Search apparatus, search system, non-transitory computer-readable medium, and search method

Abstract

Search apparatus, search system, non-transitory computer-readable medium, and search method. A search apparatus includes a receiver and a determining unit. The receiver receives search terms for a search. The determining unit determines recommended terms to be used for refining searches, each refining search using a corresponding recommended term and a search term from the recommended terms. Recommended terms are determined based on i) the amount of overlap between the results of the refining searches and / or ii) the amount of difference between the number of results in one refining search and the number of results in another refining search.

Description

Technical Field

[0001] This disclosure relates to a search apparatus, a search system, a non-transitory computer-readable medium, and a search method. Background Technology

[0002] Japanese Unexamined Patent Application Publication No. 2012-003532 discloses a query suggestion providing apparatus that provides query suggestions by using a search log stored to correlate terms used as query inputs. Specifically, the apparatus references a search log containing a series of search operations, including a search query and a re-search query, and calculates a score indicating the degree of correlation between the search queries included in the series of search operations. The score between the last query in the series of search operations and all other search queries is given a high weight in the score calculation. When a search query is received from a user terminal, the apparatus provides the user terminal with search queries that have a high score for the received search query.

[0003] Japanese Unexamined Patent Application Publication No. 2008-234559 discloses an information search device for refining document searches. Specifically, it performs morphological analysis on sentences included in documents, extracts terms and associates them with corresponding documents, creates an initial reverse index, generates a term list in which documents containing each extracted term are associated, and displays the term list on a user terminal. Then, the user is encouraged to select a term from the term list, a reconfigured reverse index is created based on a subset of documents containing the selected term, and the term list is regenerated and re-displayed on the user terminal using the reconfigured reverse index. Summary of the Invention

[0004] When inputting additional terms to further narrow down the search for content fragments retrieved using a particular term, a method for recommending terms to be entered can be used as follows: a list of recommended terms is created and recommended using an inverted index (where terms and documents serving as examples of content fragments are related to each other). However, with this recommendation method using an inverted index and a list of recommended terms, when multiple terms are recommended for refining the search, the content fragments retrieved using these terms may overlap, or the number of retrieved content fragments may vary depending on the terms used in the search. Therefore, it is necessary to input recommended terms one by one to find the necessary information from the search results.

[0005] Therefore, the object of this disclosure is to provide a search apparatus, a search system, a non-transitory computer-readable medium, and a search method that, when additional terms are input to further narrow down the scope of content retrieved using a certain term, can obtain search results that include more information needed by the user, compared to a recommendation method that creates a list of recommended terms using inverted indexes.

[0006] According to a first aspect of this disclosure, a search apparatus is provided, comprising a receiver and a determining unit. The receiver receives search terms for searching. The determining unit determines recommended terms to be used for refining searches, each of the refining searches using a corresponding one of the recommended terms and the search term. The recommended terms are determined based on (i) the amount of overlap between the results of the refining searches and / or (ii) the amount of difference between the number of results in one refining search and the number of results in another refining search.

[0007] According to a second aspect of this disclosure, in the search apparatus according to the first aspect, the determining unit obtains a relationship between terms by using a document list obtained by extracting documents including search terms received by the receiver from pre-stored documents and a pre-stored correspondence between the documents and the pre-stored terms, and determines recommended terms based on the obtained relationship between the terms.

[0008] According to a third aspect of this disclosure, in the search apparatus according to the second aspect, the determining unit obtains the interaction information between the probability of a term to be recommended being selected and the probability of refining the search type as a relationship between terms, and determines recommended terms such that the interaction information is minimized or equal to or less than a predetermined threshold.

[0009] According to a fourth aspect of this disclosure, the search apparatus according to the second or third aspect further includes a limiter that limits the number of terms to be used to obtain the relationship between terms by using search terms received by the receiver.

[0010] According to a fifth aspect of this disclosure, the search apparatus according to any one of the second to fourth aspects further includes a limiter that sets a limit on the necessary number of documents in the document list, wherein the determining unit obtains a relationship between terms by using a correspondence between a document list obtained by extracting documents including search terms received by the receiver from documents that match the limit set by the limiter and pre-stored terms, and determines recommended terms.

[0011] According to a sixth aspect of this disclosure, in the search apparatus according to the fifth aspect, the limiter determines the necessary number of documents by using the number of documents and the predetermined number of recommended terms.

[0012] According to the seventh aspect of this disclosure, in the search apparatus according to any one of the first to sixth aspects, the determining unit determines the recommended terms by using the Jaccard coefficient, the Dice coefficient, or the Simpson coefficient.

[0013] According to the eighth aspect of this disclosure, in a search apparatus according to any one of the first to sixth aspects, the determining unit determines the recommended terms by using the difference in the number of documents obtained by adding the terms to be recommended to the query.

[0014] According to the ninth aspect of this disclosure, in the search apparatus according to the third aspect, the determining unit determines recommended terms by using virtually determined virtual terms, the search results using virtual terms include a predetermined ideal number of documents, and the search results do not overlap with search results using other terms.

[0015] According to the tenth aspect of this disclosure, the search apparatus according to any one of the first to ninth aspects further includes a display unit that displays the number of documents used for the query before the refinement search, the number of documents when any recommended terms are added to the query, the overlap when any recommended terms are added to the query, the imbalance when any recommended terms are added to the query, and the loss when any recommended terms are added to the query as corresponding areas.

[0016] According to the eleventh aspect of this disclosure, the search apparatus according to the tenth aspect further includes an adding unit that adds to the query an entry corresponding to the region selected from the region.

[0017] According to the twelfth aspect of this disclosure, a search system is provided, comprising: a search device according to any one of the first to eleventh aspects; and an information processing terminal to which inputs terms to be received by the receiver are displayed, and which displays the determination results obtained by the determining unit.

[0018] According to a thirteenth aspect of this disclosure, a non-transitory computer-readable medium is provided, storing a program that enables a computer to perform a process for searching, the process comprising: receiving search terms for searching; and determining recommended terms to be used for refining searches, each of the refining searches using a corresponding one of the recommended terms and the search term. The recommended terms are determined based on (i) the amount of overlap between the results of the refining searches and / or (ii) the amount of difference between the number of results in one refining search and the number of results in the other refining search.

[0019] According to a fourteenth aspect of this disclosure, a search method is provided, comprising: receiving search terms for searching; and determining recommended terms to be used for refining searches, each of the refining searches using a corresponding one of the recommended terms and the search term. The recommended terms are determined based on (i) the amount of overlap between the results of the refining searches and / or (ii) the amount of difference between the number of results in one refining search and the number of results in the other refining search.

[0020] According to a first aspect of this disclosure, a search apparatus can be provided that, when additional terms are entered to further narrow down the scope of content retrieved using a certain term, can obtain search results that include more information needed by the user, compared to a recommendation method that uses a reverse index to create a list of recommended terms.

[0021] According to the second aspect of this disclosure, recommended terms can be determined by considering the relationships between terms.

[0022] According to a third aspect of this disclosure, recommended terms can be determined based on (i) the amount of overlap between the results of a refinement search and / or (ii) the amount of difference between the number of results in one refinement search and the number of results in another refinement search.

[0023] According to the fourth aspect of this disclosure, the computational load can be reduced compared to determining recommended terms by using all pre-stored terms.

[0024] According to the fifth aspect of this disclosure, the computational load can be reduced compared to determining recommended terms by using a document list that includes all documents.

[0025] According to the sixth aspect of this disclosure, the number of necessary documents for determining recommended terms can be determined.

[0026] According to the seventh aspect of this disclosure, recommended terms can be determined.

[0027] According to the eighth aspect of this disclosure, recommended terms can be determined.

[0028] According to the ninth aspect of this disclosure, compared to determining recommended terms by obtaining interactive information without using virtual terms, the probability of no document matching in the search can be reduced even when any of the recommended terms are added to the query.

[0029] According to the tenth aspect of this disclosure, the relationships between terms can be visually examined.

[0030] According to the eleventh aspect of this disclosure, terms can be added to a query while examining the relationships between terms.

[0031] According to the twelfth aspect of this disclosure, a search system can be provided that, when additional terms are entered to further narrow down the scope of content retrieved using a certain term, can obtain search results that include more information needed by the user, compared to a recommendation method that uses an inverted index to create a list of recommended terms.

[0032] According to the thirteenth aspect of this disclosure, a non-transitory computer-readable medium can be provided, which, when additional terms are entered to further narrow down the scope of content retrieved using a certain term, can provide search results that include more information needed by the user, compared to a recommendation method that uses a reverse index to create a list of recommended terms.

[0033] According to the fourteenth aspect of this disclosure, a search method can be provided that, when additional terms are entered to further narrow down the search by using a certain term, can obtain search results that include more information needed by the user, compared with a recommendation method that uses an inverted index to create a list of recommended terms. Attached Figure Description

[0034] Exemplary embodiments of this disclosure will be described in detail with reference to the following figures, wherein:

[0035] Figure 1 This is a diagram schematically illustrating the configuration of an information processing system according to an exemplary embodiment;

[0036] Figure 2 This is a block diagram illustrating the electrical configuration of an information processing terminal and a server in an information processing system according to an exemplary embodiment;

[0037] Figure 3 This is a functional block diagram of a server according to a first exemplary embodiment;

[0038] Figure 4 This is a diagram showing example recommended terms when "cooking" is added to the query;

[0039] Figure 5 It is a diagram that schematically illustrates the relationship between interactive information and "overlap" and indicates that the interactive information decreases as the "overlap" decreases;

[0040] Figure 6 It is a diagram that schematically illustrates the relationship between interactive information and "imbalance" and indicates that interactive information decreases as "imbalance" decreases;

[0041] Figure 7This is a flowchart illustrating an example process performed in a server according to a first exemplary embodiment;

[0042] Figure 8 This is a functional block diagram of a server according to a second exemplary embodiment;

[0043] Figure 9 This is a flowchart illustrating an example process performed in a server according to a second exemplary embodiment;

[0044] Figure 10 This is a functional block diagram of a server according to a third exemplary embodiment;

[0045] Figure 11 This is a diagram showing an example of a correspondence table.

[0046] Figure 12 The interactive information, calculated as a fraction using expressions based on a correspondence table, is shown.

[0047] Figure 13 The scores for the recommended word list, calculated based on interaction information, are shown.

[0048] Figure 14 This is a flowchart illustrating an example process performed in a server according to a third exemplary embodiment;

[0049] Figure 15 This shows the candidate recommended terms when "cooking" is added to the query, as well as the number of documents matched when each term is added to the query;

[0050] Figure 16 This is an image showing an example where "virtual terms" are set up;

[0051] Figure 17 This is a diagram illustrating an example graphical user interface (GUI) for handling "overlap," "imbalance," and "missing" elements;

[0052] Figure 18 This is a diagram used to explain an example where terms are added to a query using a GUI that handles "overlap," "imbalance," and "missing" terms; and

[0053] Figure 19 This is a diagram illustrating an example GUI using a true / false table for terms and documents. Detailed Implementation

[0054] In the following, exemplary embodiments will be described in detail with reference to the accompanying drawings. In the exemplary embodiments, an information processing system in which multiple information processing terminals and servers are connected to each other via communication lines that are any type of network will be described as an example of a search system. Figure 1This is a diagram schematically illustrating the configuration of an information processing system 10 according to an exemplary embodiment.

[0055] like Figure 1 As shown, the information processing system 10 according to an exemplary embodiment includes a plurality of information processing terminals 14a, 14b, ... and a server 16, wherein the server 16 serves as a search device. Where the information processing terminals 14a, 14b, ... need not be distinguished from each other in the description, the letter symbols at the end of the reference numerals may be omitted. In the exemplary embodiment, an example including a plurality of information processing terminals 14a, 14b, ... will be described; however, the number of information processing terminals 14 may be one.

[0056] Information processing terminal 14 and server 16 are connected to each other via communication line 12, which can be a local area network (LAN), wide area network (WAN), Internet, or intranet. Information processing terminal 14 and server 16 can send and receive various types of data to each other via communication line 12.

[0057] In the information processing system 10 according to an exemplary embodiment, server 16 provides a document management service as a cloud service for managing documents. The document management service allows, for example, various documents representing information to be stored on server 16, and allows management target documents stored on server 16 to be browsed when information processing terminal 14 accesses server 16.

[0058] The electrical configurations of the information processing terminal 14 and the server 16 according to an exemplary embodiment will now be described. Figure 2 This is a block diagram illustrating the electrical configuration of an information processing terminal 14 and a server 16 in an information processing system 10 according to an exemplary embodiment. The information processing terminal 14 and the server 16 essentially have a general-purpose computer configuration, and therefore, the information processing terminal 14 is described as a representative example.

[0059] like Figure 2As shown, the information processing terminal 14 according to an exemplary embodiment includes a central processing unit (CPU) 14A, a read-only memory (ROM) 14B, a random access memory (RAM) 14C, a hard disk drive (HDD) 14D, a keyboard 14E, a display 14F, and a communication line interface (IF) unit 14G. The CPU 14A controls the overall operation of the information processing terminal 14. The ROM 14B pre-stores various control programs and various parameters, etc. When the CPU 14A executes various programs, the RAM 14C is used as, for example, a workspace. The HDD 14D stores various types of data and application programs, etc. The keyboard 14E is used to input various types of information. The display 14F is used to display various types of information. The communication line interface (IF) unit 14G is connected to a communication line 12 to send various types of data to or receive various types of data from other devices connected to the communication line 12. The above-mentioned components of the information processing terminal 14 are electrically connected to each other via a system bus 14H. In the information processing terminal 14 according to the exemplary embodiment, HDD 14D is used as a storage unit; however, the storage unit is not limited to this, and another non-volatile storage unit, such as flash memory, can be used.

[0060] With the above configuration, in the information processing terminal 14 according to the exemplary embodiment, the CPU 14A performs control to access the ROM 14B, RAM 14C, and HDD 14D, obtains various types of data via the keyboard 14E, and displays various types of information on the display 14F. In the information processing terminal 14, the CPU 14A performs control to send and receive communication data via the communication line IF unit 14G.

[0061] In the information processing system 10 configured according to an exemplary embodiment, as described above, the server 16 provides a document management service as a cloud service for managing documents. For example, when information stored on the information processing terminal 14 is transmitted to the server 16 as a management target document, the document is managed by the server 16, and access to the document stored on the server 16 is allowed by operating the information processing terminal 14.

[0062] First Exemplary Implementation

[0063] The functional configuration of server 16 according to the first exemplary embodiment will now be described. Figure 3 This is a functional block diagram of server 16 according to a first exemplary embodiment.

[0064] In this exemplary embodiment, a function is provided in which, when searching for document information stored by the document management service provided by the server 16 from the information processing terminal 14, the server 16 recommends a list of terms corresponding to the terms entered from the information processing terminal 14 to provide search assistance. That is, when characters or strings are entered as a query from the information processing terminal 14, the server 16 recommends a list of terms corresponding to the characters or strings being entered to the information processing terminal 14. For example, such as... Figure 4 As shown, when "cooking" is added to the query, the following candidate recommended terms are recommended: "Japanese," "Italian," "French," "Chinese," "delicious," and "simple." In the following description, the term entered as a query to search for documents is called a search term, and the term related to the search term entered as a query is called a recommended term.

[0065] like Figure 3 As shown, server 16 has the functions of document database (DB) 22, term database 24, query receiver 18 used as receiver, searcher 20, score calculator 26, recommended term list calculator 28 used as determination unit, and term selector 30.

[0066] Document DB 22 stores document information that has been pre-registered on server 16 and allows registration and browsing of documents from information processing terminal 14.

[0067] When a document is registered in document DB 22, terms extracted from documents associated with that document are registered in term DB 24.

[0068] When a user operates the information processing terminal 14 to input terms for searching documents, the query receiver 18 obtains and receives the input terms as search terms from the information processing terminal 14. The query receiver 18 refers to the term database 24, searches for the received terms, and outputs the search results to the score calculator 26.

[0069] Searcher 20 references the terms received by query receiver 18, creates a list of target documents matching the search criteria, and outputs it to score calculator 26. In other words, searcher 20 searches document DB 22 for documents containing the terms received by query receiver 18 and outputs a list of the retrieved documents to score calculator 26.

[0070] The score calculator 26 calculates the score between the respective indicative terms by using the correspondence between document DB 22 and term DB 24.

[0071] The recommended term list calculator 28 calculates the optimal number of terms with the lowest scores calculated by the score calculator 26 as the recommended term list. In this exemplary embodiment, when outputting multiple recommended terms to narrow the search results obtained by using the search terms received by the query receiver 18, the recommended term list calculator 28 determines the optimal number of recommended terms, where at least one of "overlap" and "imbalance" is less than when narrowing the scope using combinations of other terms for the optimal number of recommended terms.

[0072] The term selector 30 adds the terms selected by the user from the recommended term list calculated by the recommended term list calculator 28 as search terms to the query.

[0073] Now, the score calculation of the score calculator 26 and the recommended term list calculation of the recommended term list calculator 28 will be described in detail.

[0074] In this exemplary embodiment, a refined search is performed using a list of recommended terms instead of individual terms. That is, the relationships between the recommended terms are considered. In this exemplary embodiment, the search results are scored for "overlap," "imbalance," and "loss" when the recommended term list is added to the query. Note that "overlap" refers to the overlap of the refined search results when terms are added to the query; "imbalance" refers to the difference in the number of documents in each narrowed scope when terms are added to the query; and "loss" refers to the absence of a matching document in the search even when any term from the recommended term list is added to the query.

[0075] As a method for scoring "overlap," there are methods that use similarity scores between sets (e.g., Jaccard coefficient, Dice coefficient, or Simpson coefficient). Specifically, when using r... i This indicates that an entry w is being added. i The set of documents matched in the search, and denoted by r j This indicates that an entry w is being added. i When the set of documents matched in the search is considered, the Jaccard coefficient J is... ij It is represented by the following expression (1).

[0076]

[0077] In other words, the recommended term list calculator 28 needs to select the term list with the smallest sum of Jaccard coefficients J among the recommended term lists. The sum of Jaccard coefficients J is represented by the following expression (2).

[0078]

[0079] As a method for scoring "imbalance," one approach is to use the difference in the number of documents obtained by adding recommended terms to the query. Specifically, when using r... i This indicates that an entry w is being added. i The number of documents matched in the search, and expressed in r j This indicates that an entry w is being added. j The "imbalance" score D is calculated based on the number of documents matched in the search. ij The difference is represented by the following expression (3) using the method of using the difference.

[0080]

[0081] In other words, the recommended term list calculator 28 needs to select the term list with the smallest sum of the absolute values ​​of the differences between the recommended term lists. The sum D is represented by the following expression (4).

[0082]

[0083] As a method for simultaneously scoring "overlap" and "imbalance," there exists a method that uses the interaction information between the probability of a term being selected from the recommended term list and the probability of refining the search type (AND search or NOT search). When using r... i This indicates adding a term (w). i The number of documents matched in the search, denoted by r. j This indicates adding a term (w). j The number of documents matched in the search, and expressed in r ij Indicates r i and r j When considering the union of sets r, the values ​​from the union r are calculated as follows: ij The probability p(r) of selecting a certain document ij The entropy H(p(r)) ij The probability p(r) is based on the probability p(t) of the refined search type. ij The entropy H(p(r)) ij The interaction information I is obtained by finding the difference between |t)). ij .

[0084] I ij =H(p(r) ij ))-H(p(r ij |t))…(5)

[0085]

[0086] Figure 5 and Figure 6 The relationships between interactive information and "overlap" and between interactive information and "imbalance" are illustrated schematically. When "overlap" decreases, interactive information decreases. When "imbalance" decreases, interactive information decreases. Interactive Information I ij Corresponding to "use term w" i "Refined search" and "using terms w" j The "overlap" and "imbalance" between the "refined search" are also considered. That is, the recommended term list calculator 28 needs to select the term list with the smallest sum I of the interaction information of the recommended term lists. The sum I of the interaction information is represented by the following expression (7).

[0087]

[0088] The specific processing performed in server 16 according to this exemplary embodiment will now be described. Figure 7 This is a flowchart illustrating an example process performed in server 16 according to this exemplary embodiment. When a user operates information processing terminal 14 and enters a term as a query, the process begins... Figure 7 The processing in the middle.

[0089] In step 100, the query receiver 18 receives the term entered as a query from the information processing terminal 14, and the process proceeds to step 102.

[0090] In step 102, the receiver 18 queries the reference term DB 24 and searches for the received term, and the process proceeds to step 104.

[0091] In step 104, the searcher 20 searches the document DB 22 for documents that include the terms received by the query receiver 18, and the process proceeds to step 106.

[0092] In step 106, the score calculator 26 calculates the score representing the relationship between the terms by using the correspondence between document DB 22 and term DB 24, and the process proceeds to step 108. To calculate the score, as described above, a method for scoring "overlap" can be used, a method for scoring "imbalance" can be used, or a method for scoring both "overlap" and "imbalance" can be used simultaneously.

[0093] In step 108, the recommended term list calculator 28 calculates the optimal number of terms with the lowest scores calculated by the score calculator 26 as the recommended term list, presents the recommended term list to the user, and the process proceeds to step 110.

[0094] In step 110, the term selector 30 determines whether an instruction is given to add a term selected by the user from the list of recommended terms calculated by the recommended term list calculator 28 as a search term to the query. If the result is affirmative, the term selector 30 adds the specified term to the query, the process returns to step 100, and the above processing is repeated. If the result is negative, the process proceeds to step 112.

[0095] In step 112, the term selector 30 determines whether to issue an instruction to search for documents without selecting any terms. If the result is affirmative, the process proceeds to step 114. On the other hand, if the term used as query input is reset and another term is entered as the query, or if an instruction for another processing is given, the result is determined to be negative, and the processing ends.

[0096] In step 114, CPU 16A searches for documents in document DB 22 that include the terms used as query input, presents the documents on information processing terminal 14, and the processing ends.

[0097] Second Exemplary Implementation

[0098] The functional configuration of server 16 according to the second exemplary embodiment will now be described. Figure 8 This is a functional block diagram of server 16 according to this exemplary embodiment. Note that configurations identical to those in the above exemplary embodiment are given the same reference numerals, and their detailed descriptions will be omitted.

[0099] In the above exemplary embodiment, a computational load problem arises when the score calculator 26 calculates scores. For example, if the number of terms registered in the term database 24 is W, and N terms are selected from W terms as a recommended term list, the number of term combinations is... W C N Furthermore, with a large number of entries, it may be impossible to perform calculations within actual time.

[0100] In this exemplary embodiment, such as Figure 8 As shown, a candidate recommended term calculator 32 is also provided as a limiter to limit the number of candidate recommended terms in term DB 24 to be used for score calculation based on the input query.

[0101] As techniques for limiting candidate recommended terms (e.g., nearby terms in the word embedding space), one can use word2vec (Mikolov, T., Chen, K., Corrado, G., and Dean, J. (2013), Efficient estimation of word representations in vector space, arXiv preprint arXiv: 1301.3781.) or fasttext (Joulin, A., Grave, E., Bojanowski, P., and Mikolov, T. (2016), Bag of tricks for efficient text classification, arXiv preprint arXiv: 1607.01759). Alternatively, nearby terms on a knowledge graph (ontology) can be used.

[0102] Figure 9 This is a flowchart illustrating an example flow of processing performed in server 16 according to this exemplary embodiment. Note that, with Figure 7 The same steps in the accompanying drawings are given the same reference numerals, and their detailed descriptions will be omitted.

[0103] like Figure 9 As shown, add step 103. Figure 7 In step 103, the candidate recommendation term calculator 32 calculates candidate recommendation terms. Therefore, scores are calculated for the defined terms, thereby reducing the computational load and quickly recommending a list of terms to the user.

[0104] Third Exemplary Implementation

[0105] The functional configuration of server 16 according to a third exemplary embodiment will now be described. Figure 10 This is a functional block diagram of server 16 according to this exemplary embodiment. Note that configurations identical to those in the above exemplary embodiment are given the same reference numerals, and their detailed descriptions will be omitted.

[0106] In addition to the functions in the second exemplary embodiment, the server 16 according to this exemplary embodiment also has the functions of a search results display 34 and a table creator 36.

[0107] The search results display 34 performs the process of displaying the results of the searcher 20 searching for document DB 22 on the information processing terminal 14 operated by the user.

[0108] Table creator 36 creates a mapping table that indicates the correspondence between candidate recommended terms calculated by candidate recommended term calculator 32 and documents retrieved by searcher 20. Figure 11 This is a diagram showing an example of a correspondence table.

[0109] exist Figure 11 In the example shown, for simplicity, the registered term W in term DB 24 and the registered document D in document DB 22 are defined as follows.

[0110] W={w1,w2,w3,w4,w5},D={d1,d2,d3,d4,d5}...(8)

[0111] Based on the defined registered terms and target documents, the score calculator 26 calculates the score, and the recommended term list calculator 28 calculates the recommended term list.

[0112] exist Figure 11 In the search query, "T" indicates that word w corresponds to document d, meaning document d matches in the search, while "F" indicates that word w does not correspond to document d, meaning document d does not match in the search.

[0113] Figure 12 It is shown that by using the following expression (9) based on Figure 11 The corresponding table shown calculates the interaction information as the score.

[0114]

[0115] Here, Δr represents a very small quantity and is used to prevent [the value of Δr] from being too small. j It is r i subsets of r ij -r i The case where Δr = 0 and cannot be calculated when it is about to be divided by 0. Here, assume Δr = 1.0 × 10 -5 And calculations are performed. The interactive information is asymmetric, so the same combination will have different scores when the references are different (e.g., the scores of w1 and w3 are different from the scores of w3 and w1).

[0116] Figure 13 It shows the basis Figure 12 The results shown are used to calculate the recommended term list (in Figure 13 In the examples shown, each list includes the scores of two terms. Figure 13 In the example shown, w1 and w2 have the lowest scores, and this list of recommended terms is selected. (See reference...) Figure 11Among the collocations w1 and w2, the list shows smaller overlap and imbalance for "T". For example, collocations w4 and w5 are imbalanced but not overlapped, and have an interaction information of 0.80, which is larger than other collocations. Collocations w2 and w3 are overlapped but not imbalanced, and have greater interaction information than collocations w1 and w2. These results also show that interaction information was used to score both overlap and imbalance.

[0117] In the example presented here, there are five registered terms and ten combinations of two selected terms (5C2 = 10); however, as the number of registered terms and selected terms increases, the number of combinations when creating the recommended term list also increases. Therefore, filtering is needed to, for example, limit the number of registered terms that will be used in the list calculation.

[0118] Figure 14 This is a flowchart illustrating an example flow of processing performed in server 16 according to this exemplary embodiment. Note that, with Figure 9 The same steps in the accompanying drawings are given the same reference numerals, and their detailed descriptions will be omitted.

[0119] In this exemplary embodiment, such as Figure 14 As shown, in step 104, the searcher 20 searches the document DB 22 for documents that include the terms received by the query receiver 18, and then the process proceeds to step 105A.

[0120] In step 105A, the search results display 34 performs the process of displaying the search results of the searcher 20 on the information processing terminal 14 operated by the user, and the process proceeds to step 105B.

[0121] In step 105B, table creator 36 creates a mapping table indicating the correspondence between candidate recommended terms calculated by candidate recommended term calculator 32 and documents retrieved by searcher 20. The process then proceeds to step 106, where score calculator 26 uses the created mapping table to calculate scores relating the relationships between the respective indicated terms.

[0122] In the above exemplary embodiment, the score calculator 26 can perform calculations for "overlap," "imbalance," and "missing" terms separately. For example, "overlap" and "imbalance" can be quantified using interaction information, but "missing" cannot. Therefore, "missing" terms are first scored, and interaction information is calculated based on the data from this score, thus considering "overlap," "imbalance," and "missing" terms. Therefore, "overlap," "imbalance," and "missing" terms are represented by various calculations, and the scoring method can be changed according to the objective. Furthermore, when a threshold is set in each calculation step, this can be used as a filter to reduce the computational load. Specifically, "missing" terms cannot be quantified using interaction information. To suppress "missing" terms, a lower limit (hereinafter referred to as "necessary document count") is set for the number of documents to be matched when adding terms to the query to limit the number of documents, and terms are filtered. Based on the recommended term count W... n Given the number of documents D, determine the required number of documents D. n The fraction calculator 26 selects terms from a table that meet certain conditions when calculating a fraction, and calculates interactive information to handle "missing" terms. In this case, the fraction calculator 26 acts as a restrictor.

[0123] The specific description provides an example of setting a lower limit on the number of documents for filtering purposes. Figure 15 This shows the candidate recommended terms when "cooking" is added to the query, and the number of documents matched when each term is added to the query. Assume that the number of matched documents R equals 200 when "cooking" is added to the query, and the number of recommended terms W. n Equals 5. Number of required documents D n Defined by the following expression (10). Assume the number of required documents is D. n It is the number of hits for each term that makes the "missing" value zero when the "overlap" is assumed to be zero.

[0124] D n =R / W n ...(10)

[0125] Given a document count R equals 200 and a recommended term count W... n When the number of required documents is equal to 5, the number of documents D is... n It equals 40. In Figure 15 In the example shown, "shorter time", "Egypt" and "superhot" are excluded from the calculation of the interaction information.

[0126] In the above exemplary embodiment, when the score calculator 26 scores using interactive information, "virtual terms" can be used to suppress "loss". "Loss" cannot be quantified using interactive information. Therefore, as... Figure 16As shown, terms that "include an ideal number of documents in the search results and whose search results do not overlap with those of other terms" are set as "virtual terms". Since "virtual terms" do not "overlap" with other terms, only "imbalance" is used to calculate interaction information. In other words, when using "virtual terms", "loss" is suppressed.

[0127] In the exemplary embodiment described above, server 16 can provide a GUI for handling "overlap," "imbalance," and "missing" terms. That is, it can explicitly display "overlap," "imbalance," and "missing" terms in cases where they are added from the recommended term list to the query. Specifically, when the recommended term list calculator 28 calculates the recommended term list in step 108 and presents it to the user, it can display, for example... Figure 17 The screen 50 shown is presented to the user as a GUI. In this case, the recommended glossary calculator 28 is used as the display unit.

[0128] exist Figure 17 In the context of the search, the number of documents used for the query ("cooking") before refining the search is represented by the outermost rectangular area, and the number of documents when terms from the recommended term list are added to the query is represented by the area containing those terms. Furthermore, "overlap" when terms from the recommended term list are added to the query is represented by the size of the overlapping portion of the corresponding areas; "imbalance" when terms from the recommended term list are added to the query is represented by the difference between the corresponding areas; and "missing" when terms from the recommended term list are added to the query is represented by the size of the portion of the area where no terms are present, or is explicitly indicated as a "missing" area.

[0129] When "overlap," "imbalance," and "missing" terms are explicitly displayed, users can directly visually inspect the relationships between terms, which aids in comprehension. Furthermore, it's easy to check the extent to which selecting terms narrows down the document, thus improving efficiency.

[0130] In the exemplary embodiments described above, a GUI for handling "overlap," "imbalance," and "missing" terms can be used to add terms to a query. For example, a recommended term list calculator 28 can provide a GUI that allows the user to access the information processing terminal 14 to perform queries. Figure 18 When performing an operation on a specified area on screen 52, terms corresponding to that specified area can be added to the query. For example, when a user performs an operation on... Figure 18When operating on the screen 52 showing an overlapping area, multiple terms that form an "overlapping" relationship are added to the query at once. When terms can be added using the GUI, allowing users to select queries while examining relationships between terms enables efficient, refined searching. In this case, the term list calculator 28 is recommended as the addition unit.

[0131] As a GUI, it's possible to apply a true / false table for terms and documents. Specifically, applications such as... Figure 19 The GUI shown uses a truth table, where the vertical axis represents terms and the horizontal axis represents documents. A term and document match is considered "true," and the cell is filled and represented as "white." A term and document do not match is considered "false," and the cell is not filled and represented as "black." Creating such a true / false table explicitly represents the correspondence between terms and documents.

[0132] Furthermore, when techniques such as the Infinite Relational Model (IRM) (Charles, K., Joshua, T., Thomas, G., Takeshi, Y., and Naonori, U. (2006), Learning System of Concepts with an Infinite Relational Model, AAAI) are employed, table clustering can be achieved, which helps in understanding the relationships between terms and between documents.

[0133] The processing performed in server 16 according to the exemplary embodiments described above can be performed using software, using hardware, or using a combination of software and hardware. The processing performed in server 16 can be stored as a program in a storage medium and is distributed.

[0134] The above description of exemplary embodiments of the present disclosure has been provided for purposes of illustration and description. It is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations will obviously be apparent to those skilled in the art. These embodiments were chosen and described in order to best explain the principles of the present disclosure and its practical application, thereby enabling others skilled in the art to understand the disclosure of various embodiments and various modifications suitable for the intended particular use. The scope of this disclosure is intended to be defined by the appended claims and their equivalents.

Claims

1. A search device, the search device comprising: A receiver that receives search terms for searching; as well as The determining unit determines the recommended terms to be used for refined searches, wherein each refined search uses a corresponding recommended term from the recommended terms and the search term, wherein... The recommended terms are determined by obtaining the interaction information between the probability of a term being selected and the probability of the refined search type as a relation between terms, based on i) the amount of overlap between the results of the refined search and / or ii) the amount of difference between the number of results in one of the refined searches and the number of results in the other.

2. The search device according to claim 1, wherein, The determining unit obtains the relationship between terms by using the correspondence between a document list obtained by extracting documents including the search terms received by the receiver from pre-stored documents and the pre-stored terms, and determines the recommended terms based on the obtained relationship between the terms.

3. The search device according to claim 1, wherein, The determining unit determines the recommended terms so that the interactive information is minimized or equal to or less than a predetermined threshold.

4. The search device according to claim 1, further comprising: A limiter that limits the number of terms to be used to obtain the relationship between the terms by using the search terms received by the receiver.

5. The search device according to claim 2, further comprising: A limiter that sets a limit on the necessary number of documents for the document list, wherein The determining unit obtains the relationship between the terms by using the correspondence between a document list obtained by extracting documents including the search terms received by the receiver from documents that match the restrictions set by the restrictor and the pre-stored terms, and determines the recommended terms.

6. The search device according to claim 5, wherein, The limiter determines the necessary number of documents by using the number of documents and the number of pre-defined recommended terms.

7. The search device according to any one of claims 1 to 6, wherein, The determining unit uses the Jaccard coefficient, Dice coefficient, or Simpson coefficient to determine the recommended terms.

8. The search device according to any one of claims 1 to 6, wherein, The determining unit determines the recommended terms by using the difference in the number of documents obtained by adding the terms to be recommended to the query.

9. The search device according to claim 1, wherein, The determining unit determines the recommended terms by using virtually determined virtual terms. The search results using the virtual terms include a predetermined ideal number of documents, and the search results do not overlap with the search results using other terms.

10. The search device according to any one of claims 1 to 6 and 9, wherein the search device further comprises: The display unit displays the number of documents used for the query before the refined search, the number of documents with any recommended terms added to the query, the overlap with any recommended terms added to the query, the imbalance with any recommended terms added to the query, and the loss with any recommended terms added to the query as corresponding areas.

11. The search device according to claim 10, further comprising: An adding unit adds a term to the query that corresponds to the region selected from the region.

12. A search system, the search system comprising: The search device according to any one of claims 1 to 11; as well as The information processing terminal inputs the terms to be received by the receiver into the information processing terminal, and the information processing terminal displays the determination result obtained by the determination unit.

13. A non-transitory computer-readable medium storing a program that causes a computer to perform a process for searching, the process comprising: Receive search terms used for searching; as well as The recommended terms to be used for refined searches are determined, each refined search using a corresponding recommended term from the recommended terms and the search term, wherein... The recommended terms are determined by obtaining the interaction information between the probability of a term being selected and the probability of the refined search type as a relation between terms, based on i) the amount of overlap between the results of the refined search and / or ii) the amount of difference between the number of results in one of the refined searches and the number of results in the other.

14. A search method, the search method comprising the following steps: Receive search terms used for searching; as well as The recommended terms to be used for refined searches are determined, each refined search using a corresponding recommended term from the recommended terms and the search term, wherein... The recommended terms are determined by obtaining the interaction information between the probability of a term being selected and the probability of the refined search type as a relation between terms, based on i) the amount of overlap between the results of the refined search and / or ii) the amount of difference between the number of results in one of the refined searches and the number of results in the other.

Images