words, the signal transmission coefficient of the link on the path from the node of interest to the adopted node is modified to be relatively increased with respect to the signal transmission coefficients of the other links. More specifically, the learning means 90 defines all paths from the target node to each related node as learning target paths, and among the links on the learning target path, the links on the path from the target node to the adopted node Is modified to increase its signal transmission coefficient, and to reduce the signal transmission coefficient of other links on the learning target path.

<6.3: Search Using Dynamic Link> The search means 40 performs not only a search using links (static links) constituting a link aggregate in the link storage means 50, but also It has a function to search using another link (dynamic link) generated temporarily (see §3 or §4.7). That is,
For a node that is completely connected to the node of interest by the existing static link, a search using this existing static link is performed, and the node is completely connected to the node of interest by the existing static link. No, for nodes where there is a disconnected part,
Evaluate the relevance between the nodes of the disconnected portion, and if the evaluation result satisfies a predetermined condition, temporarily define a dynamic link in the disconnected portion, and search using both the static link and the dynamic link I do.

The evaluation of the association between the nodes of the disconnected part is as follows.
It is determined based on the degree of relevance of the keyword defined for each node. For example, if a static link is not defined between the first node and the second node, but constitutes a disconnected portion, a first keyword defined in the first node, The degree of association with the second keyword defined in the second node (e.g., the degree of matching of character strings) is evaluated, and if the evaluation result satisfies a predetermined condition (e.g., the mutual character strings If they match at a ratio or more), a temporary dynamic link is defined between both nodes, and a signal transmission process is performed via this dynamic link. The signal transfer coefficient for the temporarily defined dynamic link may be determined based on the evaluation result of the relevance of the keyword. For example, the higher the matching between the character strings of both keywords, the larger the signal transmission coefficient is defined. Further, if a plurality of equivalent keywords are defined in each node, and all the equivalent keywords are to be evaluated, more flexible relevance evaluation can be performed.

When performing the learning process by the learning means 90, the existing static links constituting the link aggregate stored in the link storage means 50 and the temporarily generated dynamic links are respectively different from each other. Is handled. That is, the learning related to the static link is as described above. For the dynamic link, when the related node extracted by the search using the dynamic link is adopted by the candidate adoption unit, the dynamic link becomes the static link. The link is added to the link aggregate and modified to become a new component. At this time, the signal transfer coefficient defined for the dynamic link is increased and corrected, and the corrected signal transfer coefficient is given as the signal transfer coefficient of the promoted static link. On the other hand, if the related node extracted by the search using the dynamic link is not adopted by the candidate adoption means, the dynamic link is deleted as it is.

<6.4: Definition of Class Link> When used as a distributed database system, §4.1
As described above, it is preferable to classify the individual nodes into a plurality of classes so that each node belongs to one of the classes, and define a class link indicating the degree of association between the classes. In the example shown in FIG. 8, remote links AB and AC indicating the degree of association between different classes and local links AA and B indicating the degree of association between self and self.
B and two types of class links are defined. When determining whether or not to define a dynamic link at the time of search, the relevance between nodes in the disconnected part is evaluated. When the relevance is evaluated, each node belongs to the node. You can refer to the class link defined between the classes. For example, if the definition of the dynamic link is not permitted between the classes having no class link as in the above-described embodiment, the generation mode of the dynamic link can be controlled by the definition of the class link. become.

Also, a thesaurus dictionary is prepared corresponding to each class link, and a class defined between the classes to which each node belongs when evaluating the relevance between the nodes of the disconnected part. Using a thesaurus dictionary corresponding to the link, the relevancy of the keyword can be evaluated. Depending on the contents of the prepared thesaurus dictionary, the evaluation result of the relevancy of the keyword changes, and the direction of the relevancy evaluation method can be determined.

Further, a signal transmission coefficient is also defined for each class link, and when the search means transmits a signal between the nodes, the signal transmission coefficient of the link defined between the nodes and the signal transmission coefficient are defined between the classes. If the signal value is increased or decreased in consideration of both the defined signal transmission coefficient of the class link and the signal transmission tendency, the tendency of the signal transmission is generally controlled by the class link as described in §4.4. It becomes possible to do.

<6.5: Addition of Frequency Coefficient Storage Unit> FIG.
2 corresponds to the database system shown in FIG.
FIG. 3 is a block diagram of an embodiment in which a frequency coefficient storage unit 110 is further added. The frequency coefficient storage unit 110 is a unit that stores a frequency coefficient indicating the adoption frequency of each node. When the operator 100 determines an adopted node from the candidate nodes presented by the candidate presenting means 60, the learning means 90 sets the frequency coefficient of the selected node relative to the frequency coefficients of the other nodes. Make corrections to increase The frequency coefficient for each node is a coefficient indicating the weight of the node.
The value of the frequency coefficient increases as the node has a higher frequency of adoption by 00, and the weight increases. Such weighting of the nodes does not affect the signaling process itself, but is used to determine the priority when presenting candidate nodes, as described in §4.6. That is, the candidate presenting means 60 in this embodiment defines a priority order for each candidate node according to both the signal value of the signal transmitted to each candidate node and the frequency coefficient for each candidate node. It has a function of presenting candidate nodes based on priorities.

The frequency coefficient storage means 110 stores a separate frequency coefficient for each operator, and the learning means 90 corrects the frequency coefficient corresponding to the operator who has adopted the node. To do. Therefore, similar to the transmission coefficient for the link, learning is performed separately and independently for each operator, and the more the system is used, the more the usability for each operator is improved. Become.

§7. AND Search The database system according to the present invention is characterized by its unique search method. As described above, in order to search for desired data using this database system, a keyword that is considered to be relevant is input for the time being, and a search process for this keyword (node of interest) is performed. The keyword (candidate node) having the possibility is presented. The operator selects the target data from the presented keywords,
Keywords that are considered more relevant can be adopted, and further relevant keywords can be searched.

FIGS. 23 to 28 are conceptual diagrams for explaining such a search operation unique to the present invention. In the database system according to the present invention, a node aggregate as shown in FIG. 23 is defined. In the figure, a large number of black circles indicate individual nodes, and each node is defined with a predetermined keyword and associated with specific data. The target data that the user is seeking is associated with one of these nodes,
The search operation performed by the operator is an operation for finding a node to which target data is associated.

Now, it is assumed that the operator inputs a keyword that is considered to be relevant to the target data, and specifies the target node N1 corresponding to the keyword. FIG.
Indicates a state in which one node in the node aggregate is designated as the target node N1. Of course, if the target node N1 is a target node, a provision request (browsing instruction) is given to the data providing means 20, and the target data associated with the target node N1 is stored in the data storage means. 10 will be extracted and presented. When the operator gives a search request to the search means 40, the search is performed for the target node N1, and a related set G1 including nodes having a certain degree of relevance to the target node N1 is extracted. FIG. 25 shows the related set G1 searched in this way. The related nodes in the related set G1 thus found are presented to the operator as candidate nodes by the candidate presenting means 60. The presentation order of the candidate nodes follows the priority of each node.

The operator adopts one of the presented candidate nodes. FIG. 26 shows a state where the node N2 in the related set G1 is adopted. Node N adopted here
2 becomes a new node of interest as shown in FIG.
At this time, learning to increase the weight of the link on the path from the node N1 to the node N2 is performed. If a search is performed again with the node N1 as the target node in the future, the search in the related group G1 will be performed. The priority of the node N2 when presenting the node as a candidate node is improved.

The operator can make a request to provide data associated with the new node of interest N2 and browse the data as required. When a search request is made again to the new node of interest N2, a related set G2 including nodes having some degree of relevance to the node N2 is extracted. FIG. 28 shows the related set G2 searched in this way. The related nodes in the related set G2 thus found are presented to the operator as candidate nodes by the candidate presenting means 60. Note that the nodes N1 and N
Since the weight of the link on the path between the link 2 and the link 2 is increased, the original target node N1 is usually included in the related set G2. Thus, the original node of interest is
If it is not preferable to be presented again as a candidate node in a later search, in the series of search operations, a function of selecting a past node of interest from the candidate node may be added to the candidate presenting means 60. I just need.

In the example described above, the candidate presenting means 60 presents all the related nodes extracted by the searching means 40 to the operator 100 as candidate nodes as they are. However, as described above, if the candidate presenting means 60 is provided with a selection function, the search means 40
It is possible to present only some of the extracted related nodes as candidate nodes. The AND search described below uses the selection function of the candidate presenting means 60.

The AND search described here is an effective means for gradually narrowing down candidates in the process of performing a search operation using the database system according to the present invention. For example, in the example of the general search operation described above,
A related set G1 shown in FIG. 25 and a related set G shown in FIG.
2 is a set that is separately and collectively independent, and no candidates have been narrowed down even though the search operation has been performed twice. As a search method to narrow down the candidates,
Usually, an AN that sets a plurality of conditions for determining candidates and extracts only candidates that satisfy the logical product of these conditions
A D search is performed. The search method described here applies this general AND search to the search of the database system according to the present invention.

FIG. 29 is a block diagram of an embodiment in which a population definition means 120 is added to the database system shown in FIG. Population definition means 120
Has a function of defining a mother set indicating a specific node aggregate. The candidate presenting means 60 in this embodiment is defined in the related set extracted by the search means 40 and the mother set defining means 120. It has the function of presenting only nodes belonging to the intersection with the mother set as candidate nodes. When this intersection set is presented as a candidate node by the candidate presenting means 60, the mother set defining means 12
0 has a function of redefining the set of candidate nodes as a new population.

Hereinafter, the search operation by the AND search will be described.
A specific example will be described. Now, consider a state in which the node N1 in the node aggregate as shown in FIG. 30 is designated as the node of interest (the node aggregate includes many nodes as in FIG. In the description,
To avoid clutter, only the nodes of interest in each figure are shown with black circles). In this initial state, the entire node aggregate is defined as a mother set M1. Therefore, at this time, the population definition means 120
Defines a mother set M1 corresponding to the entire node aggregate.

Here, when a search request is made to the noted node N1, as shown in FIG. 31, an aggregate of related nodes is extracted as a related set G1, as described above. The related set G1 given by the search means 40 to the candidate presenting means 60 includes all the related nodes extracted in this way. On the other hand, the candidate presenting means 60 presents only nodes included in the logical product set “(G1) AND (M1)” of the related set G1 and the mother set M1 as candidate nodes. In the case of the example shown in FIG.
Since the mother set M1 at this point is the entire node aggregate, the candidate presenting means 60 presents all the related nodes included in the related set G1 as the candidate nodes as they are. That is, the nodes in the area indicated by hatching in FIG. 31 are presented as candidate nodes.

When the candidate nodes are thus presented,
The mother set defining means 120 defines this set of candidate nodes as
Redefine as a new population. Therefore, in this case, the related set G1 indicated by hatching in FIG. 31 is defined as a new mother set M2.

Now, here, the node N among the candidate nodes
If node 2 is adopted, this node N2 becomes a new node of interest. Now, it is assumed that the operator again gives a search request to this new node of interest N2. Further, it is assumed that a search request in the "AND search" mode is given (when the operator gives a search request to the search means 40, it is possible to specify the "normal search" mode or the "AND search" mode. You can do so). The search means 40 executes a search process for extracting a related set G2 including all related nodes for the target node N2. FIG.
2 indicates the related set G2 extracted in this way.

The candidate presenting means 60 uses the related set G2
AND set “(G2) AND (M
2) ”(nodes included in the hatched region in FIG. 32) are presented as candidate nodes. As a result, the candidate nodes presented at this time are only the nodes included in the intersection of the related set G1 at the time of the first search and the related set G2 at the time of the second search. This means that the search was narrowed down. When such a candidate node is presented, the mother set definition means 120 redefines the aggregate of the candidate nodes as a new mother set. So, in this case,
A set of nodes included in the area indicated by hatching in FIG. 32 is defined as a new mother set M3 (see FIG. 3).
3).

Here, the node N3 among the candidate nodes
Is adopted, this node N3 becomes a new node of interest. Then, it is assumed that the operator gives a search request to the new node of interest N3 again in the “AND search” mode. In this case, the search means 40 sets the related set G3 including all the related nodes to the target node N3.
Execute a search process for extracting. FIG. 34 shows the related set G3 extracted in this manner.

The candidate presenting means 60 calculates the relation set G3
AND set ((G3) AND (M
3)) (nodes included in the hatched region in FIG. 34) are presented as candidate nodes. As a result, the candidate node presented at this time is, as shown in FIG.
1, only the nodes included in the intersection set of the related set G2 at the time of the second search and the related set G3 at the time of the third search. Become. When such a candidate node is presented,
The mother set defining means 120 defines this set of candidate nodes as
Redefine as a new population. Therefore, in this case, a set of nodes included in the area hatched in FIG. 35 is defined as a new mother set M4.

Such an AND search is very effective when narrowing down using the database system according to the present invention. For example, at the time of the first search, a search is performed using a node defined as a keyword “hypertension” as a target node, and as a result, as keywords indicating candidate nodes, “abnormal blood pressure”, “arteriosclerosis”, “hypertension” It is assumed that “retinopathy”,... Are presented as candidates. Here, a node in which the keyword “arteriosclerosis” is defined is adopted as a new node of interest, and the second search is performed as “A
If the search is performed in the “ND search” mode, new keywords related to both the keyword “hypertension” and the keyword “arteriosclerosis” will be presented as candidates.
As described above, by using the AND search, candidates can be narrowed down gradually, and can be used as an effective search method for searching for target data.

§8. Negative Search Process The search process described so far is a search process for finding “items related to a certain keyword”. For example,
If a search is performed using a node defined as a keyword “hypertension” as a target node, a related node defined with keywords such as “abnormal blood pressure”, “arteriosclerosis”, “hypertensive retinopathy”, etc. can be found. . If such a search process is called "positive search process", the search process described here is a search process for finding "items not related to a certain keyword", and is called "negative search process". Should be.

<8.1: Basic Concept of Negative Search Processing> As described above, the node N1 of interest is designated from the node aggregate shown in FIG. 23 as shown in FIG. Is executed, a related set G1 composed of nodes having relevance to the target node N1 is extracted, and the nodes in the related set G1 are presented as candidate nodes (indicated by white node points). Therefore, the presented candidate node is a “node related to the node of interest N1”. On the other hand, as shown in FIG. 36, if the complementary set G1 ^* of the related set G1 is presented as a candidate node (indicated by a white node point), the presented node becomes
This means "a node not related to the node of interest N1". These nodes are "nodes having a relationship of being unrelated" if the feature of "unrelated" is considered as one of "relationships". Therefore, here, the so-called “relation in a general sense” is referred to as “positive relation”, and the “relation that is not related” is referred to as “negative relation”. Therefore, a node indicated by an outline node point in FIG. 25 is a “positive related node” having a “positive relationship” with respect to the target node N1, and the related set G1 is referred to as a “positive related set”. Will be. On the other hand, a node indicated by an outline node point in FIG. 36 is a “negative association” with respect to the target node N1.
And the related set G1 ^* is referred to as a “negative related set”.

Search means 4 shown in FIGS. 21, 22 and 29
0 is “positive search processing” for extracting “positive related set” and “negative search processing” for extracting “negative related set”
And a function for performing two types of search processing. Such “negative search processing” is particularly effective when combined with the AND search described in §7. Here, the aforementioned A
Let's combine "negative search processing" with the specific example described in the ND search.

First, the node N1 in the node aggregate as shown in FIG. 30 is designated as the node of interest, and "positive search processing" is performed as the first search. Related set G1 ”is extracted and presented as a candidate node. Further, when the node N2 is selected from the candidate nodes, the selected node N2 is designated as a new target node, and "positive search processing" is performed in the "AND search" mode as a second search. As shown in FIG. 32, the “positive related set G2” is extracted, and the logical product part (hatched part) with the mother set M2 is presented as a candidate node.
Subsequently, as shown in FIG. 33, the node N3 is selected from the candidate nodes, the selected node N3 is designated as a new target node, and "AND" is again performed as a third search.
When “positive search processing” is performed in the “search” mode, as shown in FIG. 34, “positive related set G3” is extracted, and a logical product part (hatched part in FIG. 34) with the mother set M3 is a candidate. Presented as a node. As shown in FIG. 35, by such three search processes, “(G1) AND (G2)
As described above, the nodes included in the logical product set “AND (G3)” are presented as candidate nodes.

However, in the state shown in FIG. 33, the node N3 is adopted as a new node of interest, and this new node of interest N3 is referred to as “A
If "negative search processing" is performed in the mode of "ND search", FIG.
As shown in FIG. 7, “negative related set G3 ^* ” is extracted,
The logical product of the mother set M3 and the negative related set G3 ^* (FIG. 3
7 (hatched portion 7) will be presented as candidate nodes. As a result, as shown in FIG. 38, "(G1) AND (G2) AND"
(G3 ^* ) "are presented as candidate nodes.

An AND search combining such “negative search processing” is very effective when performing a flexible narrow search. For example, consider a case in which past similar case data is searched for a patient having any cardiovascular disease. First, it is assumed that the first search is performed in the “positive search process” with a node defined as a keyword “circulatory disease” as a target node, and some candidate nodes are presented. Here, because the patient's blood pressure was high,
From these candidates, a node in which the keyword “hypertension” is defined is adopted, and “AND” is used as the second search.
Let's say you did a "positive search" in the "search" mode.
As a result, it is assumed that “abnormal blood pressure”, “arteriosclerosis”, “hypertensive retinopathy”,..., Etc. are presented as candidates as keywords indicating candidate nodes. Here, assuming that the patient does not have the symptoms of "arteriosclerosis", the third search adopts a node in which the keyword "arteriosclerosis" is defined and executes the keyword "arteriosclerosis". Is designated as the node of interest, and "A
"Negative search processing" may be performed in the mode of "ND search".
Then, candidate nodes presented as a result of the third search are narrowed down to nodes that are not related to the keyword “arteriosclerosis” (have a negative association).

In the system according to the present embodiment, when a search request is given to the search means 40, the mode is "normal search" (mode in which related nodes belonging to the extracted related set are presented as candidate nodes as they are). Or, "AND
Search ”mode (mode in which the candidate node at the previous search is used as a mother set, and only nodes belonging to the intersection of the extracted related set and mother set are presented as candidate nodes)
, And a second selection indicating "positive search processing" or "negative search processing", and a flexible search is performed by combining these two selections. The processing can be realized.

<8.2: Link Definition Considering Sign> One of the specific methods of performing the negative search process is as follows.
A method of extracting a positive related set G by performing a positive search process and extracting a negative related set G ^* as a complement thereof is conceivable. However, such a search method
This is not a method of directly finding the negative related set G ^* , but only a method of indirectly finding it using the positive related set G.
If such an indirect method is adopted, it becomes impossible to define the priority of the candidate nodes based on the link weight.

As described above, a signal transmission coefficient is defined for each link, a signal is transmitted from a node of interest, and a node having a signal value equal to or higher than a predetermined level is extracted as a related node. For example, a priority order based on the magnitude of the signal value can be defined for each related node, and when the candidate node is presented as a candidate node, the presentation can be performed in consideration of the priority order. However, when the negative search processing is performed by the indirect method described above, it becomes impossible to define the priority order for the negative related nodes based on the link weights. For example, in the case of the example shown in FIG. 36, let us consider a case in which, when obtaining the negative related set G1 ^* , a method is first employed in which the positive related set G1 is first obtained and then the complement set thereof is obtained indirectly. In this case, the node that has received signal transmission from the node of interest N1 is only the node in the positive association set G1. Therefore, for the nodes in the positive related set G1, the priority can be defined based on the signal value of the transmitted signal, but for the nodes in the negative related set G1 ^* obtained as its complement. Cannot define priorities based on signaling.

The candidate node obtained by the negative search process is a node having a negative relevance (relevance of no relevance) to the node of interest, and the negative relevance is also positive. It is convenient to be able to treat quantitatively as well as relevance. That is, it is preferable that the same negative relevance can be distinguished according to the degree thereof, and a candidate node showing a stronger negative relevance can be preferentially presented. As described above, the inventor of the present application conceived to define two types of links, a positive link and a negative link, so that the negative association can be treated quantitatively in the same manner as the positive association. That is, a positive link having a positive signal transmission coefficient is defined between nodes having a positive relation, and a negative link having a negative signal transmission coefficient is defined between nodes having a negative relation. is there.

FIG. 39 is a conceptual diagram showing a node / link aggregate defining both such a positive link and a negative link. In the figure, a link with a “+” sign (shown by a solid line) is a positive link having a positive signal transmission coefficient, and a link with a “−” sign (shown by a broken line) is a negative link. This is a negative link having a signal transfer coefficient. For example, a link between the nodes A and B (hereinafter, referred to as a link AB) is a positive link, and the nodes A and B have a positive relationship (a so-called normal relationship). It is shown that. On the other hand, for example, link BD
Is a negative link, and indicates that the node B and the node D are nodes having a negative relationship (relation that there is no relationship). Each of the positive links has a positive signal transmission coefficient, and each of the negative links has a negative signal transmission coefficient. Signal transfer coefficients with absolute values are defined).

A search process using a node / link aggregate defining both the positive link and the negative link is performed as follows. First, in the case of performing a positive search process, that is, a process of searching for a node having a positive association with the target node, a signal having a positive signal value is transmitted from the target node to another node along the link. And a node from which a signal having a positive signal value is obtained may be extracted as a positive related node. FIG. 40 is a conceptual diagram showing a state in which a positive search process in which signal transmission is limited to the number of hops H = 1 is performed for the node B of interest. The code given in parentheses to each node indicates the code of the signal value of the signal obtained at each node, and the node denoted by “0” indicates a node that has not transmitted a signal. Bold arrows indicate signal transmission paths, solid arrows indicate signal transmission paths along the positive link, and broken arrows indicate signal transmission paths along the negative link.

Since the positive links BA, BE, and BJ have positive signal transmission coefficients, when a signal having a positive signal value is given to the target node B, the positive signals are applied to the nodes A, E, and J. A signal having a signal value will be transmitted. On the other hand, since the negative links BD and BG have negative signal transmission coefficients, when a signal having a positive signal value is given to the target node B, the negative signal value is applied to both the nodes D and G. Will be transmitted. Other nodes C,
No signal is transmitted to F, H, I, and K under the condition that the number of hops is H = 1. Here, if nodes A, E, and J from which a signal having a positive signal value is obtained are extracted as positive related nodes and presented as candidate nodes, a positive search process is executed. . In practice, the signal value of the signal given to the target node B is multiplied by the signal transmission coefficient of each node, so that the signal is attenuated.
A node for which a signal value equal to or higher than a predetermined level is not obtained is treated as having no signal transmission. However, here, for convenience of explanation, such signal attenuation is not considered.

As described above, even if a negative link is defined, a signal having a positive signal value is given to a node of interest, and only nodes from which a signal having a positive signal value is obtained are extracted. Then, no problem occurs in the positive search processing. That is, nodes connected by negative links are not extracted as candidate nodes. Then, if one node is selected from the candidate nodes presented by the positive search processing, learning is performed as described above. For example, in the example shown in FIG. 40, assuming that the candidate node J is adopted, as shown in the flowchart of FIG. 20, among the learning target paths (links BA, BE, BJ),
Learning is performed to increase the signal transmission coefficient of the link BJ to the adopted node J and decrease the signal transmission coefficients of the other links BA and BE, and increase the frequency coefficient of the adopted node J to increase the other candidate nodes. Learning to reduce the frequency coefficients of A and E is performed.

When the number of hops H for signal transmission is set to 2 or more, multiplication considering the sign of the signal transmission coefficient may be performed as in the above-described example. FIG.
A part of the state in which the positive search processing in which the signal transmission is limited to the number of hops H = 2 is performed for the target node B (node B → A
FIG. 3 is a conceptual diagram illustrating only three types of paths, namely, a path of → F, a path of node B → D → K, and a path of node B → G → H (other paths are omitted). By transmitting a signal via the positive link BA, a positive signal is obtained at the node A, and this positive signal is further transmitted to the node F via the positive link AF, so that a positive signal is also obtained at the node F. ing. On the other hand, a negative signal is obtained at the node D by signal transmission via the negative link BD, and this negative signal is further transmitted to the node K via the positive link DK.
A negative signal is also obtained. In contrast, negative link B
A negative signal is obtained at the node G by the signal transmission via G. However, since this negative signal is further transmitted to the node H via the negative link GH, a positive signal is obtained at the node H. I have. As described above, when the hop number H is 2 or more, a positive signal is obtained for a node that has performed signal transmission through the negative link even number of times, and is extracted as a positive related node. Become. After all, in the example shown in FIG. 41, in addition to the nodes A and F, the node H is also extracted as a positive related node. This means that the node H is “the node of interest B
This is because the node H has a negative relation to the node H that has a negative relation to the node H. In other words, the node H Is expected to have relevance to the node B of interest.

Next, consider negative search processing, that is, a case of searching for a node having negative relevance to the node of interest. In this case, a signal having a negative signal value is transmitted from the target node to another node along the link, and a node having a signal having a positive signal value and a node having no signal transmission are negative. What is necessary is just to extract as a related node. FIG. 42 is a conceptual diagram showing a state in which a negative search process in which signal transmission is limited to the number of hops H = 1 is performed for the node B of interest.

Since the positive links BA, BE, and BJ have positive signal transmission coefficients, when a signal having a negative signal value is given to the target node B, the negative signals are applied to the nodes A, E, and J. A signal having a signal value will be transmitted. On the other hand, since the negative links BD and BG have a negative signal transmission coefficient, when a signal having a negative signal value is given to the node of interest B, the positive signal value is applied to both the nodes D and G. Will be transmitted. Other nodes C,
No signal is transmitted to F, H, I, and K under the condition that the number of hops is H = 1. Here, nodes A, E, and J from which signals having positive signal values were obtained and nodes C, F, H, I, and K without signal transmission are extracted as negative related nodes. Is presented as a candidate node, a negative search process has been executed. A set of nodes (candidate nodes obtained by positive search processing) indicated by white node points in FIG. 40 and each node (candidates obtained by negative search processing) indicated by white node points in FIG. Comparing the obtained set of candidate nodes) with each other, it can be seen that both have a complementary set relationship.

Thus, a negative link is defined,
If a signal having a negative signal value is given to the node of interest and a node having a signal having a positive signal value and a node having no signal transmission are extracted, a node obtained by a positive search process can be obtained. It can be seen that a complement (negative association) of the set (positive association) is obtained. This is exactly the same when the hop number H is set to 2 or more.

An important point here is that the negative search processing by such a method directly obtains a negative related set. For example, in the case of the example of FIG. 42, signal transmission is actually performed to nodes D and G extracted as candidate nodes, and predetermined signal values are obtained for nodes D and G, respectively. The signal values of the other candidate nodes C, F, H, I, and K become zero because no signal was transmitted, but in any case, the candidate nodes extracted as a negative related set have signal values of Priority can be defined based on the priority. For example, in FIG. 42, if the absolute value of the signal transmission coefficient of the negative link BD is larger than the absolute value of the signal transmission coefficient of the negative link BG, the signal value obtained at the node D is larger than the signal value obtained at the node G. Eventually, the candidate node has node D in the first rank, node G in the second rank, node C in the third rank, and so on.
Priorities F, H, I, K can be defined. Of course, as described in §4.6, if a frequency coefficient is defined for each node, it is possible to further define a priority order in consideration of the frequency coefficient.

As described above, if the link is defined in consideration of the sign, a positive search process can be performed if the signal value of the signal given to the node of interest is positive, and a negative search is performed if the signal value is negative. Processing can be performed.

<8.3: Learning by Negative Search Processing> FIG.
The system shown in FIGS. 1, 22, and 29 has a learning unit 90, and when the operator performs an act of selecting a new node of interest from among the candidate nodes, the link aggregation in the link storage unit 50 is performed. Is performed as described above. Further, in the system shown in FIG. 22, the adoption frequency of each node is stored in the frequency coefficient storage means 110, and the learning process for this adoption frequency is performed by the operator's adoption action, as described above. is there.

As described above, the procedure of the learning process when the selection is performed based on the positive search process is as shown in the flowchart of FIG. 20, but the selection is performed based on the negative search process. Also in this case, basically, the learning procedure shown in the flowchart of FIG. 20 can be applied in common. For example, let us consider a case in which the node D is selected as a new node of interest from among the candidate nodes in the negative search processing shown in FIG. In this case, as shown in the flowchart of FIG. 20, the learning target path (the path where effective signal transmission for extracting the related node has been performed, that is, the links BD and BG)
Among them, the signal transmission coefficient of the link BD reaching the adopted node D is increased (the absolute value is increased while the sign remains negative),
Learning to reduce the signal transmission coefficient of the other link BG (reducing the absolute value while keeping the sign negative) is performed, the frequency coefficient of the adopted node D is increased, and the nodes on the other learning target paths are increased. Learning to reduce the frequency coefficient of G is performed.

However, there is a difference between positive search processing and negative search processing in nodes extracted as (positive or negative) related nodes. In other words, the positive related nodes extracted by the positive search processing are all nodes (nodes A, E, and J in the example shown in FIG. 40) on the path (learning target path) where effective signal transmission has occurred. On the other hand, the negative related nodes extracted by the negative search process are only the nodes (nodes D and G in the example shown in FIG. 42) on the path (learning target path) where the valid signal has been transmitted. 42, the nodes that did not transmit signals (in the example shown in FIG. 42, nodes C, F,
H, I, K). Therefore, when learning about the frequency coefficient, it is preferable to add nodes that have not transmitted a signal to the learning target. That is, in FIG. 42, when the node D is adopted,
A learning process of increasing the frequency coefficient of the adopted node D and decreasing the frequency coefficients of all the remaining candidate nodes G, C, F, H, I, and K that are not adopted is performed. Of course, even when a node that has not transmitted a signal is selected, a learning process may be performed in which the frequency coefficient of the selected node is increased and the frequency coefficients of all remaining candidate nodes that are not selected are reduced. . For example, in FIG.
When the node I is adopted, the learning that increases the frequency coefficient of the adopted node I and decreases the frequency coefficient of all the remaining candidate nodes C, D, G, F, H, I, and K that are not adopted is performed. Just do it.

In the present embodiment, when a node having no signal transmission is adopted as a result of the negative search processing, an additional learning processing of defining a new negative link is executed. For example, consider the case where node I is adopted in FIG. In the state shown in FIG. 42, there is no direct link between the focused node B and the adopted node I. In this example, the reason why the node I is presented as a candidate node for the node of interest B is that no signal is transmitted to the node I under a predetermined condition of the number of hops H = 1. Between I and
Not because a positive negative link was defined. However, “the node I was selected by the operator as a result of performing the negative search processing on the node B of interest”
This fact indicates that the operator recognizes a negative association between the node of interest B and the adopted node I. As described above, in the state shown in FIG. 42, the priorities as the candidate nodes are higher in the nodes D and G than in the node I. However, if the operator dares to adopt the node I instead of adopting the nodes D and G having the higher priority, a negative search using the node B as the target node is performed again in the future. It is preferable that the priority order is raised and presented.

From such a viewpoint, in the negative search processing, when a node having no signal transmission is adopted,
The learning unit 90 newly defines a negative link indicating a negative association between the focused node and the adopted node, and performs a process of adding the negative link to the link aggregate in the link storage unit 50. I have.

For example, in the state shown in FIG. 42, if a node I to which no signal is transmitted is adopted as a new node of interest, as shown in FIG.
A negative association has been indicated by the operator between and the adopted node I. Therefore, as shown in FIG. 44, a negative link BI is newly defined between the focused node B and the adopted node I. A negative signal transmission coefficient is defined for the negative link BI, and its absolute value may be set in advance, for example, to “70%”.

FIG. 45 is a conceptual diagram showing a state in which the negative search processing using the node B as the target node is performed again after the learning processing shown in FIG. 44 is performed. Compared with the conceptual diagram shown in FIG. 42, in the search processing after learning in FIG. 45, signal transmission via the negative link BI is newly added, and a positive signal value is obtained at the node I. It turns out that it is. In other words, the priority of the candidate node I is improved as compared with the search processing shown in FIG. Here, the operator again selects the candidate node I
Is adopted, learning is performed to increase the absolute value of the signal transfer coefficient of the negative link BI on the path from the target node B to the adopted node I by the learning processing shown in the flowchart of FIG. 20 (FIG. 46). In this example, the learning result is indicated by a symbol “−−”). By such re-learning, the absolute value of the signal transmission coefficient of the negative link BI is improved, and when a negative search process is performed again with the node B as the target node in the future, the priority of the candidate node I is further promoted. And
It is preferentially presented to the operator.

FIGS. 47 and 48 are diagrams showing the above-described learning processing in an example of combination with the above-described AND search. That is, nodes in the area hatched in FIG. 37 are presented as candidate nodes in combination with the above-described AND search, and the node N4 is adopted from among these candidate nodes as shown in FIG. Suppose you did. In this case, since the adopted node N4 has been obtained by the negative search processing using the node N3 as the target node, the operator has recognized that there is a negative association between the target node N3 and the selected node N4. Will be. here,
When the adopted node N4 is a node that has become a candidate as a node that has not transmitted a signal, a negative node is newly defined between the focused node N3 and the adopted node N4, as shown in FIG. Will be.

§9. One of the features of the database system according to the present invention is as follows.
The point is that the link aggregate in the link storage unit 50 is learned by the learning unit 90. However, the learning by the learning means 90 described so far is based on the correction for increasing or decreasing the signal transfer coefficient given to each static link.
Of course, as mentioned in §3, §4.7, §6.3,
Define a dynamic link, promote it to a static link and add it as a new member of the link structure, or, as described in §8.3, add a new negative The link aggregate is also learned by the process of adding the link, but the previous embodiment does not mention the link reconfiguration for organizing or integrating the static link once defined. . Here, a database system having a function of performing an existing static link consolidation process as an incidental learning process will be described.

FIG. 49 is a block diagram of an embodiment in which a link reconstructing means 130 is added to the database system shown in FIG. Link reconfiguration means 1
30 performs link reconfiguration processing such as deletion of an existing link or addition of a new link to the link aggregate in the link storage means 50 based on the correction to the signal transfer coefficient performed by the learning means 90. Has functions. Here, 2
Specific link reconfiguration processing as described above is exemplified below.

<9.1: First Example of Link Reconfiguration Processing>
Now, for example, as shown in the left column of FIG. 50, for the three nodes N1, N2, and N3, the signal transfer coefficient of the negative link L1 between the nodes N1 and N2 is equal to or more than a "predetermined standard" and the node N2- The signal transmission coefficient of the negative link L2 between N3 is greater than or equal to a "predetermined standard", and both links L1, L
It is assumed that the frequency coefficient of the intermediate node N2 sandwiched between 2 is equal to or less than a “predetermined standard”. In this case, the negative links L1, L
It is considered that the intermediate node N2 was frequently involved in the past adopting action of the operator, but the intermediate node N2 itself was rarely adopted. In other words, the intermediate node N2 plays only a role as a mere transit node on a frequently used path from the node N1 to the node N3. In such a case, as shown in the right column of FIG. 50, the negative links L1 and L2 may be deleted and a bypass primary link L3 may be newly defined between the nodes N1 and N3. In this case, the newly defined primary link L3
If the product of the signal transfer coefficient of the negative link L1 and the signal transfer coefficient of the negative link L2 is given as the signal transfer coefficient of The signal value attenuation due to the signal transmission between the nodes N1 and N3 can be maintained in the same state as before the link reconfiguration. Also, since the node N3 is a node having a negative relationship with the “node N2 having a negative relationship with the node N1,” the node N3 is a node having a positive relationship with the node N1. Can be expected, positive link L
3, it is appropriate to directly connect nodes N1-N3.

As a result, for a particular node N2, the frequency coefficient of the node N2 is equal to or less than a "predetermined standard", and the signal transmission coefficient between the node N2 and another first node N1 is "predetermined reference". There is a first negative link L1 that is equal to or more than a "reference", and a second negative link L2 whose signal transfer coefficient between the node N2 and another second node N3 is equal to or more than a "predetermined reference". Exists, the first negative link L1 and the second negative link L2 are deleted, and a new link L3 is added between the first node N1 and the second node N3. What is necessary is just to perform a structure.

As the “predetermined criterion” as a criterion for determining whether or not to perform such link reconfiguration, a unique reference value may be set in advance. Alternatively, instead of setting a unique reference value, the relative magnitudes of the signal transmission coefficient and the frequency coefficient may be compared. For example,
In the case of the example shown in FIG. 50, the signal transmission coefficient of the negative link L1 is at least twice the frequency coefficient of the node N2, and the negative link L
If the signal transmission coefficient of No. 2 is equal to or more than twice the frequency coefficient of the node N2, it is possible to make an agreement such as performing the link reconfiguration as described above. In this case, the "predetermined criterion" for the frequency coefficient of the node N2 is the signal transmission coefficient value of the negative links L1 and L2, and conversely, the "predetermined criterion" for the signal transmission coefficients of the negative links L1 and L2. "Means the frequency coefficient value of the node N2.

<9.2: Second Example of Link Reconfiguration Processing>
Now, for example, as shown in the left column of FIG. 51, for three nodes N1, N2, and N3, the signal transfer coefficient of the negative link L1 between the nodes N1 and N2 is equal to or more than a "predetermined standard", and the node N2- The signal transmission coefficient of the primary link L2 between N3 is equal to or more than a "predetermined standard", and the two links L1, L
It is assumed that the frequency coefficient of the intermediate node N2 sandwiched between 2 is equal to or less than a “predetermined standard”. In this case, it is considered that the negative link L1 and the positive link L2 were frequently involved in the past adoption actions of the operator, but the intermediate node N2 itself was rarely adopted. In other words, the intermediate node N2 plays only a role as a mere transit node on a frequently used path from the node N1 to the node N3. In such a case, as shown in the right column of FIG. 51, the negative link L1 and the positive link L2 may be deleted, and a negative link L3 for bypass may be newly defined between the nodes N1 and N3. in this case,
If the product of the signal transmission coefficient of the negative link L1 and the signal transmission coefficient of the positive link L2 is given as the newly defined signal transmission coefficient of the negative link L3 (the product of the positive coefficient and the negative coefficient). Therefore, naturally, it becomes a negative coefficient), the node N1
The signal value attenuation due to the signal transmission between −N3 can be maintained in the same state as before the link reconfiguration. Further, since the node N3 is a node having a positive relationship with the “node N2 having a negative relationship with the node N1”, the node N3 is a node having a negative relationship with the node N1. It can be expected, and it is appropriate to directly connect the nodes N1-N3 by the negative link L3.

As a result, for a particular node N2, the frequency coefficient of the node N2 is equal to or less than a "predetermined standard", and the signal transmission coefficient between the node N2 and another first node N1 is "predetermined reference". A negative link L1 that is equal to or more than the “reference” exists and a second node N3 different from the node N2
If there is a positive link L2 having a signal transmission coefficient between the first node N1 and the second node L2, the negative link L1 and the positive link L2 are deleted.
A link reconfiguration for adding a new negative link L3 to the node N3 may be performed.

As the "predetermined criterion" as a criterion for determining whether or not to perform such a link reconfiguration, a unique reference value may be set in advance. Alternatively, instead of setting a unique reference value, the relative magnitudes of the signal transmission coefficient and the frequency coefficient may be compared. For example,
In the case of the example shown in FIG. 51, the signal transmission coefficient of the negative link L1 is at least twice the frequency coefficient of the node N2 and the positive link L
If the signal transmission coefficient of No. 2 is equal to or more than twice the frequency coefficient of the node N2, it is possible to make an agreement such as performing the link reconfiguration as described above. In this case, the “predetermined criterion” for the frequency coefficient of the node N2 is the signal transmission coefficient value of the negative link L1 and the positive link L2, and conversely, the signal transmission coefficient value of the negative link L1 and the positive link L2. The “predetermined criterion” is the frequency coefficient value of the node N2.

<9.3: Specific Example of Link Reconfiguration Processing> Finally, an example in which the above-described link reconfiguration processing is specifically performed on the link aggregate shown in FIG. 39 will be described. For example, in FIG. 39, when the absolute value of the signal transfer coefficient of the negative link BG and the absolute value of the signal transfer coefficient of the negative link GH are compared with the frequency coefficient of the node G, the former is twice or more the latter. Let's say it was determined. in this case,
Although the negative link BG and the negative link GH are relatively frequently used for node adoption, the frequency of adoption of the node G is low, so it is considered that the node G plays only a role as a mere passing node. . In such a case, the negative link BG and the negative link GH are deleted, and as shown in FIG.
H may be defined. The signal transfer coefficient of the positive link BH is
What is necessary is just to make the product of the signal transmission coefficient of the deleted negative link BG and the signal transmission coefficient of the deleted negative link GH.

In FIG. 39, when the absolute value of the signal transmission coefficient of the negative link HG and the absolute value of the signal transmission coefficient of the positive link GJ are compared with the frequency coefficient of the node G,
Suppose the former is determined to be more than twice the latter. In this case, although the negative link HG and the positive link GJ are relatively frequently used for adopting nodes, the frequency of adoption of the node G is low, so that the node G plays only a role as a mere passing node. it is conceivable that. In such a case, the negative link HG and the positive link GJ may be deleted, and a negative link HJ may be newly defined between the nodes H and J as shown in FIG. The signal transmission coefficient of the negative link HJ may be the product of the deleted signal transmission coefficient of the negative link HG and the deleted signal transmission coefficient of the positive link GJ.

[0205]

As described above, according to the database system according to the present invention, a node aggregate and a link aggregate are defined, each data is associated with each node, and a positive association is established with respect to a specific node of interest. We have added a function to search for nodes that have
It is possible to perform a flexible search with a high degree of freedom according to the intention of each user. Further, since the learning is performed on the link aggregate, the more the user uses the search, the more the search according to the user's intention becomes possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a basic database system for explaining a basic concept of the present invention.

FIG. 2 is a diagram illustrating an example of a result of a search process in the database system illustrated in FIG. 1;

FIG. 3 is a diagram showing a state in which a new node of interest N4 is adopted based on the search result shown in FIG. 2;

FIG. 4 is a diagram showing a search process in which a dynamic link L8 is temporarily defined in the database system shown in FIG. 1;

5 selects a new node of interest N6 based on the search result shown in FIG.
FIG. 8 is a diagram showing a state in which 8 has been promoted to a static link.

FIG. 6 is a diagram illustrating an evaluation method when a plurality of equivalent keywords are defined in one node in keyword evaluation when defining a dynamic link.

FIG. 7 is a diagram showing an example of a simple distributed database system including three classes A, B, and C.

8 is a diagram showing an example in which class links and thesaurus dictionaries are defined in the distributed database system shown in FIG. 7;

FIG. 9 is a diagram showing a state where static links L1 to L5 are defined in the distributed database system shown in FIG. 8;

FIG. 10 shows an example using the static link shown in FIG.
It is a figure showing the result of having performed the search which set node N1 as an attention node.

11 is a diagram illustrating a result of performing the same search processing as the search illustrated in FIG. 10 in consideration of class link weights (signal transmission coefficients).

12 is a diagram showing a state in which a learning process for correcting the weight (signal transmission coefficient) of the static link is performed based on the search result shown in FIG.

FIG. 13 is a diagram illustrating a result of performing the same search process as the search illustrated in FIG. 10 again in a state after the learning process illustrated in FIG. 12;

14 is a chart showing a method of determining a priority order when presenting a candidate node in consideration of the weight of the node after executing the search processing shown in FIG. 10;

FIG. 15 is a table showing a method of determining a priority order when presenting candidate nodes in consideration of node weights after executing the search processing shown in FIG. 13;

FIG. 16 is a diagram illustrating an example of executing a search process using both a static link and a dynamic link in the distributed database system illustrated in FIG. 8;

FIG. 17 is a diagram illustrating an example in which the definition of a dynamic link is restricted by a class link.

FIG. 18 is a flowchart showing the overall use procedure of the database system according to the present invention.

FIG. 19 is a flowchart showing in detail a procedure of a search process in step S4 in the flowchart shown in FIG. 18;

20 is a flowchart showing in detail a procedure of a learning process in step S7 in the flowchart shown in FIG. 18;

FIG. 21 is a block diagram showing a specific configuration of a database system according to an embodiment of the present invention.

FIG. 22 is a block diagram showing an embodiment in which a frequency coefficient storage means 110 is added to the database system shown in FIG. 21;

FIG. 23 is a conceptual diagram showing a node aggregate to be searched in the database system according to the present invention.

24 is a conceptual diagram showing a state where one node N1 in the node aggregate shown in FIG. 23 is designated as a target node.

FIG. 25 is a view showing the node of interest N1 in the state shown in FIG. 24;
It is a key map showing positive related set G1 searched by positive search processing about.

FIG. 26 is a conceptual diagram showing a state in which one node N2 is adopted from within the positive association set G1 shown in FIG.

FIG. 27 is a conceptual diagram showing a state in which the adopted node N2 shown in FIG. 26 has been updated to a new node of interest.

FIG. 28 is a conceptual diagram showing a positive related set G2 searched by a positive search process for a new node of interest N2 shown in FIG. 27;

FIG. 29 is a block diagram showing an embodiment in which a population set definition means 120 is added to the database system shown in FIG. 21;

30 is a conceptual diagram showing a state where a node of interest N1 is specified in a node aggregate to be searched in the database system shown in FIG. 29;

FIG. 31 is a diagram showing the node of interest N1 in the state shown in FIG. 30;
It is a key map showing positive related set G1 searched by positive search processing about.

FIG. 32 shows a case where a node N2 in the positive related set G1 shown in FIG. 31 is adopted as a new node of interest, and the new node of interest N2 is searched by a positive search process in an “AND search” mode. It is a key map showing positive related set G2.

FIG. 33 is a conceptual diagram showing a set M3 of candidate nodes presented as a positive search result in the “AND search” mode shown in FIG. 32;

34. A node N3 is selected as a new node of interest from the set of candidate nodes shown in FIG. 33, and the new node of interest N3 is searched by a positive search process in the “AND search” mode. FIG. 10 is a conceptual diagram showing a positive related set G3.

FIG. 35 is a conceptual diagram showing a manner in which candidates corresponding to the logical product of the respective search conditions are extracted by three positive search processes in the “AND search” mode.

36 is a view showing the node of interest N1 in the state shown in FIG.
It is a conceptual diagram which shows the negative related set G1 ^* searched by the negative search process about.

FIG. 37 adopts the node N3 as a new node of interest from the set of candidate nodes shown in FIG. 33, and searches for the new node of interest N3 by negative search processing in the “AND search” mode. Negative related set G3 ^*
FIG.

FIG. 38 is a conceptual diagram showing a state in which candidates corresponding to the logical product of the respective search conditions are extracted by three positive and negative search processes in the “AND search” mode.

FIG. 39 is a conceptual diagram showing an example of a node / link aggregate defined using both a positive link and a negative link.

40 is a conceptual diagram showing a state in which, in the node / link aggregate shown in FIG. 39, a node B is designated as a target node, and a positive search process is performed with the hop count H = 1.

FIG. 41 is a conceptual diagram showing a part of a state in which a positive search process in which the node B is designated as the target node and the number of hops is H = 2 is performed in the node / link aggregate shown in FIG. 39;

42 is a conceptual diagram showing a state in which, in the node / link aggregate shown in FIG. 39, a node B is designated as a target node, and a negative search process is performed with the hop count H = 1.

FIG. 43 is a conceptual diagram showing a state in which a node I has been selected from candidate nodes presented by the negative search processing shown in FIG. 42;

FIG. 44 is a conceptual diagram showing a learning process by adopting the candidate node I shown in FIG. 43.

45 is a conceptual diagram showing a state in which the node B is designated as a target node in the node / link aggregate after the learning process shown in FIG. 44 and a negative search process is performed with the hop count H = 1. .

46 is a conceptual diagram showing a learning process performed by adopting a node I from the candidate nodes presented by the negative search process shown in FIG.

FIG. 47 is a conceptual diagram showing a state in which a node N4 has been adopted from candidate nodes presented by the negative search processing shown in FIG. 37;

FIG. 48 is a conceptual diagram showing a learning process in which a new negative link is defined between the target node N3 and the adopted node N4 by adopting the node N4 shown in FIG. 47;

FIG. 49 is a block diagram showing an embodiment in which a link reconfiguration unit is further added to the database system shown in FIG. 21;

50 is a diagram showing a first example of a link reconfiguration process executed in the database system shown in FIG. 49.

FIG. 51 is a diagram showing a second example of the link reconfiguration processing executed in the database system shown in FIG. 49.

52 is a diagram illustrating an example in which the link reconfiguration processing illustrated in FIG. 50 is applied to the node / link aggregate illustrated in FIG. 39.

FIG. 53 is a diagram illustrating an example in which the link reconfiguration processing illustrated in FIG. 51 is applied to the node / link aggregate illustrated in FIG. 39;

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Database 2 ... Node / link aggregate 3 ... Operator 10 ... Data storage means 20 ... Data provision means 30 ... Target node setting means 40 ... Search means 50 ... Link storage means 60 ... Candidate presentation means 70 ... Candidate selection means 80 ... Updating means 90 learning means 100 operator 110 frequency coefficient storing means 120 population definition means 130 link restructuring means A, B, C class AA, BB class link (local link) AB, AC class link (Remote links) A to K nodes G1, G2, G3 ... positive related set (set of nodes related to the node of interest) G1 ^* , G3 ^* ... negative related set (set of nodes not related to the node of interest) H ... number of hops Hmax ... upper limit of number of hops K1-K9 ... keyword K40, K70 ... representative keyword K41- K45, K71 to K75 ... equivalent keywords L1 to L8 ... instance links (static link and dynamic link) M1 to M3 ... population set N1 to N9, N12 ... nodes Taa, Tbb, Tab, Tac ... thesaurus dictionary

Claims (13)

[Claims] 1. A node aggregate composed of a plurality of nodes is defined, predetermined data is associated with each node, and when a specific node is designated, data associated with this node is provided. A database system having a function, comprising: data storage means for storing data to be provided in association with nodes; and link storage means for storing a link aggregate consisting of a set of links indicating relationships between nodes. A target node setting unit that sets a specific target node based on an instruction from an operator; and using the link aggregate, a set of related nodes related to the target node under predetermined conditions. Search means for extracting a complement set as a negative related set, and all or some of the nodes belonging to the negative related set as candidate nodes to the operator. From the candidate nodes presented by the candidate presenting unit, a candidate selecting unit for allowing an operator to adopt a specific candidate node as an adopted node, and associating the selected node with the selected node from the data storage unit. And a data providing means for extracting the obtained data and providing the extracted data to an operator. 2. The database system according to claim 1, further comprising a set defining means for defining a set indicating a specific node set, wherein the candidate presenting means includes a logic of the negative related set and the set. A database system having a function of presenting only nodes belonging to an intersection as candidate nodes. 3. The database system according to claim 1, further comprising learning means for executing a learning process for defining a negative relationship between the target node and the adopted node, wherein the search means comprises a specific node of interest. When extracting a negative related set for, if there is a node for which a negative relevance to the node of interest is defined by past learning processing, extract this node as a member of the negative related set A database system, wherein the candidate presenting means determines a priority when presenting candidate nodes in consideration of the negative relevance. 4. A node aggregate including a plurality of nodes is defined, predetermined data is associated with each node, and when a specific node is designated, data associated with this node is provided. A database system having a function, a data storage means for storing data to be provided in association with a node, a positive link indicating a positive relationship between the nodes, and a negative link indicating a negative relationship between the nodes A link storage unit that stores a link aggregate composed of a set of the following: a target node setting unit that sets a specific target node based on an instruction from an operator; and Positive search processing for searching for a positive relation node having a positive relation under a predetermined condition, and extracting a set of the positive relation nodes as a positive relation set; By using a link aggregate, a negative associated node having a negative association with the noted node is determined under a predetermined condition, and a negative associated set is extracted as a negative associated set. Search process,
A candidate presenting unit that presents all or a part of the nodes belonging to the positive related set or the negative related set to the operator as a candidate node, and a candidate node presented by the candidate presenting unit. From among the candidate selecting means for allowing the operator to adopt a specific candidate node as the selected node, from the data storage means, extracting data associated with the selected node, data providing means for providing this to the operator, A database system comprising: 5. The database system according to claim 4, further comprising updating means for updating the setting of the focused node setting means such that the selected node becomes a new focused node, and the focused node is updated. A database system characterized in that one of a positive search process and a negative search process can be designated for a search means each time. 6. The database system according to claim 5, further comprising a mother set defining means for defining a mother set indicating a specific node aggregate, wherein the candidate presenting means is a logical product of the related set and the mother set. It has a function of presenting only nodes belonging to a set as candidate nodes, and the mother set defining means has a function of redefining the aggregate of the candidate nodes as a new mother set. Database system. 7. The database system according to claim 4, wherein a positive signal transmission coefficient is defined for a positive link, a negative signal transmission coefficient is defined for a negative link, and When performing the search process, a signal having a positive signal value is transmitted from the node of interest to another node along the link, and the value obtained by multiplying the signal value before transmission by the signal transmission coefficient defined in the transmission path is When a signal having a signal value equal to or greater than a predetermined threshold value is defined as a signal value after transmission and a node for which a signal is obtained is extracted as a positive related node, the search unit performs a negative search process. A signal having a negative signal value is transmitted from the node of interest to another node along the link, and a value obtained by multiplying the signal value before transmission by the signal transmission coefficient defined in the transmission path is referred to as a signal value after transmission. Define and signal value above a certain threshold A database system for performing processing for extracting a node from which a signal having a symbol is obtained as a negative related node. 8. The database system according to claim 7, wherein a predetermined threshold value for performing a positive search process is set to a positive value, and a predetermined threshold value for performing a negative search process is set. A database system characterized by being set to zero. 9. The database system according to claim 7, wherein learning means for correcting a link signal transmission coefficient or adding a new link to the link aggregate based on the adopted node adopted by the operator. A database system further provided. 10. The database system according to claim 9, wherein the learning unit defines all paths from the node of interest to the positive related node or the negative related node as learning target paths, and extends from the node of interest to the adopted node. When a path exists, the absolute value of the signal transmission coefficient for the link on this path is modified to increase relative to the signal transmission coefficient of the other link on the learning target path. Database system. 11. The database system according to claim 9, wherein when a node having no signal transmission is selected by the negative search processing, the learning unit newly adds a negative link between the focused node and the selected node. A database system comprising defining and adding this negative link to a link aggregate in a link storage means. 12. The database system according to claim 9, further comprising frequency coefficient storage means for storing a frequency coefficient indicating an adoption frequency for each node, wherein the learning means stores a frequency coefficient for the selected node. ,
A function of making a correction to increase relative to the frequency coefficient of the other node, wherein the candidate presenting means includes a signal value of a signal transmitted to each candidate node, a frequency coefficient of each candidate node, A database system having a function of defining a priority order for each candidate node according to both of them, and presenting a candidate node based on the priority order. 13. The database system according to claim 12, wherein, for a particular node, a frequency coefficient of the particular node is equal to or less than a predetermined criterion, and a first node is located between the node and another first node. A link is defined, a second link is defined between the node and another second node, and the absolute values of the signal transmission coefficients of the first link and the second link are predetermined. And when the first link and the second link are deleted, and between the first node and the second node, the sign of the first link and the second A database system further comprising link reconfiguration means having a function of performing link reconfiguration for adding a new link having a code corresponding to the product of the code of the second link and the link.