JPH10198682A - Database retrieving device and database registering device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to calculate the degree of association even when an undefined component is included in a feature vector in the case of calculating the degree of association between an inquiry and previously stored data by using a feature vector by a database retrieving device. SOLUTION: When an undefined component exists in an inquiry vector generated by an inquiry vector generation means 2 or the feature vector of each data stored in a data storage means 3 prior to the calculation of a distance between the inquiry vector and the feature vector of each data, an undefined component interpolating means 4 interpolates the undefined component by using a correlation table indicating the degree of correlation among the components of a feature vector stored in an inter-component correlation table storing means 5 and other defined components in the feature vector including the undefined component. When another correlated defined component does not exist, the undefined component can not be interpolated, so that a specific code for setting up a difference from any value to '0' is substituted for the undefined component so as not exert influence upon the distance calculation of the feature vector.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a database search device and a database registration device that enable efficient and accurate related data search for irregular data.

[0002]

2. Description of the Related Art Conventionally, a database that enables a search using the semantic information of data has been automatically input from a generally stored data, such as a database of technical documents and a database of library collections. Was extracted, and the degree of association between data was calculated using a thesaurus created empirically in advance.

On the other hand, in the case of data that does not depend on languages, such as images and sounds, a method is known in which the feature amount of the data is represented in the form of a vector, and the degree of association between the data is represented by the distance between the two vectors. I have. Such methods include, for example, JP-A-58-147799 and JP-A-60-1895.
No. 85, JP-A-64-17182, JP-A-Hei.
As disclosed in Japanese Patent Laid-Open No. -15777, there is a method of determining the similarity based on the Euclidean distance between feature vectors of a voice pattern or a character pattern. In addition, there is a method of determining the similarity by summing the absolute values of the differences between the components of the feature vector as disclosed in Japanese Patent Application Laid-Open Nos. 60-202491 and 63-780. In any case, the feature vector is a sequence of physical quantities characterizing audio data or image data, and it is premised that all components of the feature vector are determined in advance.

[0004] In recent years, hypertexts have attracted attention as a data format for uniformly handling documents, images, sounds, and the like. The hypertext has an information unit called a node composed of a document, an image, a sound, and the like, and the nodes can be associated with each other by a link. Also in the field of databases, by targeting the data expressed in the hypertext, it is applied as a new medium with no distinction of data types such as documents, images, and sounds.

Here, there is a demand for a technique for automatically adding a link from a new node to a related node or a link from a related node to a new node when a new node is added to the hypertext system. In such a method, as disclosed in Japanese Patent Application Laid-Open No. 3-278247, when the content of a node is changed, the difference information or all information of the node is held as a new version of the node, and the new node is added to the node before the change. Is added from the link. Also, as disclosed in JP-A-4-317172, in addition to a node having ordinary information (search target node), a node (search node) used for search is prepared and a corresponding search target node is linked. In addition, there is a method in which when there is a change in the search target node, the link of the search node that needs to be changed is updated or added.
Finally, qualitative search conditions are described in advance in each information as in Japanese Patent Application Laid-Open No. Hei 5-20363, and the search conditions are displayed at the same time when the information is displayed. There is also a technique in which information is ambiguously searched. Here, the qualitative search condition is a condition that can be determined by a physical quantity such as the contrast of an image, such as the darkness of an image. Conventionally, in a hypertext database, a list of links to nodes that match a search formula is generally displayed for displaying search results.

[0006]

The first problem is as follows.
For data search, as a means to calculate the degree of association between the query entered by the user and the data already stored,
The conventional method in which the features of the query and data are represented by vector quantities and the degree of relevance of them is the distance between the vectors is based on the assumption that all the components of the feature vector have been determined. If there is a definition,
There was a problem that distance calculation could not be performed.

[0007] As a second problem, regarding a search for a hypertext database, in the conventional method, the search result is simply a list of nodes related to the query input by the user, and the link information about each node included in the search result. Therefore, there is a problem that it is difficult for the user to predict whether or not each node represented as a search result is what he or she wants.

[0008] As a third problem, with respect to data registration, in the prior art in which a keyword is automatically extracted from data or a keyword is freely registered by a registrant in order to define the meaning of the data, for example, a user interface is involved. Like data, user characteristics, task characteristics,
If it is necessary to characterize from various viewpoints, such as system characteristics, and the appropriate vocabulary is not clear as a keyword, it is difficult to properly define a keyword that represents the meaning of the data, and therefore, appropriate characterization for the data There was a problem that it was difficult.

As a fourth problem, in the conventional method of automatically generating a link using qualitative information added to each node when registering a new node in a hypertext database, (1) ) Problems such as not being able to cope with atypical data that allows arbitrary combinations of documents, images, sounds, etc., and (2) not being able to dynamically determine the link direction for node pairs linked to each other was there.

Therefore, according to the present invention, atypical data obtained by freely combining documents, images, sounds, and the like can be characterized by a single method, and newly registered data and already registered data can be characterized. It is an object of the present invention to provide a database registration device for automatically linking with related data, and a database search device capable of performing highly accurate and efficient related data search and having a good prospect of search results. I do.

[0011]

According to the present invention, there is provided a database search apparatus comprising: a query input unit for inputting a query from a user; a data storage unit for storing data and a feature vector of the data; Query vector generating means for converting the query input in step 2 into a query vector, inter-component correlation table storing means for storing a table of the degree of correlation between the components of the feature vector, and a query generated by the query vector generating means The vector and the undefined components of the feature vector added to each data stored in the data storage unit are used to determine the correlation between the components defined by these vectors and the components stored in the inter-component correlation table storage unit. Undefined component complementing means for complementing using a table of degrees, and the query vector in which an undefined part has been complemented by the undefined component complementing means An inter-vector distance calculating means for obtaining a distance from a feature vector of the data stored in the data storing means, and a distance between the feature vectors calculated by the inter-vector distance calculating means being equal to or less than a predetermined threshold value If it is, nearby data determination means for determining the data as proximity data to the inquiry, and threshold storage means for storing a predetermined threshold value used for determination of proximity data in the proximity data determination means Search result display means for displaying a list of data determined to be near data by the nearby data determination means; and data for instructing data to be displayed by the user from a list of data displayed by the search result display means. It is characterized by comprising an instruction means and a display means for displaying the contents of the data specified by the data instruction means.

In the database search apparatus having such a configuration, the query vector and the feature vector of the already stored data are complemented by the undefined component complementing means. Even if the vector includes undefined components, their distance can be calculated to measure the degree of relevance.

The undefined component complementing means is defined to have a correlation with the undefined component which is necessary for obtaining a complement value of a certain undefined component of a query vector and a feature vector of data stored in advance. If there is no component, the undefined component is replaced with a special symbol that always becomes 0 regardless of the difference from any value. Thus, even when the undefined component cannot be complemented as described above, the distance between the vectors can be obtained to measure the degree of association.

Further, in the database search device according to the present invention, when a relation between data is determined in advance and the data is classified into several categories, each of the neighborhood data obtained by the neighborhood data determining means is provided. And the category to which the data that can be reached by tracing the relation from the neighboring data belongs to, using a predetermined inter-category rule stored by the inter-category rule storing means, to trace the relation from the neighboring data. And a relational data judging unit for judging whether or not data reachable by the search is included in the search result.

On the other hand, the database registration device of the present invention
Content registration means for the data registrant to enter the content of the new data; and keyword selection for the data registrant to select a keyword suitable for representing the characteristics of the new data from a predetermined set of keywords. Means, a keyword set storage means for storing a keyword set used by the keyword selection means, and [0,
1] weighting input means for the data registrant to input weighting by real numbers, and a feature vector of new data based on the keyword selected by the keyword selection means and the weighting input by the weighting input means And a data storage means for storing the new data and a feature vector of the new data.

In the database registration device having such a configuration, it is possible to assign an appropriate keyword to the atypical data by an arbitrary combination of text, image, sound, and the like by a single method.

The database registration device of the present invention
Further, an inter-component correlation table storage unit for storing a table of the degree of correlation between components of the feature vector, and a feature vector of the new data generated by the feature vector generation unit and each of the feature vectors stored in the data storage unit. An undefined component that complements the undefined components of the feature vector added to the data using the components defined by these vectors and the table of the degree of correlation between the components stored in the inter-component correlation table storage unit. Interpolating means; inter-vector distance calculating means for calculating a distance between a feature vector of the new data whose undefined part has been complemented by the undefined component complementing means and a feature vector of the data stored in the data storage means; If the distance between the feature vectors calculated by the inter-vector distance calculation means is equal to or less than a predetermined threshold, the data is stored in the neighborhood data for the new data. Data, a threshold value storage unit that stores a predetermined threshold value used for the determination of the proximity data in the proximity data determination unit, and a threshold value storage unit that determines the proximity data. Link direction determining means for determining a direction of a link to be newly added between the neighboring data and the new data from a category of the neighboring data and a category of the new data by using a predetermined linking rule. A node adding means for adding a link between the neighboring data and the new data in the direction determined by the link direction determining means, and storing information about the new data and the link in the data storage means; It is characterized by having.

This makes it possible to automatically link to another node that is semantically related to the creation of a new hypertext node.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described in detail with reference to the drawings.

First, a first embodiment of a database search device according to the present invention will be described in detail with reference to the drawings.

Referring to FIG. 1, the configuration of the first embodiment of the database search apparatus according to the present invention is a query input means 1 for allowing a user to input desired data characteristics.
Query vector generating means 2 for converting an input query into a feature vector, data storage means 3 for storing data and a feature vector of data, and an undefined component for complementing an undefined component included in the feature vector Complementary means 4
An inter-component correlation table storage unit 5 storing an inter-component correlation table used by the undefined component complementing unit 4, an inter-vector distance calculation unit 6 for obtaining a distance between a query and a feature vector of stored data, A neighborhood data determination unit 7 for determining stored data having a feature vector located at a distance equal to or less than a certain threshold with respect to the feature vector; a threshold storage unit 8 for storing the threshold; Judgment means 7
Search result display means 9 for displaying a list of nearby data obtained in step 1, data indicating means 10 for a user to specify arbitrary data from the list of nearby data displayed on search result display means 9, Data display means 11 for displaying the contents of the data thus obtained. Here, the position group extraction unit 12 is constituted by the undefined component complementing unit 4, the inter-component correlation table storage unit 5, the inter-vector distance calculation unit 6, the proximity data determination unit 7, and the threshold value storage unit 8.

Inquiry input means 1 and data instruction means 1
Reference numeral 0 denotes an information input device such as a mouse and a keyboard. The search result display unit 9 and the data display unit 11 include an information display device such as a VDT. The data storage unit 3 is configured by an external storage device such as a fixed disk, and the inquiry vector generation unit 2 and the position group extraction unit 12 are configured by a program operating on a PC (personal computer), WS (workstation), or the like. It consists of.

The data stored in the data storage means 3 is
It is composed of data that has any combination of documents, images, sounds, etc. and has zero or more relations to other data, and feature vectors corresponding to the data. For example, data A
And data B, and each feature vector is v
[A], v [B], and one of the relations of the data A has a relation to the data B, the data storage means 3
Then, it is saved in the form as shown in FIG.

Next, the operation of the database search device configured as shown in FIG. 1 will be described with reference to the drawings.

Referring to FIG. 1, it is assumed that a keyword is selected from a user using, for example, a keyboard or a mouse, and weighted by a real number [0, 1] in the inquiry input means 1. For example, FIG.
As shown in FIG. 3A, it is assumed that a keyword set representing the characteristics of data such as a design example of a user interface of an arbitrary system is determined in advance, and the user selects the keyword shown in FIG. From the top, respectively, "unselected, 0.8, 0.2, unselected, 0.8,
0.7, unselected, 0.1, 0.2 ". At this time, the feature vector (query vector) corresponding to the input query is generated as follows by the query vector generating means 2. (X [1], 0.8, 0.2, x [2], 0.8, 0.7, x [3], 0.1, 0.2) where x [1], x [2] and x [3] indicate undefined components.

Next, when the query vector and the feature vector of each data in the data storage means 3 to be searched include an undefined component, the feature vector of the data is determined by the undefined component complementing means 4. And the correlation table between the attribute values stored in advance by the inter-component correlation table storage unit 5, the complement value of the undefined component is determined.

FIG. 4 shows an example of a correlation table between attribute values stored in the inter-component correlation table storage means 5. In this table, the attribute values are listed in the row direction and the column direction, and the presence or absence of the correlation and the degree of correlation are described at the intersection between the row and the column. When examining another attribute value having a correlation with a certain attribute value α, the attribute value α is selected from the attribute value group listed above, and the value described at the intersection between the column and each row is examined. If the numerical value p is described, there is a correlation with the attribute value described on the left side of the line, and the degree of correlation is p, and if "-", there is no correlation. For example, when examining another attribute value having a correlation with the attribute value “for beginners”, the intersection value between the column of the attribute value “for beginners” and each row in the attribute value group listed at the top of the table is examined. In the case of the figure, the attribute value "for infants",
There is a correlation with "for the general public" and "for the elderly", and the degrees of correlation are 0.9, 0.3 and 0.8, respectively.

Now, for example, in the case of the above inquiry vector, the undefined component complementing means 4 calculates x [1], x [2] and x
Since the three undefined components of [3] are included, the following processing is performed.

Referring to the correlation table of FIG. 4, there are three types of attributes, "products for infants", "products for the general public", and "products for the elderly" that have a correlation with the attribute value "for beginners" corresponding to x [2]. In the query vector, the attribute value for “toddler” is excluded because it is an undefined component x [1], and the “correspondence” with the highest correlation among the remaining “for general” and “for the elderly” For the elderly "is selected. The meaning of this correlation value indicates that "the elderly are novice with considerable certainty." Looking at the correlation table, the correlation degree is 0.8
Since the weight assigned to “for the elderly” in the inquiry vector is 0.2, it is complemented as x [2] = 0.2 · 0.8 = 0.16.

Similarly, “visual impairment correspondence” corresponding to x [3] has a correlation of 0.7 only with “for the elderly” in the correlation table of FIG. Since the weight for the value “for the elderly” is 0.2, it is complemented as x [3] = 0.2.7 = 0.14.

On the other hand, the attribute value "for infants" corresponding to x [1] has a correlation only with the attribute value "for beginners" in the correlation table of FIG. It is an undefined component x [2]. Therefore, a complementary value cannot be obtained. In such a case, ⊥ is assigned. ⊥ indicates that the component is not used in the distance calculation when calculating the inter-vector distance in the subsequent stage. That is, when at least one of the two vector components whose difference is to be obtained is 求 め, the difference is always set to 0. By doing so, the undefined component that could not be complemented does not affect the distance calculation at the subsequent stage, so that the distance between the vectors can still be used as an index of the degree of semantic relevance between the two data. As a result of the above complementation, the inquiry vector is as follows. (⊥, 0.8, 0.2, 0.16, 0.8, 0.7, 0.14, 0.1, 0.2) ...

In the correlation table of FIG. 4, each attribute value has a correlation with at least one other attribute value, but there is an attribute value having no correlation with another attribute value. If such an attribute value appears as an undefined component, a complementary value cannot be obtained in the same manner as described above. In such a case, ⊥ is assigned.

Thereafter, the inter-vector distance calculating means 6 calculates a distance between the query vector in which the undefined component is complemented as described above and the feature vector of each data. There are two possible definitions of this distance: (1) the sum of the absolute values of the differences for each component, or (2) the Euclidean distance. An index of the degree of semantic relevance between the query and the data to be searched Both have the same effect. Therefore, here, the distance is defined as the sum of the absolute values of the differences between the components. Then, the neighborhood data determination means 7 compares the threshold value registered in advance in the threshold value storage means 8 with the distance between the vectors. Search target data is added.

The data set in the position group Gp thus obtained is a search result in response to an inquiry from the user, and a list of data contained therein and a relation between the data is displayed on the search result display device 14 such as a display device. . In the case of hypertext, the list includes the title of each data. When the user looks at such a list and specifies any data in the list using the data specifying means 10, the data display device 11 displays the detailed contents of the specified data.

As described above, it is possible to perform an intuitive input without using a search formula or the like in a user's query input, and it is possible to perform an efficient and highly accurate relevance calculation using a feature vector. Even if there are undefined components, the distance between feature vectors can be calculated without losing the meaning of the degree of semantic relevance between data.

Next, a second embodiment of the database search device according to the present invention will be described with reference to the drawings.

Referring to FIG. 5, in the configuration of the second embodiment of the database search apparatus according to the present invention, the relationship between data is determined in advance, and the data is classified into several categories. In this case, the predetermined category stored in the inter-category rule storage unit 14 is based on the category to which each piece of neighboring data obtained by the position group extracting unit 12 belongs and the category to which data that can be reached by following the relationship from the neighboring data belongs. The first embodiment shown in FIG. 1 includes a relational data determination unit 13 that determines whether or not data that can be reached by tracing a relationship from the neighboring data is included in the search result using the inter-category rule. Is different from the database search device of the embodiment. Note that the relational data determination means 13
The inter-category rule storage means 14 constitutes a related group extracting unit 15. The related group extracting unit 15 is realized by a program operating on, for example, a PC or WS.

The data stored in the data storage unit 3 includes data obtained by arbitrarily combining documents, images, sounds, and the like, and having zero or more relations to other data, a feature vector corresponding to the data, and the like. And the category to which the data belongs. For example, there are data A and data B, the respective feature vectors are v [A] and v [B], the respective categories are C1 and C2, and the data A
When one of the relations has a relation to the data B, the data is stored in the data storage means 3 in the form as shown in FIG.

Next, the operation of the database search device shown in FIG. 5 will be described in detail with reference to the drawings, focusing on the differences from the database search device shown in FIG.

Referring to FIG. 5, relational data determining means 1
3 is the position group G obtained by the neighborhood data determination means 7
For any data A belonging to p, a relationship traceable from the data is traced, and any data B beyond that is obtained. Then, using the category to which the data A and the data B belong and the predetermined inter-category rule stored in the inter-category rule storage unit 14, it is determined whether or not the data B is to be included in the data set Gr. For example, three categories, "guideline", "design case", and "evaluation case", are predetermined.
It is assumed that Gp includes data A belonging to “design case”. Data A has a relationship to data B, data C, and data D, and the category of each data is “guideline”, data C is “design case”, and data D is “evaluation case”. And It is also assumed that an inter-category rule as shown in FIG. At this time, referring to FIG. 7, the relational data determination means 13 first sets Gr to G
r = Gp is initialized and data A belonging to Gp is referred to the inter-category rule. In the cross-category rules,
Since it is defined that the relationship from “design case” to “guideline” and the relationship from “design case” to “evaluation case” are both included in Gr, data B and data D are added to Gr. On the other hand, since the relationship from “design case” to “design case” is not included in Gr, data C is not added to Gr. Therefore, for data A, data B and data D are newly added to Gr. Such processing is performed for all data belonging to Gp.

The data set Gr thus obtained is a search result in response to an inquiry from the user, and a list of data included therein and a relation between the data is displayed by the search result display means 9. Then, when the user designates arbitrary data using the data designating means 10, the content of the designated data is displayed on the data display means 11.

Thus, when determining whether each of the data included in the search result is the data desired by the user, not only the degree of relevance to the query but also what data is obtained therefrom It is possible to make a multifaceted determination by using information on whether or not it is possible.

Next, an embodiment of the database registration apparatus according to the present invention will be described with reference to the drawings.

Referring to FIG. 9, the configuration of the embodiment of the database registration apparatus according to the present invention is a content registration means 1 in which a data registrant registers the contents of data (node).
6, a keyword set storage unit 17 storing a predetermined keyword set, a keyword selection unit 18 for the data registrant to select a keyword appropriately representing the content of the node from the keyword set, and a data registrant A weighting input means 19 for weighting each selected keyword using a real number [0, 1]; a feature vector generating means 20 for generating a feature vector from a set of weighted selected keywords; (Node) data storage means 21, a position group extraction unit 12 for obtaining a set of related nodes, and a link direction for determining a link direction when adding a link between a registered node and a neighboring node Determining means 22, linking rule storing means 23 storing rules used for the determination, and data storage of the registered nodes In addition to the stage 21 and a node addition section 24 for adding a link between Node and the neighboring node in the direction determined by the link direction determination means 22.

Here, the data stored in the data storage means 21 is the same as that shown in FIG. 6 described above, and has no relation to other data obtained by arbitrarily combining documents, images, sounds, etc. It is composed of data having the above, a feature vector corresponding to the data, and a category to which the data belongs.

The internal structure of the position group extracting unit 12 is the same as that of the position group extracting unit 12 constituting the main part of the database search apparatus shown in FIG.
As shown in (1), the inter-component correlation table storage means 5 for storing a table of the degree of correlation between the components of the feature vector, and the feature vector of the new node generated by the feature vector generation means 20 and the data storage means 21 The undefined components of the feature vector added to each of the nodes are complemented using the components defined by these vectors and the table of the degree of correlation between the components stored in the inter-component correlation table storage unit 5. Undefined component complementing means 4 and a distance between vectors for calculating a distance between a feature vector of a new node whose undefined part has been complemented by the undefined component complementing means 4 and a feature vector of a node stored in the data storage means 21 If the distance between the feature vectors calculated by the calculating means 6 and the inter-vector distance calculating means 6 is equal to or less than a predetermined threshold value, the node is set to a neighborhood data for the new node. Data and determines proximity data determination unit 7
And a threshold value storage means 8 for storing a predetermined threshold value used for the determination of the proximity data in the proximity data determination means 7. In addition, the undefined component complementing means 4 correlates with the undefined component necessary for finding the complement value of the feature vector generated by the feature vector generating means 20 and the undefined component of the feature vector of the data stored in advance. If there is no defined component with, the undefined component is replaced with a special symbol that always becomes 0 regardless of the difference from any value.

The content registration means 16, the keyword selection means 18, and the weight input means 19 are information input devices such as a mouse and a keyboard, and include a keyword set storage means 17, a feature vector generation means 20, and a position group extraction part 12. The link direction determining means 22, the linking rule storing means 23, and the node adding means 24 are realized by, for example, a program operating on a PC or WS. The data storage means 21 is an external storage device such as a fixed disk.

The operation of the database registration device having such a configuration will be described with reference to the drawings.

Referring to FIG. 9, the registrant first prepares the contents of the registration node by combining documents, images, and sounds arbitrarily in advance, and inputs the contents by the content registration means 16. The category of the registered node (a guideline, a design case, an evaluation case, etc., which will be described later) is input by the content registration unit 16. And a keyword selecting means 1
In step 8, an appropriate keyword is selected from the keyword set stored in the keyword set storage unit 17 in advance according to the content of the node input by the content registration unit 16. Therefore, some keywords are not selected. Here, the keyword set may be classified into several categories so as to be easily understood by the user. For example, referring to FIG. 11, a keyword set (attribute value) such as “general”, “beginner”, and “visual” is
It is classified into four categories (attributes): "age group", "system experience", "task knowledge", and "disability". Then, the registrant assigns a real number [0, 1] to the selected keyword by the weighting input unit 19 as the importance of the keyword. At this time, a real number may be assigned to the category. In this case, the real number assigned to the category is assigned to all the keywords belonging to the category. By doing so, the keywords are selected all at once, and the user's trouble in inputting weights is reduced. Also, a real number may be assigned to each keyword, and a real number may be assigned to a category. In this case, for all keywords belonging to the category given the real number, the product of the real number assigned to each keyword and the real number given to the category is obtained, and the value is reassigned to the keyword. By doing so, it is possible to reduce the time and effort required to adjust the weight when performing a repeated search.

The feature vector generating means 20 generates a feature vector corresponding to the registered node from the results up to the previous stage. At this time, a special symbol x indicating undefined is assigned to the keyword not selected by the keyword selecting means 18. As such x, for example, -1
For example, a real number other than [0, 1] may be used, or a character such as “x” may be used. A vector in which the real numbers assigned to the respective keywords as described above are arranged in a certain order is a feature vector corresponding to the registered node. For example, for the keyword set mentioned above as an example, "general", "beginner", "expert", "poor", "normal", "rich",
When the eight keywords “visual” and “limb” are selected by the keyword selecting means 18 and subsequently the importance is determined by the user as shown in FIG.
The feature vectors corresponding to the registered data are (x, 0.8, x, 0.9, x, 0.7, 0.3, 0.8, 0.7,
0.7, x, 0.1) (x is an undefined part).

Next, the distance between the feature vector of the registered node determined by the feature vector generation unit 20 and the feature vector of each node stored in the data storage unit 22 is calculated by the position group extraction unit 12, A set Gp of nodes located at a distance equal to or less than a predetermined threshold is obtained.
That is, referring to FIG. 10, the undefined component complementing means 4
The undefined components of the feature vector of the new node generated by the feature vector generating means 20 and the feature vector added to each node stored in the data storage means 21 are defined by the components defined by these vectors and the components Interpolation is performed using the table of the degree of correlation between components stored in the correlation table storage unit 5, and the inter-vector distance calculation unit 6 calculates the feature vector of the new node whose undefined part has been completed and the data storage unit 2.
1. The distance from the feature vector of the node stored in 1 is obtained, and the neighborhood data determination unit 7 determines that the distance between the feature vectors calculated by the inter-vector distance calculation unit 6 is equal to or less than a predetermined threshold. The node is determined as neighboring data for the new node, and is included in the set Gp.

Next, the link direction determining means 22 calculates, for each node of Gp obtained by the position group extracting unit 12,
Determine the link direction to the registered node. At that time, a linking rule stored in the linking rule storage unit 23 in advance is applied. The linking rules are defined based on the category of the node, for example, "Guidelines",
The linking rules shown in FIG. 13 are applied to the three categories of “design case” and “evaluation case”. Now, assuming that the registered node belongs to the "guideline" and the nodes A, B, and C belong to the "guideline", "design case", and "evaluation case", respectively. The link direction between each of A, B, and C is determined as shown in FIG.

Finally, the node adding means 24 adds the registered node, its feature vector, category, and a link to (from) each node belonging to the position group Gp in the direction determined by the link direction determining means 22. Then, the data is accumulated in the node accumulation means 22.

As described above, when registering new data, the link direction to be added between the new data and the neighboring data is determined by the category information to which the new data and the neighboring data belong and the linking rule using the category. Since a link is automatically added in that direction to accumulate new data and its link information, it is possible to reduce the time required for searching for a related node and adding a link when creating a new node of hypertext.

[0055]

As described above, according to the present invention, the following effects can be obtained.

According to the first to third aspects of the present invention, an undefined component included in a part of a feature vector of an inquiry or stored data is replaced with a complement value from a defined component based on the degree of correlation between components. Is determined, it is possible to perform an efficient and highly accurate relevance calculation using the feature vector even if all components of the feature vector of the query and the stored data are not defined.

According to the second aspect of the present invention, when there is no defined component necessary for obtaining a complement value of a certain undefined portion of the feature vector, the difference with respect to an arbitrary value becomes 0. Since the undefined portion is replaced with such a special symbol, the portion where the undefined portion exists is ignored by the inter-vector distance calculating means in the subsequent stage. Therefore,
Even when there are many undefined portions in the feature vector, efficient relevance calculation using the feature vector can be performed.

According to the third aspect of the present invention, not only is a data set having a high degree of relevance to a query obtained,
Since the data that is related to each data belonging to the set is also included in the search results and displayed, the user can make multi-faceted judgments from the additional information other than the relevance for each data included in the search results. This makes it easier to see whether the result is what you want.

According to the fourth aspect of the present invention,
At the time of data registration, a keyword is arbitrarily selected by the user from a predetermined keyword set for the input data, and a feature vector of the input data is generated from a result of weighting each selected keyword. Therefore, it is possible to characterize atypical data formed by an arbitrary combination of documents, images, sounds, and the like in a single method that is easy for the data registrant to understand, and to reduce the time and effort required to assign appropriate keywords. .

According to the fifth aspect of the present invention, at the time of registration of new data, the new data and the neighboring data are linked to each other by the category information and the linking rule using the category. To determine the direction of the link to be added, and automatically add a link in that direction to accumulate new data and its link information, reducing the time required to search for related nodes and add links when creating new nodes for hypertext. can do.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of a database search device according to the present invention.

FIG. 2 is an explanatory diagram of data stored in a data storage unit in FIG.

FIG. 3 is a diagram showing an example of a keyword set presented to a user when an inquiry is input and an example of a keyword actually selected by the user.

FIG. 4 is a diagram showing an example of an inter-component correlation table.

FIG. 5 is a block diagram of a second embodiment of the database search device according to the present invention.

6 is an explanatory diagram of data stored in a data storage unit in FIG.

FIG. 7 is an explanatory diagram of an operation of a related data determination unit.

FIG. 8 is a diagram illustrating an example of an inter-category rule.

FIG. 9 is a block diagram of an embodiment of a database registration device according to the present invention.

FIG. 10 is a block diagram illustrating a configuration example of a position group extraction unit in the database registration device.

FIG. 11 is a diagram illustrating an example of a keyword set (attribute value) and an example of a category stored in a keyboard set storage unit.

FIG. 12 is a diagram illustrating an example of a keyword (attribute value) selected for a new node by a data registrant at the time of registration of the node and its importance (weight).

FIG. 13 is a diagram illustrating an example of a linking rule.

FIG. 14 is a diagram illustrating an example of a link provided to a registered new node.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Query input means 2 Query vector generation means 3 Data storage means 4 Undefined component complementing means 5 Inter-component correlation table storage means 6 Vector distance calculation means 7 Neighbor data judgment means 8 Threshold value storage means 9 Search result display means 10 Data Instruction means 11 Data display means 12 Position group extraction unit 13 Relational data determination means 14 Inter-category rule storage means 15 Relation group extraction unit 16 Content registration means 17 Keyword set storage means 18 Keyword selection means 19 Weighting input means 20 Feature vector generation means 21 data storage means 22 link direction determination means 23 linking rule storage means 24 node addition means

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FIG06F 15/403 380E 15/419 320

Claims (6)

[Claims] A query input unit for inputting a query from a user; a data storage unit for storing data and a feature vector of the data; and a query for converting the query input by the query input unit into a query vector. A vector generation unit, an inter-component correlation table storage unit for storing a table of the degree of correlation between components of the feature vector, and an inquiry vector generated by the inquiry vector generation unit and each of the data stored in the data storage unit. An undefined component that complements the undefined components of the feature vector added to the data using the components defined by these vectors and the table of the degree of correlation between the components stored in the inter-component correlation table storage unit. Complementing means, the query vector whose undefined part has been complemented by the undefined component complementing means, and the data stored in the data storing means. An inter-vector distance calculating means for obtaining a distance from the feature vector, and if the distance between the feature vectors calculated by the inter-vector distance calculating means is equal to or less than a predetermined threshold value, the data is neighborhood data for the inquiry. Proximity data determination means, threshold value storage means for storing a predetermined threshold value used for determination of proximity data in the proximity data determination means, and proximity data determination means in the proximity data determination means Search result display means for displaying a list of searched data, data instructing means for instructing data to be displayed by the user from the data list displayed in the search result display means, and data instructing by the data instructing means. And a display means for displaying the contents of the data. 2. The undefined component complementing means includes a defined component having a correlation with the undefined component necessary for obtaining a complement value of the undefined component of the query vector and a feature vector of data stored in advance. 2. The database search device according to claim 1, wherein when there is no, the undefined component is replaced with a special symbol that always becomes 0 regardless of the difference from any value. 3. A relation between data is determined in advance,
And if the data is broken down into several categories,
The predetermined category stored by the inter-category rule storage unit is determined from the category to which each of the nearby data obtained by the nearby data determination unit belongs and the category to which the data that can be reached by following the relationship from the nearby data belongs. The database search device according to claim 1 or 2, further comprising a relational data determination unit configured to determine whether to include, in a search result, data that can be reached by following a relation from the neighboring data using a rule. 4. A content registration means for a data registrant to input the content of new data, and a data registrant selects a keyword suitable for representing characteristics of the new data from a predetermined keyword set. A keyword set for storing the keyword set used by the keyword selecting means, and weighting each keyword selected by the keyword selecting means with a real number [0, 1]. And a feature vector generating means for generating a feature vector of new data from the keyword selected by the keyword selecting means and the weight inputted by the weight input means. And data storage means for storing the new data and a feature vector of the new data. Database registration apparatus characterized by. 5. An inter-component correlation table storage unit for storing a table of the degree of correlation between components of a feature vector, a feature vector of new data generated by the feature vector generation unit, and a feature vector stored in the data storage unit. The undefined components of the feature vector added to each data are complemented by using the components defined by these vectors and the correlation degree table between the components stored in the inter-component correlation table storage unit. An undefined component complementing means, and a distance between vectors for finding a distance between a feature vector of the new data whose undefined part has been complemented by the undefined component complementing means and a feature vector of the data stored in the data storage means. Calculating means, if the distance between the feature vectors calculated by the inter-vector distance calculating means is equal to or less than a predetermined threshold value, the data is compared with the new data. Neighborhood data determination means for determining proximity data to be stored, threshold value storage means for storing a predetermined threshold value used for determination of proximity data in the proximity data determination means, determination by the proximity data determination means A link direction for determining a direction of a link to be newly added between the proximity data and the new data from the category of the proximity data and the category of the new data by using a predetermined linking rule. Determining means; adding a link between the neighboring data and the new data in the direction determined by the link direction determining means, and storing information on the new data and the link in the data storage means; 5. The database registration device according to claim 4, further comprising: means. 6. The undefined component complementing means, which is required to obtain a complement value of an undefined component of a feature vector generated by the feature vector generating means and a feature vector of data stored in advance, 6. The method according to claim 5, wherein when there is no defined component having a correlation with the component, the undefined component is replaced with a special symbol which is always 0 regardless of the difference from any value. Database registration device.