JP5225350B2 - Information retrieval device - Google Patents

Description

  Embodiments described herein relate generally to an information search apparatus that outputs information that satisfies a search condition by keyword specification, for example.

  In the field of information retrieval, there is a problem of acquiring necessary information without excess or deficiency, and in information retrieval using keywords (words) in some form, there are many techniques that use the relationship between keywords to improve search accuracy. It is used.

  Search techniques for processing the relationship between keywords include: (1) It is possible to specify the keyword appearance conditions in detail using logical operators such as AND, OR, NOT, and (2) There is known a technique that can express the condition of information required by the searcher more accurately by specifying the position of the keyword in the search target, for example, by specifying the paragraph in which the keyword appears. However, these methods only improve in terms of strictly determining the presence / absence of keywords, and do not deal with the relationship between keywords in search target information. For example, even if “invention intended for computers” is assumed or “computers using the invention” is assumed, the conventional keyword specification type search device uses “computer AND invention” or There is a problem that the search condition can only be expressed as “invention AND computer” and the difference cannot be reflected in the search processing.

  In order to solve the above problem, a search device has been proposed in which a natural language sentence is received as a search condition and a condition is determined based on a dependency relationship between keywords included therein. However, a search device that performs condition determination based on the dependency relationship does not support search conditions expressed by enumerating only keywords. For example, dependency analysis is performed on a natural language sentence (clause), and there is a problem that a dependency structure based on parts of speech must be included in the data.

  Furthermore, dependency analysis uses grammatical relationships in natural language sentences (sections), which do not necessarily coincide with semantic relationships. For example, “red car”, “red car” and “red car” mean the same thing, but their dependency structures are all different, and only “red car” is related to “color”. The keyword is not included. Therefore, for example, when performing a search determination using dependency analysis under the condition of “red car”, “red car” satisfies the condition because “red” has a structure related to “car”. Since “red” is a structure related to “red” in “color”, the structure does not match and it is determined that the condition is not satisfied. As described above, if the same dependency structure is not specified by a keyword that can be determined to be the same, it cannot be determined that the semantic structure is the same, which causes a reduction in search recall.

  On the other hand, there is an apparatus that performs information retrieval using a semantic relationship between keywords instead of a grammatical relationship. However, in this device, the searcher must construct the semantic relationship specified as the search condition by himself / herself, and it is compatible with other information search systems that accept search conditions by specifying keywords as well as search devices by dependency analysis. There was no problem.

  On the other hand, there is a technique using concept search as a technique for improving the search accuracy by accepting only a keyword designated by a search performer as a search condition and statistically processing the relationship between the keywords. In this conceptual search, the distance between keywords is calculated from the keyword co-occurrence frequency in the given learning data, and the distance between the search target and the search condition is obtained using this, and the search is close to the search condition. The target is output as satisfying the condition. However, in a search method using statistical evaluation results as represented by concept search, condition determination is performed by quantitative comparison that abstracts the degree of “similarity” or “relevance”. The information search apparatus using the technique can only determine whether or not certain data satisfies the search condition from the relative relationship in the entire data to be searched. This may lead to a situation in which the result of whether or not a certain object satisfies the search condition changes depending on other registered data. In addition, unless a fixed threshold value is separately prepared to determine whether or not the condition is satisfied, a decisive factor that does not depend on the registered data, such as “XX cases are applicable and XX cases are not applicable”. It is not possible to evaluate. However, since such a threshold value cannot be intuitively specified by humans, it is necessary to perform an extremely expensive operation of repeatedly adjusting the threshold value through trial and error until a result that can be accepted by a searcher is obtained. I must.

Patent No. 2742115 JP-A-2005-276183 JP 2009-276826

Retrieval of subject information in patent information: Concept search and its limitations, The journal of Information Science and Technology Association 54 (7) pp.355-362 (Takehashi Akihiro (Tell Research)), 2004

  The present invention provides an information search apparatus that enables a search with a high recall and relevance rate at a low cost.

  An information search apparatus as an embodiment of the present invention includes an ontology database, a search target database, an index keyword database, a search condition reception unit, a semantic network generation unit, a search condition network generation unit, and an index network generation unit. A search condition determination unit and a search result output unit.

  The ontology database includes a plurality of nodes representing a plurality of concepts and connected in a hierarchy, and a plurality of directional links that connect the two nodes and represent the relationship between the concepts corresponding to the two nodes. And a second ontology having a plurality of nodes representing a plurality of the relations and connected in a hierarchical manner.

  The search target database stores a plurality of search target data.

  The index keyword database stores a list of index keywords for each search target data.

  The search condition receiving unit receives a list of search keywords specified as a search condition by a user.

  The semantic network generation unit receives a plurality of keywords, specifies a node or link representing a concept or relationship corresponding to the received keyword in the first ontology, and a path connecting between the specified nodes or links to the first ontology. A semantic network in which the identified node or link is connected by the route found by the search is generated by tracing the node and the link.

  The search condition network generation unit gives the search keyword received by the search condition reception unit as the keyword to the semantic network generation unit, and acquires the semantic network generated by the semantic network generation unit as a search condition network.

  The index network generation unit gives an index keyword included in the list for each search target data as the keyword to the semantic network generation unit, and acquires the semantic network generated by the semantic network generation unit as an index network.

  The search condition determination unit has the same structure as the search condition network, or the same structure as that obtained by replacing any node or link included in the search condition network with a node or link representing a lower concept or a lower relationship. An index network including the index network is detected from the index networks generated by the index network generation unit.

  The search result output unit acquires search target data corresponding to the index network detected by the search condition determination unit from the search target database, and outputs the acquired search target data.

It is a block diagram showing the internal structure of the whole information search device in this embodiment. FIG. 2 is a block diagram showing an internal configuration of an index network processing unit in FIG. FIG. 2 is a block diagram showing an internal configuration of a search processing unit in FIG. FIG. 2 is a block diagram showing an internal configuration of a semantic network generation unit in FIG. It is a figure which shows the example of ontology. It is a figure which shows the example of a concept table. It is a figure which shows the example of a relationship table. It is a figure which shows the example of a system | strain list | wrist. It is a figure which shows the example of a system | strain candidate table. It is a figure which shows the example of three search object data. It is a figure which shows the example of an index keyword table. It is a figure which shows the example of an index network. It is a figure which shows the other example of an index network. It is a figure which shows the further another example of an index network. It is a figure which shows the example of a concept instance table. It is a figure which shows the example of a relationship instance table. It is a figure which shows the example of a search keyword input screen. It is a figure which shows the example of a search keyword input screen. It is a figure which shows the example of a keyword holding table. It is a figure which shows the example of a search condition network. It is a figure which shows the example of a concept instance table. It is a figure which shows the example of a relationship instance table. It is a figure which shows the example of a query concept table. It is a figure which shows the format of a search result display screen. It is a figure which shows an example of a search result display screen. It is a figure which shows the other example of a search result display screen. It is a figure which shows the example of a display screen without a search result. It is a figure which shows the example of route data. It is a figure which shows the example of the semantic network in the middle of a production | generation. It is a figure which shows the example of a route search tree. It is a figure which shows the example of a route search tree table. It is a figure which shows the example of a search candidate queue with a priority. It is a figure which shows the example of the updated route search tree. It is a figure which shows the example of the search queue with a priority updated. It is a figure which shows the example of a search condition network evaluation table. It is a figure which shows the example of an application waiting stack. It is a figure which shows the example of a stack being applied. It is a flowchart which shows the whole processing flow of the information search of this embodiment. It is a flowchart which shows the production | generation processing flow of a system | strain connection matrix. It is a flowchart which shows the production | generation flow of a system | strain list | wrist. It is a flowchart explaining the production | generation process of the index network with respect to search object data. It is a flowchart explaining the production | generation process of a semantic network. It is a flowchart explaining a route data generation process. It is a flowchart which shows the identification flow of a system | strain. It is a flowchart which shows a route data generation flow. FIG. 46 is a partially enlarged view of FIG. 45. FIG. 46 is a partially enlarged view of FIG. 45. FIG. 46 is a partially enlarged view of FIG. 45. FIG. 46 is a partially enlarged view of FIG. 45. FIG. 46 is a partially enlarged view of FIG. 45. FIG. 46 is a partially enlarged view of FIG. 45. It is a flowchart which shows a connection instance acquisition flow. It is a flowchart which shows a semantic network expansion flow. FIG. 48 is a partially enlarged view of FIG. 47. FIG. 48 is a partially enlarged view of FIG. 47. It is a flowchart which shows the determination flow of search conditions. FIG. 49 is a partially enlarged view of FIG. 48. FIG. 49 is a partially enlarged view of FIG. 48. FIG. 49 is a partially enlarged view of FIG. 48. FIG. 49 is a partially enlarged view of FIG. 48. It is a flowchart which shows the acquisition flow of applicable matching. FIG. 50 is a partially enlarged view of FIG. 49. FIG. 50 is a partially enlarged view of FIG. 49. It is a flowchart which shows the determination flow of the applicability of matching. FIG. 51 is a partially enlarged view of FIG. 50. FIG. 51 is a partially enlarged view of FIG. 50. FIG. 51 is a partially enlarged view of FIG. 50. FIG. 51 is a partially enlarged view of FIG. 50.

  Embodiments of the present invention analyze an semantic relationship between keywords specified as a search condition based on an ontology in an information search apparatus that outputs information that satisfies a search condition based on keyword specification, and use this in search determination. The present invention relates to a technique for improving search accuracy and improving data sharing and reusability.

  More specifically, the embodiment of the present invention relates to an information search apparatus having the following functions.

1) A function that accepts keywords listed as search conditions and their word order, and refers to the ontology to generate a semantic network (search condition network) that includes concepts or relationships corresponding to the keywords.
2) A function that accepts keywords assigned as indexes to search targets and their word order, and refers to the ontology to generate a semantic network (index network) that includes concepts or relationships corresponding to the keywords.
3) A function that determines whether or not the search condition is satisfied depending on whether or not the search condition network generated in 1) above is included in the index network generated in 2) above.
4) Select the keyword received in 1) above and the search target that performs condition determination for the word order, and output the search target determined as satisfying the search condition in 3) above as the condition satisfied function

  The outline of the embodiment of the present invention will be described as follows.

  (1) In the issue of acquiring information necessary for information search by keyword specification without excess or deficiency, from the keyword specified as the search condition, the type of semantic relationship between them, the direction of the relationship, and the relationship By identifying other keywords to be used and making a determination using a semantic network generated therefrom as a search condition, it is possible to perform a search with a higher recall and relevance rate than the conventional method based on the presence / absence determination of each keyword.

  (2) Refer to the ontology and enter text data with parts of speech by changing the analysis procedure for the type and direction of the relationship between the keywords to be associated according to the type and word order of the keywords specified as the search conditions. For example, it compensates for the disadvantage of the conventional method that the relationship between keywords cannot be analyzed, and accepts only enumerated keywords as a search condition so that the relationship between them can be analyzed.

  (3) By referring to the ontology and supplementing the keywords that represent the meanings that are implicitly included as necessary, the keywords are selected to such an extent that the implicit semantic content can be identified by analyzing the relationship between the keywords. If specified, the same semantic structure is derived from different keyword pairs and processed.

  (4) By accepting enumerated keywords as search conditions and estimating the semantic relationship between them, it compensates for the drawbacks of the conventional method in which the searcher must directly specify the relationship between keywords. This saves input time and maintains compatibility with other information retrieval systems that accept search conditions by specifying keywords.

  (5) Handles the type of relationship between keywords, and determines whether a target satisfies the search condition by determining whether there is a specific type of relationship between specific keywords. To compensate for the disadvantages of the conventional method that can only perform relative evaluations and the disadvantages of the conventional method that requires adjustment work of judgment criteria, and the ontology to be used does not change, the search condition and only the individual judgment objects are used. It is possible to make a determination, and to achieve a search that does not require adjustment work of the determination criteria.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  1 to 4 are block diagrams showing an embodiment of the present invention. Among these, FIG. 1 shows the internal configuration of the entire information search apparatus in the present embodiment, FIG. 2 shows the internal configuration of the index network processing unit 7 in FIG. 1, and FIG. 3 shows the search processing unit in FIG. 9 shows the internal configuration of FIG. 9, and FIG. 4 shows the internal configuration of the semantic network generation unit 3 in FIG.

  In the present embodiment, the ontology 100 shown in FIG. 5 is recorded in the ontology DB 1 in advance. An ontology represents a concept or relationship that appears in the world that the system targets. An ontology can be expressed as a graph network in which the concept is a node and the relationship is a link in terms of data structure. For the ontology recording method and the ontology processing method, the ontology is expressed using, for example, the ontology description language OWL (Web Ontology Language) established by the World Wide Web Consortium (W3C), and Jena (URL: http: //jena.sourceforge.net/) and The OWL API (URL: http://owlapi.sourceforge.net/) may be used to manage and process the content. Ontology DB1 may record the network of Ontology 100 using a tabular memory structure, or if it is a DB (such as XML DB) that can hold structured documents, record the OWL description as it is in the DB. Also good.

  Details of ontology 100 in FIG. 5 will be described. The ontology 100 includes an ontology (first ontology) representing a conceptual system related to an elevator and an ontology (second ontology) representing a conceptual system related to a relationship. A square in the figure corresponds to one concept, and a character string in the square represents a concept label.

  In the first ontology, a plurality of elevator-related concepts are combined in a hierarchical manner, and a link representing a general-special relationship (is-a relationship) is set between concepts in a parent-child relationship. For example, “gearless elevators are specialized elevators” and “marble is a generalized grade”. In the first ontology, a link with a directional arrow is also set between concepts between non-parents. The relationship represented by this link is represented by an attached tag with <>. For example, “board” is connected with “material type” and “material”. Here, the transition rule holds for the general-special relationship. Since “gearless elevator is a specialization of elevator” and “elevator is a specialization of things”, it is considered that the relationship of “gearless elevator is a specialization of things” is established ( Inheritance).

  Also in the second ontology, a plurality of concepts related to relationships are combined in a hierarchical manner, and a link representing a general-special relationship (is-a relationship) is set between concepts in a parent-child relationship. The concepts included in the second ontology are “relationships” at the topmost concept, and all the concepts below it are relationships set between the concepts in the non-parent-child relationship of the first ontology. Similar to the first ontology, the transition rule holds.

  In the present embodiment, the ontology 100 is recorded in the format of the concept table 110 in FIG. 6 and the relationship table 120 in FIG.

  In the search target DB 4, data to be searched is recorded in advance in the table format of the search target data table 200 (FIG. 10). An index keyword is assigned to each search target data. Assume that they are recorded in the index keyword DB 5 in the table format of the index keyword table 210 shown in FIG.

  The information search processing flow of this embodiment will be described below along the flow of FIG. 38 starting from processing S0000.

(Processing S0001)
The index network processing unit 7 executes a process S0300 (FIG. 41 described later) to generate an index network corresponding to each data to be searched.

  In the present embodiment, index networks 310, 320, and 330 having the contents shown in FIGS. 12, 13, and 14 are generated for the search target data of search target IDs 2100, 2200, and 2300 in FIG. Recorded in the network DB72 (Figure 2). In this embodiment, it is assumed that the index network is recorded in the index network DB 72 in the table format configured as shown in FIGS. 15 and 16. For example, the index network 310 is shown in the concept instance table 311 and the concept instance table 312. It shall be recorded as a record.

  Here, the concept instance and the relation instance are nodes and links constituting the semantic network (search condition network or index network) in the present embodiment. A concept instance and a relationship instance each refer to one of the concepts or relationships described in the ontology. For example, the record of the concept instance “3101” in the concept instance table 311 of FIG. 15 is a node of the index network in which the concept instance is generated for the search target data of the search target ID “2100” (FIG. 10). It is also recorded that it refers to the concept “gearless elevator” (concept ID 117) (FIG. 6) defined in the ontology. Further, for example, the record of the relation instance “3121” in the relation instance table 312 of FIG. 16 is included in the index network in which the relation instance is generated for the search target data of the search target ID “2100” (FIG. 10), And the directional link from concept instance “3101” (see gearless elevator) to concept instance “3102” (see door) and the relation “part (0202)” defined in the ontology (Figure 7) Is recorded. The index networks 320 and 330 (FIGS. 13 and 14) are recorded in the same format, and details are omitted. (The method of creating a semantic network (index network) will be described in detail later)

(Processing S0002)
Next, the search condition receiving unit 8 receives a keyword (search keyword) and a word order as search conditions and records them as a keyword list. The search condition receiving unit 8 provides the search keyword input screen shown in FIG. 17 for example to the search executor, acquires the input contents of the search condition input unit at the timing when the search button is pressed, and decomposes this into keyword units Record the completed list. In the present embodiment, the character string “elevator side plate mirror-finished HL processed material SUS” input in FIG. 18 (a) search keyword input screen example 1 is acquired, and the input is divided for each blank character, and each of them is a keyword. Are recorded in the format of the keyword holding table 400 shown in FIG. The keyword holding table 400 holds keywords input on the search keyword input screen as a keyword list with ranking, and assigns higher rankings to the left keyword on the input screen.

(Processing S0003)
Next, the search condition network generation unit 91 of the search processing unit 9 executes the process S0400 (FIG. 42 described later) using the keyword list recorded in the keyword holding table 400 of FIG. 19, and the search shown in FIG. A condition network 410 is generated and recorded in the format of a concept instance table shown in FIG. 21 and a relation instance table shown in FIG. In this embodiment, the structure of the table that records the search condition network is obtained by deleting the column of the search target ID from each of the concept instance table 311 (FIG. 15) and the related instance table 312 (FIG. 16) representing the index network. The configuration is the same. (The semantic network (search condition network) generation method will be described in detail later)

(Processing S0004)
The index network acquisition unit 92 refers to the concept instance table 411 of FIG. 21 recorded in step S0003, and determines whether each concept instance is recorded as a related party concept or not by referring to the relationship instance table 412 of FIG. Determine with reference to. Then, the concept instance “4101” (elevator) that is not recorded as the related concept instance is set as the root element of the concept instance table 411, and “elevator” that is the concept referred to by the concept instance “4101” is acquired. Next, referring to the concept table 110 (FIG. 6) of the ontology DB 1 via the ontology management unit 2, “gear-type elevator” and “gearless elevator”, which are subordinate concepts of the concept “elevator”, are acquired. Then, a set having these acquired concepts as elements is recorded as one record in the query concept table 500 (FIG. 23).

  Here, “a concept that is a subordinate concept of the concept“ elevator ”” refers to a concept in which the concept “elevator” is directly or indirectly a superordinate concept. In addition, “a concept in which the concept“ elevator ”is directly a superordinate concept” refers to a concept in which the value of the superordinate concept matches “elevator” in the concept table 110 (FIG. 6). This acquisition can be acquired, for example, by searching a higher concept column in the concept table 110. In addition, “the concept“ elevator ”is indirectly a superordinate concept” means that the concept “elevator” appears directly somewhere in the concept where the concept “elevator” appears. It refers to a concept that does not have the concept “elevator” as a superordinate concept. To acquire these concepts, for example, first prepare a cue that contains only concepts whose concept “elevator” is directly a superordinate concept, record “polled concepts”, and make them all direct superordinate concepts. It can be obtained by repeatedly performing the process of “putting the concept of“ to the queue ”until the queue becomes empty.

  When there are a plurality of root elements in the search condition network, the above process is repeatedly executed, and a record of the concept set is recorded for each root element.

(Processing S0005)
The index network management unit 73 sequentially acquires the concept sets recorded in the process S0004, and is indexed on the condition that “all records have a concept instance that refers to any one element included in the concept set”. Get the index network recorded in DB72 (Fig. 15, Fig. 16). In the case of the present embodiment, the query concept table 500 is first referred to, and the concept set “{elevator, geared elevator, gearless elevator}” is acquired. Next, based on the concept instance table 311 (FIG. 15), the index network (that is, the index network 310) of the search target ID 2100 in which “gearless elevator” that is one element of the concept set is recorded is acquired. Similar processing is performed, and the index networks 320 and 330 are also acquired.

(Processing S0006)
Whether the search condition determination unit 93 designates each of the search condition network 410 and the index networks 310, 320, and 330, executes the processing S1100 (FIG. 48 described later), and whether or not the search condition is satisfied for each index network. Judgment is made. In the present embodiment, the index network 310 is determined as “does not satisfy the search condition”, and the index networks 320 and 330 are both determined as “the search condition is satisfied”. (The search condition determination method will be described in detail later)

(Processing S0007)
The search processing unit 9 receives the result of the process S0006, executes the process S0008 for the index networks 320 and 330 that are determined to satisfy the “search condition”, and the index network 310 that is determined to not satisfy the “search condition” Performs the process S0010.

(Processing S0008)
The index network management unit 73 acquires the search target IDs “2200” and “2300” of the record from the concept instance table 311 in which each of the index networks 320 and 330 determined to satisfy the search condition is recorded. The search target data management unit 6 acquires corresponding search target data from the search target data table 200 of FIG.

(Processing S0009)
The search result output unit 10 acquires the search target data acquired in step S0008, and uses the search result display screen (FIG. 24) to display the search target ID, the index network, and the search target data as a search target ID. Displayed in the area, index network display area, and search target data display area. In the present embodiment, search result display screen example 1 (FIG. 25) is displayed.

(Processing S0010)
If the search processing unit 9 determines whether or not the search condition is satisfied for all the index networks acquired in the process S0005, the process S0011 is executed, and if there is an unexecuted index network, the process S0006 is executed. And continue processing.

(Processing S0011)
Ends output of search results. At this time, if there is no data to be output by the search result output unit 10 (this can be easily done by recording the number of times of output or using a flag indicating whether output processing has been performed or not). May be notified using the search result display screen shown in FIG. 27, such as “no corresponding data”.

  Next, an index network generation process for search target data in the present embodiment will be described along the flow of FIG. 41 starting from processing S0300.

  In this embodiment, there is a search target DB 4 having the search target data table 200 of FIG. 10 in which data to be searched is recorded in advance, and an index table 210 in which keywords assigned to the search target data in order are recorded. A case where the index keyword DB5 of FIG. 11 is prepared will be described. Each row of the search target data table 200 corresponds to one search target data. Each row of the index keyword table 210 (FIG. 11) corresponds to one of the keywords assigned to the search target data. The search target ID, the order of assignment (word order), and the assignment of the search target to be given It is assumed that it is composed of the keyword strings that have been processed.

  In the present embodiment, a case where text data is used as search target data will be described. However, any data search target may be used, such as a table, a figure, an image, a moving image, and voice. In addition, devices that manage data to be searched together with keywords, a method for extracting and generating keywords for that purpose, and a method for editing and managing keywords to be assigned are well-known techniques (Tokuto 4224131, No. 3173411) may be used.

(Processing S0301)
The search target data management unit 6 accesses the search target DB 4 (FIG. 10), and first acquires the search target ID “2100” in the order of records. Next, all keywords and word orders corresponding to the search target ID “2100” are acquired from the index keyword DB 5 in FIG.

(Processing S0302)
The index network generation unit 71 of the network processing unit 7 acquires the keyword and word order acquired in the processing S0301 “gearless elevator (1), door plate (2), mirror finish (3), aluminum (4), side plate (5), SUS (6), HL processing (7) "(numbers in parentheses are in word order) is specified as a keyword list, and processing S0400 (FIG. 9 described later) is executed, and the generated index network 310 (FIG. 12) Is recorded in the index network DB as an index network for the search target data with the search target ID “2100”. As described above, in this embodiment, the index network is recorded in the index network DB 72 in the table format having the configuration shown in FIGS.

(Processing S303)
It is determined whether or not an index network has been generated and recorded for all search target data registered in the search target DB4. At this time, if there is an unprocessed item, the process returns to step S0301, and the process is continued. In this embodiment, following the process assigned to the search target ID “2100”, the keywords “gearless elevator (1), side plate (2), SUS (3), HL processing” assigned to the search target ID “2200”. The index network 320 (Fig. 13) is generated from (4), mirror finish (5), side plate (6), mirror finish (7), aluminum (8) ", and the keyword" 2300 "assigned to the search target ID" 2300 " Index network 330 from gear type elevator (1), side plate (2), SUS (3), HL processing (4), mirror surface processing (5), side plate (6), satin processing (7), aluminum (8) 14) is generated and recorded in the index network DB 72, the process S0304 is executed.

(Processing S0304)
The index network generation / recording process for the search target data is terminated.

Next, the generation process of the semantic network (search condition network, index network) in the present embodiment will be described along the flow of FIG. 42 starting from the processing S0400. Here, the keywords `` gearless elevator (1), door plate (2), mirror finish (3), aluminum (4), side plate (5), given to the search target data of the search target ID `` 2100 '' in FIG. A processing example for a list of keywords related to “SUS (6), HL processing (7)” (numbers in parentheses are in word order) will be described.
(Processing S0401)
First, the semantic network generation unit 6 generates and records a search end list with 0 elements and 0 elements.

  Here, the search end list is a record of concept instances that end the search when found when an instance corresponding to a certain keyword is used as a starting point.

(Processing S0402)
The semantic network generation unit 6 has the keyword list “gearless elevator (1), door plate (2), mirror finish (3), aluminum (4), side plate (5) specified in the above process S0302 (FIG. 41). , SUS (6), HL processing (7) "(numbers in parentheses are in word order), and obtain the keyword" gearless elevator "and word order" 1 "that have the smallest word order value. Next, using the acquired keyword “gearless elevator”, the keyword columns in the concept table 110 in FIG. 6 and the relationship table 120 in FIG. 7 are searched to acquire concepts and relationships in which the keywords completely match, and the list (object list) ) Hold as. If keywords match for a plurality of concepts or relationships, all of them are recorded as a list (object list). In this embodiment, since only the concept “gearless elevator” matches the keyword “gearless elevator”, a list “{gearless elevator}” having only the concept as an element is recorded. Note that if there is no corresponding concept or relationship, the keyword may be deleted from the keyword list.

(Processing S0403)
The semantic network generation unit 6 designates the list of objects (concepts or relationships) “{gearless elevator}” recorded in process S0402 and the search end list with 0 elements, and executes process S0600 (FIG. 43). Then, route data 760 (FIG. 28 (a)) is acquired, and a null value is acquired as a connection instance. (Route data and connection instance generation / acquisition method will be described in detail later)
The route data is a list of concepts and relationships representing the connection between keywords, and the connection instance is a concept instance in the semantic network that becomes a contact point for connecting the route data. As will be described later, the semantic network is expanded by generating an instance of an object included as an element in the route data and sequentially connecting to the concept instance designated as the connection instance.

(Processing S0404)
The semantic network generation unit 6 specifies the route data 760 (FIG. 28 (a)) acquired in process S0403 and a null value as a connection instance, executes process S1000 (FIG. 47 to be described later), and executes the semantic network. To expand. In the case of this embodiment, before the process, the number of nodes is expanded from 0 to the semantic network with 1 node as shown in FIG. (Semantic network expansion method will be described in detail later)

(Processing S0405)
Whether or not the semantic network expansion processing has been executed for all the keywords is checked based on whether or not there is a keyword having a word order larger than the keyword acquired in processing S0402 (that is, a lower keyword). If there is, processing S0406 is performed. Is executed to update the search end list, and if it does not exist, the process is executed for all the keywords, and the process S0407 is executed.

  In the case of the present embodiment, following the process of “gearless elevator” with the word order 1 and “door plate” with the word order 2 exists, after executing the process S0406, the process returns to the process 0402 and obtained in the process S0403. The semantic network 312 (FIG. 29 (b)) is obtained by executing the process S0404 on the route data 761 (FIG. 28 (b)) and the connection instance “3101” (see the gearless elevator).

  Also, after the process S0406, the path data 762 (FIG. 28 (c)) obtained in the process S0403 and the connection instance “3103” (refer to the board) after the process S0402 for the “mirror finish” with the word order 3 )), The semantic network 312 (FIG. 29 (c)) is obtained.

  The same is repeated thereafter, and the processes S0402 to S0404 are executed for the “HL processing” whose word order is 7, and when the semantic network (in this embodiment, the index network) is in the state of FIG. Since there is no keyword greater than 7, process S0407 is executed.

(Processing S0406)
After the semantic network generation unit 6 clears the elements of the search termination list, it acquires the end-of-path instance of the semantic network at that time, and exists on the path from the end-of-path instance to the concept instance that is the root node of the semantic network. A list of concept instances to be created is created and recorded as a search termination list. This search end list is used in the next process S0403.

  Here, the end-of-path instance is a concept instance (that is, a so-called leaf node) farthest from the root in the expanded portion when the index network is expanded.

  In the present embodiment, the process for the “gearless elevator” having the word order of 1 yields the path end instance “3101” (see the gearless elevator), which is the root node (the parent concept in the semantic network). Therefore, the search end list “{3101}” is obtained and used in the process for the “doorboard” having the word order of 2.

  In addition, in the process for “doorboard” with word order 2, route end instance “3103” (see board) is obtained, concept instances are added in order from the node of the board, and processing is continued until there is no parent node. By repeating, the search end list “{3103, 3102, 3101}” (refer to the plate, door, and gearless elevator, respectively) is obtained and used in the process for “mirror finish” with word order 3.

(Processing S0407)
The semantic network generation process ends.

  Next, route data generation processing in the present embodiment will be described along the flow of FIG. 43 starting from processing S0600. Here, the keywords `` gearless elevator (1), door plate (2), mirror finish (3), aluminum (4), side plate (5), SUS (6 ), HL processing (7) ”(numerical values in parentheses are in word order), a case will be described in which path data is generated for“ mirror processing ”in which the word order is 3. At this time, the semantic network is in the state of the semantic network 312 (FIG. 29 (b)), the search termination list is “{3103, 3102, 3101}” (refer to the board, door, and gearless elevator, respectively). The corresponding object list is “{mirror finish}”.

(Processing S0601)
The maximum searched route length receiving unit 34 acquires the maximum route length L. The maximum path length receiving unit 34 may be acquired from, for example, the search maximum path length input unit on the search keyword input screen shown in FIG. In the present embodiment, as shown in FIG. 18 (a) search keyword input screen example 1, the maximum path length L = 5 is acquired.

  Here, the maximum search path length is a specified value of the maximum number of steps to be followed when searching for a path (ontology) from a specified keyword (here, mirror processing). Even if this value is reached, if the specified keyword does not reach one of the concepts in the search termination list, the route data generation is terminated, the connection instance is set to a null value, and the route data generation process is terminated. (In this case, there can be multiple semantic networks generated).

(Processing S0602)
If the number of elements in the designated search end list is zero, process S0603 is executed, and otherwise, process S0605 is executed. In the present embodiment, since the number of elements in the search end list is not 0, process S0605 is executed.

(Processing S0605)
The route data generation unit 31 calculates and records a search priority matrix V _{m, L-1} (V _{3,4 in} this embodiment) for the word order value m = 3 and the maximum search route length L = 5.

から求められる行列である。
まず、オントロジー経路データ生成部31が、処理S0100（後述する図39）を実行し、系統接続行列O（後述）および系統リスト（後述）を求める。本実施形態では、オントロジー100より、

と図8の系統リスト130「｛板、モノ、属性値｝」が得られる。系統は、オントロジーを複数に分割したときの各部分的オントロジーである。この例では「板」、「モノ」、「属性値」は各系統を識別し、各系統の最上位のノードでもある。 Here, the search priority matrix V _{x, y} is (number of elements in the system list + 1) × 1 matrix,

Is a matrix obtained from
First, the ontology path data generation unit 31 executes a process S0100 (FIG. 39 described later) to obtain a system connection matrix O (described later) and a system list (described later). In this embodiment, from ontology 100,

And the system list 130 “{plate, thing, attribute value}” of FIG. 8 is obtained. A system is each partial ontology when the ontology is divided into a plurality of parts. In this example, “board”, “thing”, and “attribute value” identify each system and are also the highest nodes of each system.

を記録する。（系統接続行列および系統リストの生・取得方法は後で詳述する） Next, the ontology path data generation unit 31 generates a zero matrix of (number of elements in the system list + 1) × 1 and records it as the matrix V _3,0 . Next, the keywords “gearless elevator” and “doorboard” with word order less than 3 are acquired in order, and all concepts / relationships that match the keywords are acquired using ontology management unit 2 and recorded so as not to overlap. . In this embodiment, “{gearless elevator} {door plate (0203)}” is obtained. Next, processing S700 (FIG. 44 to be described later) is performed on the concept and the relationship “gearless elevator” and “door plate (0203)” as elements, and the system “mono” and “plate” including each are acquired. Then, the acquired system is compared with the system list 130 to obtain the ranking “2” and “1” of each system, and 1 is added to each of the 2 rows and 1 column element and the 1 row and 1 column element of V _3,0 . In this way

Record. (The system connection matrix and system list generation / acquisition method will be described in detail later)

(Processing S0606)
The route data generation unit 31 sequentially acquires the elements of the object list “{mirror finish}” corresponding to the keyword of the word order = 3, and searches the end list “{3103, 3102, 3101}” (plate, door, gearless respectively) The priority matrix V ₃ and ₄ calculated in the process S0605 is designated, and the process S0800 (FIG. 45 described later) is executed to acquire route data and connection instances. In the present embodiment, route data 762 “{plate, finish (0208), mirror finish}” (FIG. 28 (c)) and connection instance “3103” (plate) are provided as route data for the object “mirror finish”. To be acquired. (The route data generation and connection instance acquisition methods will be described in detail later)

(Processing S0607 ~ S0609)
Each time the process of process S0606 ends, the path data generation unit 31 compares the number of concept instances of the path data acquired in process S0606 with the number of concept instances of the path data recorded at that time. Record as route data and delete the larger one. At this time, if route data is not recorded, the route data acquired in step S0606 is recorded as it is. Then, when the comparison of the path data is completed for all the objects corresponding to the keyword of word order = 3, the process S0610 is executed.

  In the present embodiment, since there is one element of the object corresponding to the keyword of word order = 3, the process S0610 is executed without repeating the comparison process.

(Processing S0610)
The route data generation process is terminated.

  Next, the generation process of the system connection matrix O will be described along the flow of FIG. 39 starting from the process S0100.

(Processing S0101)
The ontology management unit 2 designates the ontology 100 and executes the process S0200 (FIG. 40 described later) to acquire the system list 130.

(Processing S0102)
The ontology management unit 2 calculates and records the system connection matrix O from the system list 130 and the ontology 100.

を要素とする行列である。 Here, the system connection matrix O is a matrix of (number of elements of system list + 1) × (number of elements of system list + 1),

Is a matrix.

  In the case of the present embodiment, first, based on the number of elements “3” in the system list 130, the 4 × 4 matrix O is recorded in the working memory, and the values of all elements are set to 0.

が求まり、これを記録する。 Next, processing S0700 (FIG. 44 described later) is performed on each of the relation source concept “mono” and the relation destination concept “mono” in the relation “part (0201)” of the relation table 120 (FIG. 7). The included systems are acquired and compared to determine whether they are the same. In this case, since the included systems are both “mono” and equal to each other (that is, diagonal components), the matrix O is not updated.
Next, with respect to the relationship “part (0202)” in the relationship table 120, the system including the relationship source concept “elevator” and the relationship destination concept “door” is acquired, and whether or not they are equal is determined. Also in this case, since the included systems are both “mono” and equal to each other, the matrix O is not updated.
Next, with respect to the relationship “door plate (0203)” in the relationship table 120, the system including the relationship source concept “door” and the relationship destination concept “plate” is acquired, and whether or not they are equal is determined. In this case, the systems to which the relation source concept “door” and the relation destination concept “board” belong are “thing” and “board”, which are not equal. Therefore, each system is collated with the system list 130 to obtain the number of each system. In this case, it is obtained that the “thing” on the relation source side is the “2” th, and the “board” on the relation destination side is the “1” th system. Therefore, 1 is added to the value of the element O ₁₂ of the matrix O.
Similarly, for all the records in the relationship table 120 (FIG. 7), the system including the relationship source concept and the relationship destination concept is compared, and if the systems are different, the corresponding elements according to the definition of the matrix O. By repeating the operation of adding 1 to

Is obtained and recorded.

(Processing S0102)
The calculation process of the system connection matrix O is terminated.

  Next, the flow of generating a system list will be described along the flow of FIG. 40 starting from processing S0200.

(Processing S0201)
The ontology management unit 2 refers to the relationship table 120 (FIG. 7), sets the concept recorded as the related party concept as a system candidate, and records it in the system candidate table 140 together with the number of records (referenced number). In addition, a system | strain candidate is a concept which can become a system | strain, It changes as a system | strain by subsequent processes, or is deleted from a system | strain candidate.

  In the present embodiment, the system candidate table 140 shown in FIG. 9 is recorded, which is obtained as a result of the above processing for the relation table 120 (FIG. 7). For example, if the number of references to the concept “board” is “3”, the concept “board” is recorded in the relationship table 120 as 3 related records with the relationship IDs “0203”, “0204”, and “0205”. By being. For example, if the concept recorded in the related concept is included in the system candidate concept in the system candidate table 140, the number of the referenced concept is as follows. The number may be added by 1 and if not included, a new line may be added and recorded as a systematic candidate concept, and the process may be obtained by repeating the process of “referring to 1 referenced concept number”.

(Processing S0202)
The ontology management unit 2 refers to the system candidate table 140 (FIG. 9) and checks whether there is a system candidate. If it exists, the process S0203 is executed, and if it does not exist, the process S0207 is executed. When the system candidate table is in the state of FIG. 9, since there are system candidates, the process S0203 is executed.

(Processing S0203)
The ontology management unit refers to the system candidate table 140 (FIG. 9) and selects the concept with the largest number of references (assuming that the concept ID with the smallest concept ID number is selected). In the present embodiment, first, “board” is added to the system list.

(Processing S0204 to S0206)
It is determined whether or not a concept that is a superordinate concept of the concept selected in process S0203 is included in the system list, and if not included, the concept acquired in S0203 is added to the system list, and then the system candidate table Delete from.

  In the case of the present embodiment, after “board” is added to the system list, “mono” and “attribute value” are added to the system list, and the system list is in the state of FIG. The “board”, “thing”, and “attribute value” are deleted from the system candidate table. “Door”, “Metal”, “Finish Type”, and “Security Window” are deleted from the system candidate table without being added to the system list. Then, when the processing is completed for all the system candidates, processing S0207 is executed.

(Process 0207)
The system list acquisition process ends.

  Next, a system identification flow will be described along the flow of FIG. 44 starting from processing S0700.

(Processing S0701)
The ontology management unit 2 acquires an object designated as a system acquisition, and if the object is a concept, records the object as a system acquisition target. On the other hand, if the object is a relationship, the related party concept of the relationship is recorded as a system acquisition target.

  As the present embodiment, a case where the concept “gearless elevator” and the relationship “door plate (0203)” are designated will be described. First, in the case of the concept “gearless elevator”, the system acquisition target is recorded as “gearless elevator”. On the other hand, when the relationship “doorboard (0203)” is designated, the relationship table 120 (FIG. 7) is referred to, and the relationship destination concept “board” of the relationship is recorded as a system acquisition target.

(Processing S0702)
The ontology management unit 2 determines whether or not the system acquisition target “gearless elevator” and “plate” recorded in step S0701 are included in the system list 130 (FIG. 8). If included, the system acquisition target is recorded, and the process S0703 is executed. If not included, the process S0704 is executed.

  In the present embodiment, since the system acquisition target “gearless elevator” is not included in the system list 130, the process S0704 is executed. On the other hand, since the system acquisition target “board” is included in the system list 130, the process S0703 is executed.

(Processing S0703)
The ontology management unit 2 records the system acquisition target as a system. In the case of the system acquisition target “board”, “board” is recorded as the system.

(Processing S0704 to S0706)
The ontology management unit 2 sets the superordinate concept of the system acquisition target as the system acquisition target, and executes the process S0702. At this time, when a concept having no superordinate concept is a system acquisition target, the system is recorded as a null value, and the process S0707 is executed.

  In the case of this embodiment, for the system acquisition target “gearless elevator”, the superordinate concept “elevator” is set as the system acquisition target again, and the process is repeated. Since it is included in FIG. 8), the process S0707 is executed.

(Processing S0707)
The route identification process is terminated.

Next, the route data generation and connection instance acquisition flow will be described along the flow of FIG. 45 starting from processing S0800. 45A, 45B, 45C, 45D, 45E, and 45F are partially enlarged views of FIG. Here, the object list “{mirror finish}”, the search end list “{3103, 3102, 3101}” (refer to the plate, door, and gearless elevator, respectively) and the priority matrix V _3,4 are specified. The processing in the case will be described.

(Processing S0801)
The route data generation unit 31 generates a route search tree and records it. In addition, as a means for expressing the route search tree, the search node as a component thereof, and the object (concept or relationship) referred to by the search node on the working memory, for example, the route search for the route search tree 730 (FIG. 30). A tabular data structure such as the tree table 740 (FIG. 40) can be used, but other methods capable of processing equivalent contents may be used.

  Here, the route search tree is data of a tree structure, and has a search node that refers to an object corresponding to a specified keyword as a root. A search node is a node that is an element of a route search tree, and always has one reference to an object defined by an ontology. In the route search process, it is assumed that the route is found by expanding the route search tree while referring to the ontology, and reaching one of the concepts referred to by the instances included in the search end list.

(Processing S0802)
The route data generation unit 31 acquires the designated object, and sets a search node that refers to the object as a root node of the route search tree 730 (FIG. 30). For example, in the route search tree table 740 of FIG. 31, it can be identified that the value of the parent node is empty (null value) so that it is the root.

  In this embodiment, a node 7301 having a reference to the concept “mirror processing” is recorded as the root of the route search tree 730.

(Processing S0803 to S0804)
The route data generation unit 31 generates a priority search candidate queue 750 (FIG. 32), and pushes the root node of the route search tree 730 with priority 0.

  The search candidate queue with priority is one of the queues with priority, and is a queue for storing a node that refers to an object to be searched next in order to generate search data.

In the case of this embodiment, first, the mirror processing node 7301 (priority = 0) is stored in the queue.

(Processing S0805)
The route data generation unit 31 refers to the element of the priority-added search candidate queue 750 and executes the process S0806 because the node 7301 exists in the queue.

(Processing S0806)
The route data generation unit 31 pulls one search node and its priority p from the priority search candidate queue 750, and acquires the node 7301 (priority = 0) that is the lowest priority node.

(Processing S0807)
At this time, since the object “mirror finishing” referred to by the search node 7301 (priority “0”) acquired first is a concept, processing S0808 is executed.

(Processing S0808 to Processing S0816)
The route data generation unit 31 acquires a relationship having the concept “mirror finish” as a related concept. In this case, since there is no relationship that satisfies the condition, the route search tree 730 is not updated.

(Processing S0817 to Processing S0819)
The path data generation unit 31 refers to the ontology management unit 2 to acquire the “finishing type” that is a superordinate concept of the concept “mirror finishing”, generates a node 7302 that refers to this, and generates a path as a child of the node 7301. Add to search tree 730. Also, the node 7302 is pushed to the priority queue 750 with the priority “1”.

(Processing S0815)
When processing S0814 is executed in a state where the route search tree is 731 (FIG. 33) and the priority search queue is 751 (FIG. 34) by repeating the processing S0805 to 0819, the connection instance from the processing S0900 (described later) Since 3103 is acquired, this and the node 7305 at the end of the priority queue 751 are specified, and the process S0822 is executed. In FIG. 34, the priority value is rounded off to the second decimal place. (The connection instance acquisition method will be described in detail later)
(Processing S0822 to S0827)
First, the concept “0104” (board) referred to by the node 7305 is added to the path data, and the objects referred to by the parent node are sequentially added to the path data, so that the path data 761 “{board, finish (0208), Mirror finish} ”(FIG. 28C) is generated.

(Processing S0828)
Since the element “plate” at the head of the route data matches the concept “plate” referred to by the connection instance 3103, the processing S0838 is executed, and the route data generation and connection instance acquisition processing is terminated.

  Next, a connection instance acquisition flow will be described along the flow of FIG. 46 starting from processing S0900.

  Here, an embodiment will be described in which the search termination list is “{3103, 3102, 3101}” (refer to plates, doors, and gearless elevators, respectively), and the priority queue is in the 751 state. At this time, the concept “board” referred to by the concept instance “3103” that is the first element of the search termination list is equal to the concept “board” referenced by the node at the end of the priority-added queue 751, so the connection instance “ 3103 "is recorded, and the process 0900 ends.

  Next, the semantic network expansion flow will be described along the flow of FIG. 47 starting from processing S1000. 47A and 47B are partial enlarged views of FIG.

  In this embodiment, when the semantic network 312 in FIG. 29B, the route data 762 “{plate, finish (0208), mirror finish}” in FIG. 28C, and the connection destination instance 3103 are given. Processing will be described.

  First, the connection destination instance 3103 is recorded as a parent instance. Since the parent instance is not a null value, the first concept “board” is acquired from the route data 762 in FIG. Since this matches the concept “board” referenced by the parent instance 3103, the reference concept of the parent instance is not updated. Then, the concept “plate” as the first element is deleted from the route data 762. Since there are still elements in the route data 762, the relation “finish (0208)” which is the next element is acquired. First, a relation instance 3124 that refers to this is generated and recorded. Next, the concept instance 3103 designated as the parent instance is designated and recorded as the relation source concept instance of the relation instance 3124. Then, the relationship “finish (0208)” that is the first element is deleted from the route data 762. Next, the concept “mirror finish” as the leading element is acquired from the path data 762. First, a concept instance 3104 that refers to this is generated and recorded. Then, the concept instance 3104 is designated as the related concept instance of the related instance 3124 and recorded. Then, the concept “mirror finish” as the first element is deleted from the path data 762.

  At this stage, the route data 762 has no elements, and the process 1000 ends. At this time, the semantic network 312 is recorded in the state updated to the state 313 in FIG. 29 (c).

  Next, the search condition determination flow will be described along the flow of FIG. 48 starting from step S1100. 48A, 48B, 48C, and 48D are partially enlarged views of FIG.

  Here, the index network 320 in FIG. 13 and the search condition network 410 in FIG. 20 are designated, and an example is a process of determining whether or not there is a corresponding instance in the index network 320 for all instances of the search condition network 410. explain.

  Note that associating an instance of the search condition network with an instance of the index network is called “matching”, and the association is called “matching”. In matching, an instance on the search condition network side is called a “matching source (instance)”, and an instance on the index network side is called a “matching destination (instance)”. At this time, the concept / relationship referenced by the matching destination instance must be equal to or lower than the concept / relationship referenced by the matching source instance. In addition, “matchable / unmatchable” means that a new matching can be added / impossible within a range that matches a predetermined matching. In addition, the state of a search condition network instance for which no matching destination is defined is referred to as “matching undecided”.

(Processing S1101 to Processing S1102)
The search condition determination unit 93 of the search processing unit 9 creates an application waiting stack and an application stack.

  The application waiting stack is a stack for recording matching candidates. On the other hand, the stack being applied is for recording the matching that is actually applied. The correspondence recorded in the application stack is a restriction on other correspondences, but the correspondence recorded in the application waiting stack is a candidate before application.

(Processing S1103 to S1108)
The search condition determination unit 93 records the degree of influence and the number of candidates for the instances that do not have matching in the search condition network 410.

  The degree of influence is a numerical value indicating how much restriction is imposed on other instances in the search condition network when a corresponding instance in the index network is determined for the instance. The influence degree of a concept instance is the number of undecided matches among relation instances that refer to the concept instance as a relation destination concept instance or a relation source instance. The influence degree of the relationship instance is the number of undecided matches among the relationship destination concept instances or the relationship source concept instances.

  On the other hand, the number of candidates is the number of matching destinations that can be matched in the index network for an instance of the search condition network. It is equivalent that the number of candidates = 0 cannot be matched to any instance of the index network.

  In the case of the present embodiment, since no matching is recorded in the applied stack, first, the number of candidates and the degree of influence are obtained for all concept instances and relation instances included in the search condition network 410. Recorded as an evaluation table 800 (FIG. 35 (a)). The number of candidates can be obtained by designating an instance, an index network, and a stack being applied, and executing processing S1200 (FIG. 49 described later) to obtain matching candidates. For example, by specifying the conceptual instance “4102” (board) in FIG. 20, the index network 320 in FIG. 13, and an empty applied stack, and executing the process 1200 (FIG. 49), a candidate “{ 3202 and 3206} ”(see the respective plates), the number of elements 2 is recorded as the number of candidates for the concept instance“ 4102 ”. On the other hand, in the search condition network 410, the relationship instances in which the concept instance 4102 (board) is the relationship source or destination concept instance are “4121”, “4122”, “4123”, “4124” (side plate, finish, finish, (Refer to material)), the impact number = 4 is recorded.

(Processing S1109 to S1110)
Instances “4101”, “4104”, and “4105” (refer to elevator, HL processing, and SUS, respectively) are acquired as instances with the smallest number of candidates from the search condition network evaluation table 800 in FIG. Since the number of these candidates is not 0 and the influences are all equal (impact = 1), applicable matching for all of them is obtained by executing the process 1200 (FIG. 49) and pushed to the application waiting stack. . At this time, the application waiting stack becomes the state of the application waiting stack 900 (FIG. 36 (a)). (Applicable matching acquisition methods will be described in detail later)
(Processing S1111 ~)
Next, since the stack to be applied is not empty, pop and obtain matching “4105-3203”. Since the stack being applied is empty, all the contents of the search condition network evaluation table 800 are cleared and applied. The matching “4105-3203” is added to the stack (FIG. 37 (a)).

  Next, processing S1103 to S1110 is repeated. First, the search condition network evaluation table 810 (FIG. 35 (b)) is recorded for the instances other than the concept / relationship existing in the applied stack. At this time, the instance having the smallest number of candidates is obtained, and “4101”, “4104”, and “4124” are obtained as the instances having the greatest influence from among them, and the processing 1200 (FIG. 49) is executed for each of them. Push the obtained matching destination to the stack waiting for application. At this time, the application waiting stack becomes the application waiting stack 910 (FIG. 36B). Then, since the application waiting stack is not empty, it is popped to obtain the matching “4124-3222”. At this time, since the matching source “4124” of the matching is different from the matching source “4105” at the end of the applied stack, all the recorded contents of the search condition network evaluation table 800 are cleared, and the matching “4105” is applied to the applied stack. −3203 ”is added (FIG. 37 (b)).

  Next, processing S1103 to S1110 is repeated. First, the search condition network evaluation table 820 (FIG. 35 (c)) is recorded. At this time, the instance with the smallest number of candidates is obtained, and “4102” is obtained as the instance with the largest degree of influence, and the matching destination obtained by executing the process 1200 (FIG. 49) for this is selected. Push to the stack waiting for application. At this time, the application waiting stack becomes the application waiting stack 920 (FIG. 36C). Then, since the application waiting stack is not empty, it is popped to obtain the matching “4102-3202”. At this time, since the matching source “4102” of the matching is different from the matching source “4124” at the end of the applied stack, all the recorded contents of the search condition network evaluation table 820 are cleared, and the matching “4102” is applied to the applied stack. −3202 ”is added (FIG. 37 (c)).

  Hereinafter, in the same manner, when the applied stack is in the state shown in FIG. 37 (d), it is determined that there are matching destinations for all concept instances / relationship instances included in the search condition network 410, and the determination result is “ Record "true" and exit.

  Next, an applicable matching acquisition flow will be described along the flow of FIG. 49 starting from processing S1200. 49A and 49B are partially enlarged views of FIG. Here, the index network 320 (FIG. 13) is targeted, the applied stack is empty, the concept instance “4102” (board) of the instance of the search condition network 410 is designated, and the concept instance is set as the matching source. A process for acquiring matching to be performed will be described. First, the search condition determination unit 93 acquires the elements of the index network one by one, and determines whether or not the designated concept instance “4102” is an instance of the concept “plate”.

  Here, whether an instance a is an instance of an object (concept or relationship) B is determined by whether or not the object B is positioned in a superordinate concept (superordinate relationship) of the object A referred to by the instance a. At this time, the object A referred to by the instance a is acquired and compared with the object B, and if it does not match, the object A is further acquired and compared with the object B. It can be judged by repeating until the superordinate concept of is gone.

  At this time, for example, the concept instance “3201” (refer to the gearless elevator) and the related instance “3222” (refer to the material) included in the index network 320 are not determined as instances of the concept “plate”, but the concept instance “3202”. Only “3206” (refer to the plate respectively) corresponds to the instance of the concept “plate”. At this time, since the matching under application is not registered, a matching list “{“ 4102-3202 ”“ 4102-3206 ”}” for each instance is obtained.

  When the index network 320 is targeted and the applied stack is in the state shown in FIG. 37 (b), the concept instance “4102” (board) of the search condition network instance is designated, and the concept instance is set as the matching source. A process for acquiring matching to be performed will be described. First, the search condition determination unit 93 acquires the elements of the index network one by one, and acquires the instance of the concept “board” referred to by the designated concept instance “4102”. In the present embodiment, concept instances “3202” and “3206” are acquired. Next, a relationship instance having the concept instances “3202” and “3206” as the relationship source concept instance or the relationship destination concept instance is acquired. In this case, the relationship instances “3221”, “3222”, “3223”, and “3224” are obtained from the concept instance “3202”, and the relationship instances “3220”, “3225”, and “3227” are acquired from the concept instance “3206”. . Next, from the matching registered in the applied stack, a matching with any of the acquired relation instances as a matching destination is acquired. In this embodiment, “4124-3222” is obtained. Therefore, the matching “4124-3222” and the concept instance “4102” are specified, and the process S1300 (FIG. 50 described later) is executed to obtain whether “4102-3202” is applicable to the matching. In this case, since the execution result of the process S1300 (FIG. 50) is “true”, “4102-3202” is recorded at the end of the applicable matching list. (The method for determining whether to apply matching will be described in detail later)

  Next, a flow for determining whether or not to apply matching will be described along the flow of FIG. 50 starting from processing S1300. 50A, 50B, 50C, and 50D are partially enlarged views of FIG. Here, the result of specifying and executing the matching “4124-3222” and the concept instance “4102” will be described.

(Processing S1301 to S1305)
First, since the designated concept instance “4102” is a concept instance, it is checked whether or not it is a relation source concept of the matching source instance “4124” of the designated matching. From the search condition network 410, the determination result is true.

(Processing S1306)
Next, it is checked whether the concept instance “4102” is an instance of the concept “board” referenced by the concept instance “3202” that is the relation source concept instance of the relationship instance “3222”, or a higher-level concept instance. To do. Since the concept referred to by the concept instance “4102” is “board”, the determination result is true.

(Processing S1311 to S1312)
The determination result is true and the process is terminated.

<Effect in embodiment>
As described above, when an information search is performed with the content entered in the search condition input part of the search keyword input screen 1 in FIG. 18 (a), the search result display screen example 1 in FIG. 25 is displayed as the search result. Finish the explanation. By the way, when this embodiment having the same embodiment is executed with respect to another input, when information search is performed with the content input in the search condition input section of FIG. 18 (b) search keyword input screen example 2 When search result display screen example 2 in FIG. 26 is displayed as a search result, and information search is performed using the contents entered in the search condition input part of FIG. 18 (c) search keyword input screen example 3, FIG. A search result no display screen is displayed as a search result.

  Here, in order to clarify the effect of the present embodiment, the determination result obtained by giving the same input as in the present embodiment to the information search apparatus using the conventional method is compared with the result of the present embodiment. First, by using an ontology, it is possible to process “if a lower keyword is included, treat it as if a higher keyword is included”, but search condition determination is a conventional keyword that determines whether or not an individual keyword exists. A matching information retrieval device is assumed. Here, the keyword enumerated as “Elevator side plate Mirror surface processed HL processed material SUS” was input to this device with the intention of “Elevator with side plate that is made of SUS material and subjected to HL processing and mirror surface processing”. Consider the case of performing an AND type search. This result is processed assuming that the search target data of the search target ID 2200 does not satisfy the search condition (because the keyword “material” is not assigned). On the other hand, with the same intention, if the keyword “Elevator side plate Mirror surface processing HL processing material SUS” is enumerated and input, the search target data with the search target ID “2200” has the search condition according to this embodiment. Output as satisfying. Here, when interpreted in the range of data and ontology targeted by the present embodiment, `` SUS '' given to the search target data of the search target ID `` 2200 '' plays the role of the material in the side plate, It can be said that it is more appropriate that the search target data with the search target ID “2200” is treated as satisfying the search condition. For this reason, according to the information search apparatus of the present embodiment, in the conventional method, data that has been determined not to satisfy the search condition because “the keyword specified in the search condition is not included in a certain index” is used. On the other hand, it is possible to determine that the search condition is satisfied because “the concept instance / relationship instance of the search condition network corresponds to a certain concept instance / relational instance of the index network”, and the search reproduction rate is improved. It can be said that an effect is obtained.

  Also, consider the case where an AND-type search is executed by inputting the keyword listed as “elevator side plate mirror surface processing HL processing SUS” with the same intention to the above conventional keyword matching information search device. As a result, the search target data of the search target ID “2100” is processed as satisfying the search condition (if the keyword “elevator” is included by using ontology 100, “gearless elevator” or “ It ’s assumed that the keyword “gear elevator” is included). On the other hand, according to the present embodiment, even when an input of “elevator side plate mirror finish HL machining SUS” is accepted, a search condition network similar to that when an input of “elevator side plate mirror finish HL machining material SUS” is accepted is formed. Thus, the search target data with the search target ID “2100” is processed as not satisfying the condition. Here, when interpreted in the range of the data and ontology targeted by this embodiment, the search target of the search target ID “2100” and the “mirror finish” in the index network 310 are not applied to the side plate, but the door plate Therefore, it can be said that it is appropriate that the search target data with the search target ID “2100” be handled as not satisfying the search condition. From this, according to the information search apparatus of the present embodiment, for the data that was determined to satisfy the search condition because “the keyword specified in the search condition is included in a certain index” in the conventional method, By making it possible to determine that the search condition network concept instance / relationship instance does not correspond to the index network concept instance / relationship instance, the keyword specified as the search condition can be determined. Prevents the decline in search precision and search recall due to determining only whether or not it is included in the keyword assigned to the search target data, and enables searches with higher recall and precision than conventional methods .

  The above-described information retrieval apparatus shown in FIG. 1 can also be realized by using, for example, a general-purpose computer apparatus as basic hardware. That is, each element provided in the apparatus of FIG. 1 may be realized by causing a computer to execute a program describing an instruction to perform processing of each element. At this time, the information classification hierarchy management device may be realized by installing the above program in a computer device in advance, or may be stored in a storage medium such as a hard disk, a memory device, or an optical disk, or via a network. This program may be distributed and this program may be installed in a computer device as appropriate. In addition, the ontology DB, index keyword DB, and search target DB are stored in a memory device such as a hard disk or a CD-R, a CD-RW, a DVD-RAM, a DVD-R, etc. It can be realized by appropriately using.

Claims (4)

A first ontology having a plurality of nodes representing a plurality of concepts and connected in a hierarchy, and a plurality of directional links that connect the two nodes and represent a relationship between the concepts corresponding to the two nodes; An ontology database representing a second ontology having a plurality of nodes representing a plurality of the relationships and connected in a hierarchical manner;
A search target database for storing a plurality of search target data; and
An index keyword database for storing a list of index keywords for each search target data;
A search condition receiving unit for receiving a list of search keywords specified by the user as a search condition;
A plurality of keywords are received, a node or link representing a concept or relationship corresponding to the received keyword is identified in the first ontology, and a path connecting the identified nodes or links is traced to the nodes and links in the first ontology. And a semantic network generation unit that generates a semantic network in which the identified node or link is connected by a route found by the search;
A search condition network generation unit that gives the search keyword received by the search condition reception unit as the keyword to the semantic network generation unit, and acquires the semantic network generated by the semantic network generation unit as a search condition network;
An index network generation unit that gives an index keyword included in the list for each search target data as the keyword to the semantic network generation unit, and acquires a semantic network generated by the semantic network generation unit as an index network;
An index network including the same structure as the search condition network, or the same structure obtained by replacing any node or link included in the search condition network with a node or link representing a subordinate concept or subordinate relationship, A search condition determination unit for detecting from the index network generated by the network generation unit;
A search result output unit that acquires search target data corresponding to the index network detected by the search condition determination unit from the search target database, and outputs the acquired search target data;
An information retrieval apparatus comprising: The search condition accepting unit accepts a search keyword order specification,
The search condition network generation unit notifies the meaning network generation unit of the ranking of search keywords,
The semantic network generation unit searches a node or link corresponding to the n-th ranking keyword in the reverse direction of the link with direction, and selects any one of the first to n−1 ranking keywords. 2. The information according to claim 1, wherein a route to a corresponding node or link is found, and this process is performed while sequentially increasing the value of n from 1 to the number of keywords received from the search condition network generation unit. Search device. A ranking is specified for the index keywords in the list in the index keyword database,
The index network generation unit notifies the semantic network generation unit of the index keyword ranking;
The semantic network generation unit searches a node or link corresponding to the n-th ranking keyword in the reverse direction of the link with direction, and selects any one of the first to n−1 ranking keywords. 3. The route to the corresponding node or link is found, and this process is performed while sequentially increasing the value of n from 1 to the number of keywords notified from the index network generation unit. Information retrieval device. The ontology comprises a plurality of partial ontologies;
The semantic network generation unit
A search priority is set for each of the plurality of partial ontologies using the number of nodes or links corresponding to the 1st to n−1th ranking keywords respectively belonging to the plurality of partial ontologies,
When there are a plurality of links that can be searched in the search, a link to be prioritized for the search is identified from the plurality of links based on the search priority of the partial ontology to which the connection source node of the plurality of links belongs. 4. The information search device according to claim 2, wherein the search is performed preferentially from the specified link.

Images