JP2012053605A - Information retrieval device, information retrieval method and information retrieval program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To quickly acquire, when calculating scores, those values, and to output the scores of each document with respect to retrieval words even when a document set is limited under retrieval conditions.SOLUTION: An information retrieval device includes: retrieval condition input means for inputting retrieval conditions including one or more retrieval words and classification keys; storage means for storing index information in which a population of documents to be retrieved is associated with each of those classification keys, and document storage numerical values and mean document length values of the documents in the population and document frequency values including each of words in a plurality of documents to be retrieved are calculated and stored; retrieval means for acquiring the document storage numerical values and the mean document length values in the population associated with the classification key and the document frequency values in the population including a word matched with a retrieval word among the documents in the population associated with the classification key, and for calculating the degree of adaptation of each document in the population based on the acquired values, and for retrieving the document; and retrieval result output means for outputting a retrieval result by attaching the degree of adaptation to the document.

Description

  The present invention relates to an information search apparatus, an information search method, and an information search program.

  In recent years, search technology for electronic data or search result ranking search technology has become an increasingly important technology because of the increase in the number of search results due to an increase in the amount of information to be searched. This is because the desired information is buried in a large amount of search results, making it difficult to find.

  The ranking search used in the search system is a function for outputting a search result set satisfying the search conditions in order by a certain evaluation method. If the search conditions and the evaluation method are appropriate, there is a high possibility that data required by the user will be output at the top of the ranking. If there is no data close to the data required by the user at the top of the ranking, the search conditions can be reviewed at an early stage. Thus, the ranking search is an indispensable function for reducing the search burden on the user.

  Evaluation methods that influence the accuracy of the ranking search include a method that uses the number of referenced pages that are often used in Web page searches and a method that uses the appearance frequency in a database of search terms included in the search conditions. In particular, the latter is effective even for data that does not have references / references to other data such as hyperlinks of Web pages and citations of academic papers, and has a wide range of uses.

  Now, the evaluation method of the search result set using the appearance frequency of the search word database may not always be appropriate. Specifically, in the evaluation method of the importance level of a search word using the appearance frequency, the importance of the search word is evaluated higher as the data including a certain search word is less in the database. Then, an evaluation method is used in which a document including many search words having such high importance and each search word frequently appears in the data has a higher score. That is, the importance of the search term depends on the population of information to be searched in the database, and the range of this population affects the score.

  For example, when searching for a patent document in Patent Document 1, classification information such as IPC (International Patent Classification) or FI (File Index) is set in addition to designation of a word included in the document. This is the case. When classification information is set in this way, a search using a search word (keyword) is performed within a range of designated classification information, that is, within a limited population.

  Thus, in the search condition, not only the search term but also the document to be searched is specified (that is, when the population of information to be searched is limited), the mother that determines the importance of the search term After the group is changed to the document set specified by the search condition, the importance of the search term can be evaluated.

  However, the method of changing the document population used for the evaluation as described above to a limited set can evaluate the importance of the search word within the limited range of the document set, so that a more appropriate score can be calculated. On the other hand, the conventional index structure has a problem that this evaluation cannot be performed at high speed.

  In a conventional full-text index, in order to quickly search which document contains an index word, at least a set of an index word and a list of document IDs corresponding to the document including the index word is stored. Yes. Further, when performing a ranking search, index word frequency (tf value: number of target index words in the target document), index word document frequency (df value: document including the target index word in the target document set) Number), the number of stored documents, the document length of the target document, the average document length of the target document set, and the like are stored in a full-text index.

  Each value is a parameter used when calculating the score, and is an important value related to the accuracy of the ranking search. Among these, values such as index word document frequency, average document length, and document storage count are classified as search conditions. If there is a condition to narrow down the search target such as information, the document population is changed to a limited set by the classification information, so those values stored in the full-text index cannot be used as they are, so the search It had to be calculated again. This means that there is a trade-off between search accuracy and speed.

  The present invention has been made in view of the above points, and in order to avoid a trade-off relationship between search accuracy and speed, even when a document set is limited by a search condition, the document set is limited. An information search device having an index that can quickly acquire the document frequency, average document length, and document storage number of the corresponding index word, and acquiring these values at high speed when calculating the score, and outputting the score of each document for the search word An object of the present invention is to provide an information search method and an information search program.

  In order to achieve the above object, an information search apparatus according to the present invention is an information search apparatus that determines an order in which a plurality of search target documents are displayed based on a degree of suitability for a specified search condition. A search condition input means for inputting a search condition including the above search terms and a classification key for limiting a population of search target documents to be determined in the order among the plurality of search target documents; A population of the search target document is associated with each classification key, a document storage value in the population, an average document length value of the document in the population, and a plurality of search target documents Corresponding to an input classification key when the search condition is input, and storage means storing index information in which at least a document frequency value that is the number of documents in the population including the word for each word is calculated and stored Attached Document storage values and average document length values of documents in the population, and documents in the population including words that match the input search word among the documents in the population associated with the input classification key A document frequency value, which is a number, is obtained from the search information of the storage means, and based on the acquired document storage numerical value, average document length value, and parameter value for calculating a fitness including the document frequency value, A search means for searching for a document by calculating a goodness of fit for each document in the group, and a search result output means for adding a goodness of fit to the searched document and outputting it as a search result .

  In order to achieve the above object, in the information search apparatus, the index information in the storage means is created by associating the word with each classification key, and the search means includes an input classification key. The document frequency value is acquired by tracing the index in the order of input search terms.

  In order to achieve the above object, in the information search apparatus, the index information of the storage means is created by associating the classification key with each word, and the search means includes an input search word. The document frequency value is obtained by tracing the index in the order of the input classification key.

  In addition, what applied the arbitrary combination of the component of this invention, expression, or a component to a method, an apparatus, a system, a computer program, a recording medium, etc. is also effective as an aspect of this invention.

  According to the present invention, even when a document set is limited by a search condition, an index is provided that can quickly acquire a document frequency, an average document length, and a document storage number of an index word corresponding to the limited document set, and a score It is possible to provide an information search apparatus, an information search method, and an information search program that acquire these values at high speed at the time of calculation and output the score of each document for the search word.

It is a figure which shows the example of the information search system which concerns on this Embodiment. An example of a search condition designation screen is shown. It is a block diagram which shows the hardware constitutions of the information search device 1 which concerns on this embodiment. The structural example of the conventional index information (a full-text index and a non-full-text table) is shown. An example of the flow of inserting document data into conventional index information (full-text index and non-full-text table) is shown. A search flow example using conventional index information (full-text index and non-full-text table) is shown. The structural example of the composite index 141 of the full text and non-full text data which concerns on this embodiment is shown. An example of a flow of inserting document data into the composite index 141 of full text and non-full text data according to the present embodiment is shown. An example of a search flow using the composite index 141 according to the present embodiment is shown. An example of a search flow using the composite index 141 according to the present embodiment is shown. An example of a flow of inserting document data into the composite index 142 of full text and non-full text data according to the present embodiment is shown. An example of a search flow using the composite index 142 according to the present embodiment is shown.

  The best mode for carrying out the present invention will be described below with reference to the drawings. In the present embodiment, an information search system including an information search device for searching a document will be described as an example.

[System configuration]
FIG. 1 is a diagram illustrating an example of an information search system according to the present embodiment. As shown in FIG. 1, the information search system according to the present embodiment includes an information search device 1, a client device 2, and a search target information database (DB) 150. The client device 2 is configured by a general information processing device such as a PC (Personal Computer). The information retrieval apparatus 1 is connected to the client apparatus 2 via a network, and is operated as a server that retrieves document information stored in the retrieval target information DB 200 in response to a retrieval request from the client apparatus 2.

  The search target information DB 200 (hereinafter simply referred to as DB) stores documents and documents as search target information. That is, the search target information according to the present embodiment is document information (document file) stored in the DB 200. As shown in FIG. 1, in this embodiment, an example in which the DB 200 is provided separately from the information search apparatus 1 will be described. However, the DB 200 can be configured inside the information search apparatus 1. The DB 200 is configured by a nonvolatile storage medium such as an HDD.

  As shown in the figure, the information search apparatus 1 according to this embodiment includes a condition input unit 110, a search unit 120, a search result output unit 130, and an index DB 140.

  The condition input unit 110 inputs search conditions from the user client device 2. This is realized by the I / F 50 shown in FIG. The search condition can also be input from the operation unit 70 provided in the information search apparatus 1.

  The search unit 120 searches for a target document using the search information in the index DB 140 based on the input search condition. The search result includes a document list display in a ranking format according to the degree of matching (score) of the search condition.

  The search result output unit 130 outputs the searched search result. When a search condition is input from the client device 2 of the user, the search result is output to the client device 2.

  The index DB 140 is a DB storage unit that stores index information (index data) of documents stored in the DB 200, and is realized by, for example, the HDD 40 or the RAM 20 illustrated in FIG. If the number of documents to be searched is enormous, the time required for the search will become long. However, index information that enables high-speed searching by scanning the document group in the DB 200 to be searched in advance. Improve search performance by preparing. The search information is also called an index, and a character string is a search key. A specific example of the index will be described later with reference to the drawings.

  Here, the search condition input to the condition input unit 110 can be input from, for example, the client device 2 (or the operation unit 70). FIG. 2 shows an example of a search condition designation screen. The user inputs search conditions in “ordinary text designation (or keyword designation)” and “classification information designation” on the screen. In “ordinary sentence designation (or keyword designation)”, a search term is input in a sentence format or a word format. In “classification information designation”, a predetermined classification key or classification code for designating a population of documents to be searched is input. For example, when the search target document is a patent document, “G06F 17/30” or the like is input here. As described above, the information search apparatus 1 according to the present embodiment performs a search process based on at least one or more input search terms and one or more predetermined classification keys.

  Next, a hardware configuration of the information search apparatus 1 according to the present embodiment will be described. FIG. 3 is a block diagram illustrating a hardware configuration of the information search apparatus 1 according to the present embodiment. As shown in FIG. 3, the information search apparatus 1 according to the present embodiment has the same configuration as an information processing terminal such as a general server or a PC (Personal Computer). That is, the information search apparatus 1 according to the present embodiment includes a CPU (Central Processing Unit) 10, a RAM (Random Access Memory) 20, a ROM (Read Only Memory) 30, an HDD (Hard Disk Drive) 40, and an I / F 50. 80 is connected. Further, an LCD (Liquid Crystal Display) 60 and an operation unit 70 are connected to the I / F 50.

  The CPU 10 is a calculation means and controls the operation of the entire information retrieval apparatus 1. The RAM 20 is a volatile storage medium capable of reading and writing information at high speed, and is used as a work area when the CPU 10 processes information. The ROM 30 is a read-only nonvolatile storage medium and stores a program such as firmware. The HDD 40 is a non-volatile storage medium that can read and write information, and stores an OS (Operating System), various control programs, application programs, and the like. The I / F 50 connects and controls the bus 80 and various hardware and networks. The LCD 60 is a visual user interface for the user to check the state of the information search device 1. The operation unit 70 is a user interface such as a keyboard and a mouse for the user to input information (for example, a search condition) to the information search apparatus 1. As described with reference to FIG. 1, the information search apparatus 1 according to the present embodiment is operated as a server for the client apparatus 2. Therefore, user interfaces such as the LCD 60 and the operation unit 70 can be omitted.

  In such a hardware configuration, a program stored in a storage medium such as the ROM 30, the HDD 40, or an optical disk (not shown) is read into the RAM 20, and operates according to the control of the CPU 10, thereby configuring the software control unit (FIG. 1). Is done. A functional block that realizes the function of the information search apparatus 1 according to the present embodiment is configured by a combination of the software control unit configured as described above and hardware.

[Conventional example]
Here, first, the configuration and operation of a conventional full-text index will be described, and then the configuration and operation of a full-text and non-full-text composite index according to the present embodiment will be described. The system configuration may be substantially the same as that described above in the conventional example. However, there is a difference in the search operation using the index information (index data) in the index DB 140 and the index information by the search unit 120.

  FIG. 4 shows a configuration example of conventional index information (full-text index and non-full-text table). A full-text index is an index word (indexkey = A, indexkey = B, etc. in the figure) and an entry that stores the document ID of the document containing the index word (unique value attached to each document, docID in the figure), etc. Has a data structure that can be retrieved quickly. If a ranking search is to be performed, the word frequency of the index word (tf: Term Frequency in the figure), which is the number of times the search word included in the specified search condition appears in each document, and the search word are included. The entry also includes the document frequency of the index word (df: Document Frequency in the figure), which is the number of documents. In addition, the document length table includes the document length of each document (len in the figure), the average document length of the document set (aveLen in the figure), the number of stored documents (N in the figure), and the like. The full-text index is indexed in advance by extracting words in the DB 200 document group to be searched and calculating the above values (td, tf, len, aveLen, etc.).

  Regarding the non-full-text table, the document of the DB 200 is categorized in advance, and a classification key (key = # 1, key = # 2, etc. in the figure) is designated in advance. In the non-full-text table, the document ID (docID in the figure) belonging to the classification key is indexed for each classification key, and the data structure that can quickly obtain the document ID of the population of the search target document from the input classification key have. If reverse lookup is necessary from the document ID, the classification key can be acquired by reverse lookup.

  In this figure, as an example, the structure for acquiring entries corresponding to index terms is assumed to be a tree structure, and the structure for storing document lengths and non-full-text tables are assumed to be arrays. Of course, there are various other methods for implementing individual data structures.

  FIG. 5 shows a flow example of document data insertion into conventional index information (full-text index and non-full-text table). That is, when a document is added to the DB 200, this additional document is additionally updated to the index of the index information DB 140. Assume that the input to the index includes a document ID of the additional document, full-text data of the additional document, and non-full-text data (for example, a classification key of the additional document). Also, it is assumed that the full text data has been analyzed, and a set of words: index word (indexkey) and word frequency (tf) included in the full text data in the additional document is obtained through calculation.

  First, the non-full-text table is accessed (S101). Data with non-full text data as a key and a value corresponding to the non-full text key as a document ID is added to the non-full text table (S102).

  Next, the full text index is accessed (S103). The entry of the index word (indexkey), which is the value of the full-text index, is acquired using one word in the additional document as the key of the full-text index (S104). One line of a set of document ID (docID) and word frequency (tf) is added to the entry (S105). Since one set of document ID (docID) and word frequency (tf) is added, the document frequency (df) is incremented by 1 (S106).

  Next, the same processing is repeated using the new one word in the additional document as a key for the full text index (S107 → S104). If there are no more new words, the process moves to the next step (S107 → S108).

  Next, the document length table is accessed (S108). One line of data with the document ID as a key and the value for the document ID as the document length (len) is added to the document length table (S109). Since one document ID is added, the number of stored documents (N) in the document length table is incremented by 1, and the average document length (aveLen) is updated according to the increase in the document length of the added document (S110).

  FIG. 6 shows an example of a search flow using conventional index information (full-text index and non-full-text table). In the search, the search condition input by the user is a search word and a classification key (for example, FIG. 2). Matching by a search word (also referred to as a full-text condition) and matching by a classification key (also referred to as a non-full-text condition) A search result that satisfies both conditions is used. For example, when the full-text condition is input as “ordinary sentence designation” as shown in FIG. 2, the sentence “D of C of A by B” is changed to “A / of / B / in / C / dated / D ”, And a meaningful word may be extracted and used as a search word (A, B, C, D). Put this in the search term list.

  First, the non-full-text table is accessed (S201). All document IDs satisfying the conditions are acquired using the classification key of the non-full text condition as a key (S202). Further, the number of acquired document IDs is counted and acquired as a narrowed document storage number (N). The narrowed document storage number (N) is the number of classified set documents limited by the classification key. If there are a large number of document IDs here, the acquisition and the cost of securing the memory area are required.

  Next, the full text index is accessed (S203). Using the index word as a key of the full-text index, an entry of the index word that is the value of the full-text index is acquired (S204). Then, only the set that matches the document ID acquired in the non-full text table among the sets of the document ID (docID) and word frequency (tf) included in the acquired entry is stored in the memory (S205). The acquisition cost is required because the group to be held is determined while checking whether or not they match. The narrowed document frequency (df) in the document population limited by the classification key is obtained from the number of pairs of the document ID (docID) and the word frequency (tf) held (S206).

  Next, the document length table is accessed (S207). All document lengths of document IDs that match the document IDs obtained in the non-full-text table are acquired, as well as the document lengths of the pairs that match the document IDs obtained from the entries. Here, since the number of document IDs obtained in the non-full-text table is larger than the number of document IDs obtained from the entry, the acquisition cost of the document length is required. From the acquired document length, the narrowed-down average document length (aveLen) in the document population limited by the classification key is calculated (S209).

  Next, the degree of conformity with each document for each index word, word frequency (tf) acquired so far, narrowed document frequency (df), number of narrowed documents stored (N), document length (len), narrowed average Calculation is performed using the document length (aveLen) (S210). For specific calculation of the degree of matching (score) of the document with respect to the input index word, a well-known one can be applied. For example, Japanese Patent Application Laid-Open No. 2009-271659 (paragraph numbers 0055-0057, 0085, etc.) by the present inventor Can be helpful. Since the document frequency, the average document length, and the document storage number here are values obtained from a limited population of documents that are narrowed down by a classification key that is a non-full-text condition, the accuracy of the fitness is high.

  Next, if there is a search word remaining in the full-text condition index word list, the access to the full-text index is repeated (S211 → S204), and if there is no remaining, the process proceeds to the next step (S211 → S212).

  Next, when a plurality of search terms are input, since the fitness of the document is calculated for each search term, the fitness is synthesized for each document (S212). In order to display documents in a ranking format as a search result, document IDs and the like are returned to the calling side such as the client apparatus 2 in descending order of fitness (S213).

[Embodiment 1]
Next, the configuration and operation of the full-text and non-full-text composite index according to the first embodiment will be described.

  FIG. 7 shows a configuration example of the composite index 141 of full text and non-full text data according to the present embodiment. Compared with FIG. 4 of the conventional example, the composite index 141 shown in FIG. 4 integrates the full-text index and the non-full-text table of FIG. 4, and combines the index structure of the full-text index and the document length table into the non-full-text of the non-full-text table. It is divided by the value of data (key = # 1, key = # 2, etc. in the figure). Since the non-full-text data is, for example, a classification code, it is divided into numbers according to the number of classification categories of the document.

  The characteristics of the composite index 141 are that the index is divided into non-full-text data values and indexed, so that the index word narrowing document frequency (df) and the narrowed average document length (aveLen), which change according to the non-full-text data values, Since the number of stored narrowed documents (N) can be calculated in advance and registered in the index, these values can be acquired at a low cost. This will be described later in the search operation.

  In each index structure thus divided, the branch from the root node (Root) of the composite index to the value of non-full-text data (# 1, # 2, etc. in the figure) leads to the root node of the search term of the full-text index It has a structure that can be traced. A plurality of document length tables divided by non-full-text data values also have a structure that can be uniquely selected by non-full-text data values (# 1, # 2, etc. in the figure).

  FIG. 8 shows an example of a flow of inserting document data into the composite index 141 of full text and non-full text data according to the present embodiment. That is, when a document is added to the DB 200, the additional document is additionally updated to the composite index 141 of the index information DB 140. It is assumed that the input to the index includes a document ID of the additional document, full text data of the additional document, and non-full text data (for example, a classification code of the additional document). Further, it is assumed that the full text data in the additional document has been analyzed, and a set of words: index word (indexkey) and word frequency (tf) included in the full text data is obtained through calculation.

  First, the composite index 141 is accessed, and the index structure (tree structure) indicated by the non-full-text data (additional document classification code) is accessed (S301). Using one word included in the full-text data in the additional document as a key of the full-text index, an index word entry that is the value of the full-text index is acquired (S302). One line of a set of document ID (docID) and word frequency (tf) is added to the entry (S303). Since one set of document ID (docID) and word frequency (tf) is added, the narrowed-down document frequency (df) is incremented by 1 (S304).

  Next, the same processing is repeated using the new one word in the additional document as the key of the full text index (S305 → S302). If there are no more new words, the process moves to the next step (S305 → S306).

  Next, the document length table indicated by the non-full-text data is accessed (S306). One line of data with the document ID as a key and the document ID (len) as the value for the document ID is added to the document length table (S307). Since one document ID has been added, the number of stored documents (N) in the document length table is incremented by 1, and the reduced average document length (aveLen) is updated in accordance with the increase in the document length of the added document (S308). . It should be noted that the number of stored items (N) indicates the number of documents in the document population classified by the classification key. The narrowed-down average document length (aveLen) indicates the average document length of documents in the document population classified by the classification key.

  As described above, even when compared with the conventional example of FIG. 6, there is almost no increase in the document data insertion cost due to the composite indexing.

  FIG. 9 shows an example of a search flow using the composite index 141 according to the present embodiment. In the search, the input search conditions are a search word and a classification code (for example, FIG. 2), and both matching by the search word (referred to as full text condition) and matching by the classification code (referred to as non-full text condition) are satisfied. Condition. For example, as shown in FIG. 2, when the full-text condition is input with “ordinary sentence designation” as shown in FIG. 2, as described above, the sentence may be divided into words and meaningful words may be extracted and used as search words. Put this in the search term list.

  First, the composite index 141 is accessed, and the index structure (tree structure) indicated by the non-full-text data is accessed using the inputted classification code of the non-full-text condition as a key (S401). For example, the tree of “key = # 1” is accessed (FIG. 7). Using the input index word as a key of the full-text index, an entry of the index word that is the value of the full-text index is acquired (S402). For example, if the input index word is “A”, a child tree of “indexkey = A” is accessed, and an entry of “indexkey = A” is acquired. All pairs of document ID (docID) and word frequency (tf) included in the acquired entry are acquired (S403). For example, if it is a set (docID, tf), it will be (1,3), (3,1), etc. Further, the narrowed document frequency (df) is acquired from the entry (S404). For example, df = 10 is acquired.

  Next, the document length table indicated by the non-full-text data is accessed (S405). For example, the document length table of “key = # 1” is accessed (FIG. 7). From the document length table, all document lengths (len) of document IDs that match the document IDs acquired in S403 are acquired (S406). For example, the document length “500” of the document ID “1”, the document length “1500” of the document ID “3”, etc. Further, the narrowed-down average document length (aveLen) and the narrowed-down document storage number (N) are acquired (S407). For example, aveLen = 900 is acquired and N = 50 is acquired. Here, since the narrowed document frequency (df), the narrowed average document length (aveLen), and the narrowed document storage number (N) have already been registered in the composite index 141, these values need only be acquired, and special calculation is performed. Etc. are not required.

  Next, the degree of conformity with each document for each index word, word frequency (tf) acquired so far, narrowed document frequency (df), number of narrowed documents stored (N), document length (len), narrowed average Calculation is performed using the document length (aveLen) (S408). Note that, as described above, a publicly known one can be applied to the specific calculation of the degree of matching (score) of a document with respect to an input index word (for example, JP 2009-271659 A). Since the narrowed document frequency, the narrowed average document length, and the number of narrowed documents stored here are values obtained by narrowing down from a population of documents limited by a classification code that is a non-full-text condition, the accuracy of the fitness is high.

  Next, if there is a remaining search word in the full-text condition index word list, the access to the full-text index is repeated (S409 → S402), and if there is no remaining, the process proceeds to the next step (S409 → S410).

  Next, when a plurality of search terms are input, since the fitness of the document is calculated for each search term, the fitness is synthesized for each document (S410). In order to display documents in a ranking format as a search result, document IDs and the like are returned to the calling side such as the client apparatus 2 in descending order of fitness (S410).

  As described above, the word frequency (tf), the document frequency (df), the document storage number (N), the document length (len), the average document length (aveLen) are calculated when calculating the fitness with each document for each index word. ), Etc., but the values of document frequency, average document length, and number of stored documents are limited document frequency based on the document population limited by the classification code that is a non-full-text condition, Since the document length and the number of narrowed document storages are used, the degree of conformance can be made high. Since the composite index 141 according to the present embodiment has the above-described index structure, the narrowed-down document frequency (df), the narrowed-down average document length (aveLen), and the number of narrowed-down document storages (N) are stored in advance. Since these values are simply acquired when the search unit 120 performs a search (no special calculation or the like is required), the search speed can be improved.

  In other words, the conventional full-text index is expanded to create a composite index of full-text elements and non-full-text elements, and the index is divided in correspondence with each set divided according to the condition that limits the population in the composite index. Thus, the document frequency of each index word in each set, the average document length and the number of documents stored in each set can be stored in advance in each divided index. These values can be used without calculation.

[Embodiment 2]
Next, the configuration and operation of a full-text and non-full-text composite index according to the second embodiment will be described.

  FIG. 10 shows an example of a search flow using the composite index 141 according to this embodiment. The composite index 142 shown in the figure is substantially the same as the composite index 141 according to the above-described embodiment. The full-text index and non-full-text table of Fig. 4 are integrated, and the index structure of the full-text index and the document length table are the values of non-full-text data of the non-full-text table (such as key = # 1, key = # 2 in the figure). Divided by.

  However, the composite index 141 according to the second embodiment is compared, and the composite index 142 according to the present embodiment is different in terms of a route to reach an entry for acquiring the document ID and the document frequency. Each index structure divided here is traced to the root node of non-full-text data by branching from the root node (Root) of the composite index to the value of full-text data (indexkey = A, indexkey = B, etc. in the figure). It has a structure that can.

  In any case, this compound index 142 is also divided into non-full-text data values and indexed, so that the index word document frequency (df) and average document length (aveLen) change depending on the non-full-text data values. Can be obtained at low cost.

  FIG. 11 shows an example of a flow of inserting document data into the composite index 142 of full text and non-full text data according to the present embodiment. However, compared with FIG. 8, the difference is S501 and S502.

  First, the composite index 142 is accessed, and the index structure (tree structure) is accessed from the root node using one word included in the full-text data in the additional document as a key of the full-text index (S501). Next, using the non-full-text data (additional document classification code) as a key, an index word entry that is the value of the full-text index is acquired (S502). In short, in comparison with FIG. 8, before the entry point for storing the combination of the document ID and the word frequency is reached as a new entry, the non-full-text data (classification of the additional document) is first reached. Code) or a word in the full-text data of the additional document. In any case, finally, the document ID (docID) and the word frequency (tf) of the additional document at the entry point A new entry consisting of a set is added. The narrowed document frequency (df) is also incremented by one and updated. Since the subsequent steps are the same as those in FIG. 8, the description thereof is omitted.

  FIG. 12 shows an example of a search flow using the composite index 142 according to the present embodiment. However, the difference is S601 and S602 as compared with FIG.

  First, the composite index 142 is accessed, and the index structure (tree structure) indicated by the full text data is accessed from the root node using the input index word as the key of the full text index (S601). Next, using the classification code of the input non-full-text condition as a key, an index word entry that is the value of the full-text index is acquired (S602). In short, as compared with FIG. 9, after access to the composite index 142 and before reaching the entry for obtaining a set of document ID and word frequency, the non-full-text data (additional document classification code) is first drawn. Or a word in the full-text data of the additional document, and in any case, finally, an entry of a set of the document ID (docID) and word frequency (tf) of the additional document is acquired. The narrowed document frequency (df) is also acquired. Since the subsequent steps are the same as those in FIG. 9, the description thereof is omitted.

[Summary]
As described above, the information search apparatus 1 according to the present embodiment calculates the word frequency (tf), the document frequency (df), the document storage number (N), the document length (len) when calculating the fitness with each document for each index word. ), Average document length (aveLen), etc. are required. Of these values, the document frequency, average document length, and number of stored documents are not included in the document population limited by the classification code that is a non-full-text condition. Since the narrowed-down document frequency, the narrowed-down average document length, and the number of narrowed-down document storages are acquired from the composite index and used, the fitness level can be increased. Since the composite index according to this embodiment has the above-described index structure, the narrowed document frequency (df), the narrowed average document length (aveLen), and the number of narrowed document storages (N) are registered in the composite index in advance. When the information search apparatus 1 performs a search, these values need only be acquired (no special calculation or the like is required), so that the search speed can be improved.

  That is, according to the above-described embodiment, even when a document set is limited by a search condition, an index that can quickly acquire the document frequency, average document length, and document storage number of index words corresponding to the limited document set. It is possible to provide an information search apparatus that obtains these values at high speed when calculating the score and outputs the score of each document for the search word.

  Although the present invention has been described based on each embodiment, the present invention is not limited to the requirements shown here, such as combinations with other elements listed in the above embodiments. With respect to these points, the present invention can be changed within a range that does not detract from the gist of the present invention, and can be appropriately determined according to the application form. Moreover, what applied the component, expression, or arbitrary combinations of the component of this invention to a method, an apparatus, a system, a computer program, a recording medium, etc. is also effective as an aspect of this invention.

1 Information Retrieval Device 2 Client Device 10 CPU
20 RAM
30 ROM
40 HDD
50 I / F
60 LCD
70 Operation unit 80 Bus 110 Condition input unit 120 Search unit 130 Search result output unit 140 Index DB
200 Search target information DB

JP 2009-271659 A

Claims (5)

An information search device that determines an order of displaying a plurality of documents to be searched based on a degree of conformity to a specified search condition,
A search condition input means for inputting a search condition including one or more search terms and a classification key for limiting a population of search target documents to be determined in the order among the plurality of search target documents;
A group of documents to be searched is associated with each classification key, a document storage value in the group, an average document length value of documents in the group, and a plurality of documents in the plurality of search objects Storage means for storing index information in which at least a document frequency value that is the number of documents in the population including the word is calculated and stored for each word of
When the search condition is input, the document storage numerical value and the average document length value of the document in the population associated with the input classification key, and the documents in the population associated with the input classification key The document frequency value that is the number of documents in the population including the word that matches the input search word is acquired from the search information of the storage means, and the acquired document storage value, average document length value, and document frequency value are acquired. Search means for searching for a document by calculating a fitness for each document in the population based on a parameter value for calculating a fitness including
A search result output means for attaching the degree of relevance to the retrieved document and outputting it as a search result;
An information retrieval apparatus comprising: The index information of the storage means is created by associating the word with each classification key,
The search means obtains the document frequency value by tracing the index in the order of an input classification key and an input search term;
The information retrieval apparatus according to claim 1. The index information of the storage means is created by associating the classification key with each word,
The search means obtains the document frequency value by tracing the index in the order of input search terms and input classification keys;
The information retrieval apparatus according to claim 1. An information search method in an information search apparatus for determining a display order of a plurality of search target documents based on a degree of suitability for a specified search condition,
Among the plurality of search target documents, the search target document population is associated with each classification key that limits the search target document population for which the order is determined, and within the population A document storage value, an average document length value of documents in the population, and a document frequency value that is the number of documents in the population including the word for each word in the plurality of documents to be searched are calculated. A storage procedure for storing the stored index information;
A search condition input procedure for inputting a search condition including one or more search terms and the classification key;
When the search condition is input, the document storage numerical value and the average document length value of the document in the population associated with the input classification key, and the documents in the population associated with the input classification key The document frequency value that is the number of documents in the population including the word that matches the input search word is acquired from the search information of the storage means, and the acquired document storage value, average document length value, and document frequency value are acquired. A search procedure for searching for documents by calculating the fitness for each document in the population based on the parameter value for calculating the fitness including
A search result output procedure for outputting a search result with a goodness of fit to the searched document,
A method for retrieving information, comprising:   An information search program for causing a computer to execute the information search method according to claim 4.

Images