JP2006323575A - Document retrieval system, document retrieval method, document retrieval program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve document retrieval processing if a retrieval condition includes at least numerical information. <P>SOLUTION: The document retrieval system comprises a numerical information classification means classifying one or more pieces of numerical information extracted from an input document for each unit kind of numerical information; a characteristic quantity calculation means calculating a characteristic quantity for estimating a document including numerical information in a predetermined numerical range for each unit kind; a characteristic quantity storage means storing the characteristic quantity calculated by the calculation means for each unit kind; a numerical importance calculation means acquiring, based on a unit kind of input numerical information inputted as a retrieval condition, the characteristic quantity of the corresponding unit kind from the characteristic quantity storage means and calculating, based on at least the acquired characteristic quantity, the weight of the input numerical information inputted as the retrieval condition; and a document retrieval means retrieving, based on the weight of input numerical information calculated by the numerical importance calculation means, a document from registered documents. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a document search system, a document search method, a document search program, and a recording medium, and in particular, a document search system and a document search that enable a document search when a numerical value is included as a search condition in an input keyword or sentence. The present invention can be applied to a method, a document search program, and a recording medium.

  2. Description of the Related Art Conventionally, a document search system that performs document search based on one or a plurality of keywords, as shown in Patent Document 1, uses a keyword appearance frequency tf (Term Frequency) and a document appearance number df (Document Frequency) as a keyword. Relevance was calculated and output as a search result from a document with high relevance.

  Here, the word appearance frequency tf is the appearance frequency of a keyword in a certain document, and the document appearance number df is the number of documents in which the keyword appears.

  At this time, df is used to measure the degree of contribution of the keyword to document narrowing. That is, a keyword with a small df that appears only in a small number of documents is likely to contribute to narrowing down the documents, and a keyword with a large df (a keyword that appears in most documents in extreme cases) is not very useful for narrowing down the search. Therefore, it is common to set the calculation formula of the fitness so that the word having a smaller df has a greater influence on the calculation of the fitness.

  Further, tf is used to measure that a document containing many keywords matches the search condition more, and is used so that the fitness of the document increases as tf increases.

  As described above, the product of df and tf is added for all keywords for each document, thereby calculating the degree of fitness of the document with respect to the keyword and performing document search.

JP 2000-322416 A

  By the way, using the above-described conventional document search system, if an attempt is made to perform a document search using numerical values as search conditions, the conventional method cannot efficiently use what corresponds to df, and processing related to document search will be performed. The problem of slow speed occurs. This will be described below.

  In the case of a word keyword, the number of documents in which the word appears, that is, df, can be counted for each appearing word and stored in advance in association with the word at the time of entering the document DB (database). Then, when searching for documents, it is only necessary to refer to the value for each keyword. That is, it is not necessary to perform an operation for searching and counting documents including keywords at the time of searching, and the processing speed can be improved. Such a thing is performed in the conventional general search.

  However, in the case of numerical values, the number of documents to be narrowed differs depending on the numerical value and range specified in the condition. For example, when “10 m” is specified and when “1 m to 100 m” is specified, df varies depending on the value and range of the search condition, such as fewer documents that match the former. In other words, df information reflecting these is required, but there are innumerable variations of such numerical search conditions, and it is not possible to count and store df in advance for any numerical search condition. Therefore, it is impossible to take measures for improving the processing speed as in the keyword.

  For this reason, at the time of the search process, df is obtained only after actually performing a search, comparing numerical conditions, and counting the number of documents falling within the range of the conditions. This makes the processing at the time of retrieval very heavy compared to obtaining df for a conventional keyword, and becomes more prominent as the number of documents to be retrieved is larger.

  In order to solve such a problem, a document search system according to a first aspect of the present invention is a document search system that searches a document that matches a search condition from registered documents registered with at least numerical information as a search condition. 1) Numeric information classification means for classifying each of one or a plurality of numerical information extracted from an input document for each unit type of numerical information, and (2) a feature amount for estimating a document including numerical information is represented by unit type. Feature amount calculating means for calculating with a predetermined numerical range for each, (3) feature amount storing means for storing the feature amount calculated by the feature amount calculating means for each unit type, and (4) input as a search condition Based on the unit type of the input numerical information, the feature amount of the corresponding unit type is acquired from the feature amount storage means, and the weight of the input numerical information input as the search condition based on at least the acquired feature amount And numerical importance calculating means for calculating, characterized in that it comprises a document retrieval means for retrieving a document from the registration document based on the weight of the input numerical value information calculated (5) numerical importance calculating means.

A document search method according to the second aspect of the present invention is a document search method for searching a document that matches a search condition from a registered document registered with at least numerical information as a search condition, and (1) extracted from an input document A numerical information classification step of classifying each of the numerical information pieces that have been performed for each unit type of numerical information, and (2) a feature value for estimating a document that includes the numerical information in a predetermined numerical range for each unit type And (3) feature quantity storage means for storing the feature quantity calculated in the feature quantity calculation process for each unit type, and (4) input numerical value information input as a search condition Based on the unit type, the feature quantity of the corresponding unit type is acquired from the feature quantity storage means, and the weight of the input numerical information input as the search condition is calculated based on at least the acquired feature quantity. And (5) a document search step for searching a document from a registered document based on the weight of the input numerical information calculated in the numerical importance calculation step. Further, the document search program according to the third aspect of the present invention Is a document search program for searching a document that matches a search condition from a registered document registered with at least numerical information as a search condition. (1) One or a plurality of numerical information extracted from an input document Numerical information classification means for classifying each of the above information for each unit type of numerical information, and (2) feature quantity calculation means for calculating a feature quantity for estimating a document containing the numeric information with a predetermined numerical range for each unit type. (3) Feature quantity storage means for storing the feature quantity calculated by the feature quantity calculation means for each unit type, (4) Based on the unit type of the input numerical information input as the search condition, (5) Numerical importance calculation; (5) calculating numerical importance; and (5) calculating the numerical importance, which obtains the weight of input numerical information input as a search condition based on at least the acquired feature A document search program for causing a document search unit to search for a document from a registered document based on the weight of the input numerical information calculated by the unit.

  A recording medium according to the fourth aspect of the present invention is a computer-readable recording medium that records a document search program for searching for a document that matches a search condition from a registered document registered with at least numerical information as a search condition. A computer-readable recording medium having a document search program recorded thereon, wherein the document search program corresponds to the document search program of the third aspect of the present invention.

  According to the present invention, at the time of document registration, for each unit type of numerical information, by calculating a feature amount for a document including a numerical value and / or a numerical range, information regarding the document including predetermined numerical information at the time of document registration is obtained. Since it can be obtained in advance, even when a numerical value and / or a numerical value range is used as a search condition, the search processing can be performed without obtaining information relating to a document including numerical information. Thereby, the processing time concerning a search process can be shortened.

  Hereinafter, embodiments of a document search system, a document search method, a document search program, and a recording medium according to the present invention will be described with reference to the drawings.

  Note that the document search system in the following embodiments can be realized as a computer-executable program. The program may be stored in a computer-readable recording medium, or may be stored in a hard disk of the apparatus.

(A) First Embodiment First, the first embodiment of the document search system, document search method, document search program, and recording medium of the present invention will be described below.

  The present embodiment provides a document search system for outputting a document having a high evaluation value as a search document using a keyword, a numerical value, and / or a range of numerical values as a search condition from a plurality of registered documents. A case where a document search system, a document search method, and a document search program are applied will be described.

  In the present embodiment, a numerical value serving as a search condition is reflected by approximating the number of appearance documents (df) for each predetermined numerical section.

(A-1) Configuration of First Embodiment FIG. 1 is a functional block diagram illustrating functions of a document search system according to the present embodiment.

  As shown in FIG. 1, the document search system 1 of the present embodiment includes a document input unit 3000 that inputs input document information into a database, a document storage unit 2000 that stores input document information as a database, input conditions ( For example, it includes at least a document search unit 1000 that searches the document storage unit 2000 for a document suitable for a keyword or the like.

  First, the configuration of the document storage unit 2000 will be described. The document storage unit 2000 includes at least a DF data storage unit 2200, a numerical range distribution data storage unit 2300, and a document DB (database) unit 2400.

  The DF data storage unit 2200 stores the result counted by the document number counting unit 3200, which will be described later. The df data indicating the number of documents including the word serving as the headword among the input documents is used as the headword. It is memorized for each word.

  The numerical value range distribution data storage unit 2300 stores the calculation result calculated by the numerical value range distribution statistical unit 3400, which will be described later, and the number of documents including numerical values in a predetermined numerical interval preset for each numerical unit type. Is memorized.

  Here, the predetermined numerical section for each unit type of numerical value refers to a section obtained by dividing a numerical range into a finite number by a predetermined rule for each unit type indicated by the numerical value. For example, in this embodiment, an example of the unit type of a numerical value is “bps”, and the numerical value section of this unit is described as a single-digit numerical value range (1 to less than 10, 10 to less than 100, etc.) as one section.

  It should be noted that the predetermined numerical interval for each unit type of numerical values does not need to be “digits”. For example, in the case of numerical values having a wide range, the logarithm of the numerical value is set to a predetermined width so that it is a constant interval. You may make it delimit. In this way, numerical values having a wide range can be divided into a finite number of sections. Also, different numerical sections can be set for each numerical unit type.

  For each document (or document identification information for identifying a document) input to the document input unit 3000, the document DB unit 2400 appears in the document as a word that is a headword that appears in the document and the number of occurrences (tf). Numerical information and the number of appearances are stored. Here, the numerical information refers to information having at least a numerical value and a unit, for example, information obtained by extracting a character string including a numerical value and a unit in a document by a numerical value extraction unit 3300 described later. In the following description, a numerical string extracted from a document and a character string made up of units converted into a predetermined numerical expression will be described as numerical information. In addition, the expression method of numerical information is not specifically limited, A different expression can be used according to unit types. Note that the document DB unit 2400 may store a word that becomes a headword in all registered documents and the total number of appearances for each numerical value information.

  Next, the configuration of the document input unit 3000 will be described. The document input unit 3000 includes at least a numerical value extraction unit 3300, a headword extraction unit 3100, a headword count unit 3200, and a numerical value range distribution statistical unit 3400.

  The headword extraction unit 3100 searches and extracts a word as a headword from a document to be searched. The headword extraction unit 3100 extracts the word as a headword and the number of appearances (tf) as a document or sentence identification information. The data is stored in the document DB unit 2400 in association with each other. Here, various methods can be applied to the method of extracting the word that becomes the headword by the headword extraction unit 3100. For example, the word word extraction unit 3100 has a word dictionary that stores the word that becomes the headword and a rule related to the part of speech. Then, it is possible to apply a method based on morphological analysis by referring to the word dictionary and searching for a word as a headword included in the input document and extracting the search word.

  The document number counting unit 3200 counts the number of documents in which the word obtained as the headword by the headword extraction unit 3100 appears for each word used as the headword, and the counted number of documents per word is DF. The data is stored in the data storage unit 2200.

  The numerical value extraction unit 3300 searches and extracts a character string representing a numerical value from a document to be searched. The numerical value extracting unit 3300 interprets the character string representing the extracted numerical value as a predetermined numerical expression, and in accordance with the interpretation, for each unit type. It converts to numerical information. The numerical value extraction unit 3300 provides numerical information interpreted in a predetermined numerical expression to the document DB unit 2400 and the numerical value range distribution statistical unit 3400. Here, various methods can be applied to the method of extracting a character string representing a numerical value by the numerical value extraction unit 3300. For example, a numerical value is searched from a document, and a character string including the numerical value is set in advance. Based on a method of extracting a numerical value and a unit type by performing morphological analysis with reference to a word dictionary storing words (items) related to the numerical value, or a rule based on, for example, a preset character type or part of speech The method of extracting numerical values and unit types from documents can be applied.

  When the numerical value range distribution statistical unit 3400 receives numerical information from the numerical value extracting unit 3300, the numerical value range distribution statistical unit 3400 classifies the numerical information according to a predetermined numerical section for each unit type. As a result, numerical values can be classified into preset finite intervals for each unit type.

  For example, if the numerical information is “5 × 10 ^ 7 bps (“ ^ ”indicates square),” the number of digits is 7, so this numerical information is classified as “7-digit section”. For example, if the numerical information is “1 × 10 ^ 9”, the number of digits is nine, so the numerical information is classified into “9-digit sections”.

  Further, when the numerical value information is classified, the numerical range distribution statistical unit 3400 counts the number of appearing documents (df) in the numerical section to which the numerical information belongs, and updates the number of appearing documents (df) for each numerical section. Thereby, the number of documents including numerical information belonging to the numerical value section can be managed. Note that the number of appearing documents for each numerical section counted by the numerical range distribution statistical unit 3400 is given to the numerical range distribution data storage unit 2300 for storage.

  For example, in the above example, if the number of appearing documents in the “7-digit section” is “71”, the number of appearing documents is set to “72” according to the classification of the current numerical information “5 × 10 ^ 7 bps”. In addition, when the range of numerical information covers several digits (when it extends over a plurality of sections), processing can be performed by counting each numerical section.

  The numerical range distribution statistical unit 3400 counts and stores the number of documents appearing in each numerical section for each unit type, but the total number of appearances (CF) of each numerical section in all registered documents is stored. You may make it count and memorize | store. In this way, by calculating and storing the number of occurrences of each numerical section in all registered documents, when the numerical section is used as a search condition, the number of appearances of each numerical section can be used as an estimated amount. it can.

  Next, the configuration of the document search unit 1000 will be described. The document search unit 1000 includes at least an input unit 1100, a word importance calculation unit 1200, a numerical importance calculation unit 1300, a document evaluation value calculation unit 1400, and an output unit 1500.

  The input unit 1100 captures keywords, numerical values, numerical ranges, and the like, which are document search conditions, and supplies the captured information to the word importance calculation unit 1200 and the numerical importance calculation unit 1300. Here, the information that can be taken in by the input unit 1100 corresponds to a keyword, a numerical value, a numerical value range, and the like, and each of these pieces of information or a combination thereof is used as a search condition.

  When an input keyword is given from the input unit 1100, the word importance calculation unit 1200 obtains df data for the input keyword from the DF data storage unit 2200, and calculates the importance as a word search key. Here, the word importance calculation method by the word importance calculation unit 1200 is obtained by, for example, extracting the total number of registered documents registered in the document DB unit 2400 and calculating the log (<total number of registered documents> / df). A general evaluation method using idf or the like can be applied. Further, the word importance calculation unit 1200 gives the document evaluation value calculation unit 1400 importance as a search condition for the calculated keyword.

  When the numerical value input from the input unit 1100 is given, the numerical importance calculation unit 1300 refers to the importance of the search condition based on the input numerical value in the numerical range distribution data storage unit 2300 and data that matches the search condition. To calculate. Here, the importance calculation method by the numerical importance calculation unit 1300 is, for example, as long as the numerical value range distribution data storage unit 2300 stores the unit type and the number of appearing documents for each digit. The importance is calculated assuming that the number of appearing documents with respect to the digit corresponds to df in the keyword. As the calculation formula, various methods can be applied as in the case of the word importance calculation unit 1200. For example, a generally known method such as idf can be used. Also, the numerical importance calculation unit 1300 gives importance to the document evaluation value calculation unit 1400 as a search condition for the calculated input numerical value.

  The document evaluation value calculation unit 1400 obtains, for each document in the document DB unit 2400 including a keyword and / or a numerical search condition, the number of word appearances that are the keyword and / or the number of numerical information that conforms to the numerical search condition. It is what. In addition, the document evaluation value calculation unit 1400 stores the document DB unit 2400 based on the tf of each document and the importance of each search condition obtained from the word importance calculation unit 1200 and / or the numerical importance calculation unit 1300. The evaluation value of each document is calculated. Here, the calculation method of the evaluation value of each document by the document evaluation value calculation unit 1400 can be realized by using a general search method such as taking the sum of tf × idf for each search condition. Further, the document evaluation value calculation unit 1400 gives the calculated evaluation value of each document to the output unit 1500.

  The output unit 1500 determines and outputs a document to be output as a search result based on the evaluation value of each document obtained by the document evaluation value calculation unit 1400. If the document DB unit 2400 stores not the document itself but the document identification information, the output unit 1500 outputs the document identification information or the corresponding document based on the document identification information. May be obtained and output.

(A-2) Operation of First Embodiment Next, processing in the document search system 1 of this embodiment will be described with reference to the drawings.

  First, the operation of the document input unit 3000 when a document is input will be described with reference to the flowchart of FIG. Hereinafter, a case where the document 11 shown in FIG. 3 (FIG. 3 is a part of the document) is input will be described as an example.

  4 shows an example of stored contents of the DF data storage unit 2200, FIG. 5 shows an example of stored contents of the numerical range distribution data storage unit 2300, and FIG. 6 shows an example of stored contents of the document DB unit 2400.

  In FIG. 2, when the input document is taken into the document input unit 3000 (S1), the document is subjected to morphological analysis by the headword extraction unit 3100. From the result of the morphological analysis, for example, a headword such as a noun Are extracted (S2).

  For example, when the document 11 is captured, from the result of the morphological analysis by the headword extraction unit 3100, the words “communication speed” of W11, “provider” of W13, “provider” of W15, which are words registered in advance as headwords. Is extracted. Here, the headword can be extracted by referring to a word dictionary prepared in advance.

  When a word that becomes a headword is extracted by the headword extraction unit 3100, the extracted word is associated with a document or document identification information together with the number of appearances, and stored in the document DB unit 2400 (S3).

  For example, in the document 11, the number of occurrences of the word “communication speed” as the headword is “1”, the number of occurrences of the word “provider” is “2”, and these are associated with the document or document identification information. And stored in the document DB unit 2400 (see FIG. 6).

  Further, when a word that becomes a headword is extracted from the document by the headword extraction unit 3100, the extracted word of the headword is given to the document number counting unit 3200, and the headline extracted by the document number counting unit 3200 is displayed. The df of the word is counted (S4). Then, the df of the headword word counted by the document number counting unit 3200 is stored in the DF data storage unit 2200 (S5).

  For example, when the word “W11” of “communication speed” is extracted as a headword from the document 11 by the headword extraction unit 3100, the word “W11” of “communication speed” is given to the document number counting unit 3200. Then, the document number counting unit 3200 increments df by one for the “communication speed” of the word W11 (for example, “120” when the document 11 is input is “119”), and the numerical range distribution data It memorize | stores in the memory | storage part 2300 (refer FIG. 4).

  In S1, when the input document is taken in, the numerical value extraction unit 3300 extracts numerical information including numerical values and units from the document (S6).

  For example, when the document 11 is input, the numerical value extraction unit 3300 extracts the word W12 of “50 Mbps”. Since the extraction process of the character string composed of the numerical value and unit by the numerical value extraction unit 3300 has been described above, a detailed description thereof will be omitted.

  When the numerical value extraction unit 3300 extracts the numerical value information of the character string composed of the numerical value and the unit, the numerical value extraction unit 3300 allows the numerical value information to be divided into preset numerical value intervals. The numerical expression of information is converted into a predetermined format so as to be interpreted as a predetermined numerical expression (S7).

  For example, in this embodiment, since the division of the numerical value section is performed based on the “digit” of the numerical value, the numerical value extraction unit 3300 expresses “50 Mbps” extracted from the document 11 as a numerical value “5 × 10 ^ 7 bps”. And converted into numerical information of a mantissa part “5”, an exponent part “7”, and a unit “bps”.

  When numerical value information is interpreted by the numerical value extraction unit 3300, the numerical value information is stored in the document DB unit 2400 in association with the document or document identification information together with the number of appearances (S8).

  For example, in the document 11, “50 Mbps” extracted as numerical information is interpreted as a numerical expression “5 × 10 ^ 7 bps”, so the number of occurrences of the numerical information “5 × 10 ^ 7 bps” is “1”. This is stored in the document DB unit 2400 in association with the document or document identification information (see FIG. 6).

  When numerical value information is interpreted by the numerical value extraction unit 3300, the numerical value information converted by the numerical value extraction unit 3300 is given to the numerical value range distribution statistical unit 3400. Then, the numerical range distribution statistical unit 3400 confirms the unit of the numerical information, and counts the number of appearance documents of the number of digits of the numerical information of the unit in the numerical range distribution data (S9). In addition, the number of appearing documents of the number information in the unit in the numerical range distribution data counted by the numerical range distribution statistical unit 3400 is stored in the numerical range distribution range data storage unit 2300 (S10).

  For example, information on the mantissa part “5”, the exponent part “7”, and the unit “bps” is given to the numerical value range distribution statistical part 3400 as numerical information from the numerical value extraction part 3300. When the numerical range distribution statistical unit 3400 confirms the unit “bps”, the numerical range distribution statistical unit 3400 increases the number of appearing documents in the section corresponding to the exponent part “7” in the numerical range distribution data of the unit “bps” by one. (For example, if the number of appearing documents before inputting the document 11 is “71”, it is set to “72”) (S9).

  As described above, the above processing is performed on the input document. For example, partial data when 2000 documents are registered in the document storage unit 2000 as registered documents is illustrated in FIGS. 4, 5, and 6. .

  Next, an operation for retrieving a document based on an input from the document storage unit 2000 will be described with reference to the drawings.

  For example, it is assumed that “wireless”, “provider”, “1 Gbps or more” is input as a search condition from the input unit 1100.

  When the search condition is taken into the input unit 1100 (S21), the input search condition is given to the word importance calculation unit 1200 and the numerical importance calculation unit 1300.

  When the input search condition is given to the word importance calculation unit 1200, the word importance calculation unit 1200 extracts the df of each keyword from the DF data storage unit 2200, and the df data of each keyword and the document DB unit Based on the number of registered documents 2400, the importance of each keyword is calculated (S22). Further, each input keyword and the calculated importance of each keyword are given to the document evaluation value calculation unit 1400.

  For example, the word importance calculation unit 1200 refers to the DF data storage unit 2200 illustrated in FIG. 4 and extracts df = 36 for the keyword “wireless” of the input word. In addition, since the number of registered documents registered in the document DB unit 2400 is 2000, the word importance calculation unit 1200 calculates the importance “idf = log (2000/36) = 1.74” of the keyword “wireless”. . In addition, the word importance calculation unit 1200 similarly extracts df = 452 from the DF data storage unit 2200 for the keyword “provider”, and calculates importance idf = 0.64.

  When the input search condition is given to the numerical importance calculation unit 1300, the numerical importance calculation unit 1300 interprets the input condition into a predetermined numerical expression, and the unit of the interpreted numerical value and the numerical value The number of digits is confirmed, and for the unit of the numerical information, the number of appearing documents having the number of digits of the numerical information is extracted from the numerical range distribution data storage unit 2300. Also, based on the number of appearing documents of the number of digits of the numerical information of the unit extracted from the numerical value range distribution data storage unit 2300 by the numerical importance calculation unit 1300 and the number of registered documents registered in the document DB unit 2400. Thus, the importance of the numerical value is calculated (S23). Also, the numerical value as the input condition and the importance of the calculated numerical value are given to the document evaluation value calculation unit 1400.

  For example, the numerical importance calculation unit 1300 performs processing on “1 Gbps or more” that is a numerical value among the search conditions obtained from the input unit 1100. Here, when interpreting “1 Gbps or more”, since “1G = 10 ^ 9”, the number of numerical values is 9 digits, and the numerical value range is “above”. The number of appearance documents stored for “9 digits or more” of the unit “bps” is obtained from the numerical range distribution data storage unit 2300. That is, the numerical importance calculation unit 1300 adds “9” “14” and “10” “3” to “17” as the number of appearance documents in FIG. Then, this is equivalent to df, and log (2000/17) = 2.07 corresponding to idf is calculated as the importance of this numerical condition.

  When the document evaluation value calculation unit 1400 receives a keyword and a numerical value, which are input conditions, from the word importance calculation unit 1200 and the numerical importance calculation unit 1300, the document evaluation value calculation unit 1400 refers to the document DB unit 2400, The number of occurrences (tf) of each keyword and numerical value in each is obtained. Then, the evaluation value of each document is calculated based on the number of appearances of each keyword and numerical value (tf) and the importance (idf) calculated by the word importance calculating unit 1200 and the numerical importance calculating unit 1300 ( S24).

  For example, the document evaluation value calculation unit 1400 searches the document DB unit 2400 illustrated in FIG. 5 for documents including “wireless”, “provider”, and “1 Gbps or more” as input conditions. For example, when the document 11 in FIG. 5 is taken out, in the document 11, the appearance count tf = 0 of “wireless”, tf = 2 of “provider”, tf = 0 of “1 Gbps or more = (1 × 10 ^ 9 bps or more)” Is calculated.

  Then, the document evaluation value calculation unit 1400 calculates the evaluation value Σtf · idf = (0 × 1.7) + (2 × 0.64) + (0 × 2.07) = 1.28 of the document 11. .

  Similarly, for the document 12, tf = 0 for “wireless”, tf = 2 for “provider”, and “1 Gbps or more” conform to “1-4 × 10 ^ 9 bPs”, so tf = 1 is obtained. As a result, the evaluation value Σtf · idf = (0 × 1.74) + (2 × 0.64) + (1 × 2.07) = 3.35 of the document 12 is calculated.

  Further, for the document 13, “wireless” tf = 1, “provider” tf = 2, and “1 GbPs or more” tf = 0 are obtained. As a result, the evaluation value Σtf · idf = (1 × 1.74) + (2 × 0.64) + (0 × 2.07) = 3.02 of the document 13 is calculated.

  The output unit 1500 determines a document to be output based on the evaluation value of each document calculated by the document evaluation value calculation unit 1400, and outputs the determined document (S25).

  For example, as described above, in the documents 11 to 13 in FIG. 5, there is no document that satisfies both the conditions of “wireless” and “1 Gbps or more”, but there are the document 12 and the document 13 including either one of them. The document 12 including “1 Gbps or more” in which the number of appearances is smaller (df is smaller), that is, the document is further narrowed, has a higher evaluation value than the document 13.

  Further, for example, even when the conditions “wireless”, “provider”, and “10 to 80 Mbps” are given from the input unit 1100, the same operation (the process shown in FIG. 7) is performed.

  Regarding the importance levels of “wireless” and “provider”, the word importance level calculation unit 1200 obtains the same values idf “1.74” and “0.64” as described above.

  The importance of “10 to 80 Mbps” is interpreted by the numerical importance calculation unit 1300 as “1-8 × 10 ^ 7 bps” which is a predetermined numerical expression. Therefore, the numerical importance calculation unit 1300 obtains the number of appearance documents “72” having the same unit “bps” and “7 digits” from the numerical range distribution data storage unit 2300 of FIG. As a result, the importance of “10-80 Mbps” is calculated as log (2000/72) = 1.44.

  Thereafter, the document evaluation value calculation unit 1400 obtains tf of “wireless” and “provider” in each document in the same manner as described above.

  As for “10 to 80 Mbps”, the evaluation value of each document is calculated with reference to tf for “5 × 10 ^ 7 bps” in which this condition and the number of digits match.

  As a result, Σtf · idf = (0 × 1.74) + (2 × 0.64) + (1 × 1.44) = 2.72 for document 11 and (0 × 1.74) for document 12. ) + (2 × 0.64) + (0 × 1.44) = 1.28, and for document 13, (1 × 1.74) + (2 × 0.64) + (0 × 1.44) = An evaluation value of 3.02 is obtained.

  As described above, when there is no document satisfying the search conditions of both “wireless” and “10 to 80 Mbps”, the document 13 including “wireless” having a smaller number of appearing documents and having a narrowing effect has a high evaluation value.

(A-3) Effects of the First Embodiment As described above, according to the present embodiment, the number of occurrences of data corresponding to df in the numerical search condition is determined for each fixed section (in the embodiment, for each digit). The information obtained by counting is stored in the numerical value range distribution data storage unit 2300 to obtain finite information, and this information can be created at the time of registration.

  As a result, it is possible to calculate the weight reflecting the value of the numerical search condition without counting the number of documents that match the numerical search condition at the time of the search processing, so that the processing speed related to the document search can be improved. .

  In addition, according to the present embodiment, as exemplified in the embodiment, weighting based on the narrowing of documents according to a given search condition can be performed regardless of whether the search condition is based on a word or a search condition based on a numerical value. Search results can be obtained.

(B) Second Embodiment Next, a second embodiment of the document search system, document search method, document search program, and recording medium of the present invention will be described with reference to the drawings.

  In the present embodiment, an approximate value of df that reflects the numerical range of the numerical search condition is calculated based on the statistical amount of the numerical information width in all registered documents.

(B-1) Configuration of Second Embodiment The configuration of the second embodiment corresponds to the configuration of the first embodiment shown in FIG. However, the numerical value range statistic obtained by the numerical value range distribution statistical unit 3400, the numerical value range statistic stored in the numerical numerical value range distribution data storage unit 2300, and the numerical importance calculation unit 1300 search by numerical values. The calculation method of the importance of the condition is different from that of the first embodiment.

  Therefore, the document search system according to the second embodiment will be described below using the configuration and corresponding codes of the text search system shown in FIG.

  FIG. 8 shows an example of the contents stored in the numerical range distribution data storage unit 2300 together with the functional configuration of the numerical range distribution statistical unit 3400 of the second embodiment.

  As shown in FIG. 8, the numerical value range distribution statistical unit 3400 of the second embodiment includes, as its functions, a minimum value / maximum value update unit 3401, a standard width calculation / update unit 3402, an appearance number update unit 3403, and an appearance document. It has at least a number update unit 3404 and an appearance document expected value update unit 3405.

  The numerical range distribution statistical unit 3400 receives the numerical information from the numerical extraction unit 3300, classifies it by the unit type of the numerical information, counts the number for each classification, and distributes the number of numerical information for each classification to the numerical range distribution The data is stored in the data storage unit 2300.

  The minimum value / maximum value update unit 3401 obtains the maximum value and the minimum value of the numerical information for each unit type based on the numerical information from the numerical value extraction unit 3300 and stores it in the numerical value range distribution data storage unit 2300. Is. Further, when the numerical information of the same unit type is newly given, the minimum value / maximum value update unit 3401 refers to the minimum value 2301 and the maximum value 2302 of the numerical value range distribution data storage unit 2300 and refers to the minimum value for each unit type. It is determined whether or not to update the value and the maximum value, and when it is necessary to update, the minimum value and the maximum value are updated. Here, for example, when the numerical information is an exponential numerical value, the minimum value and the maximum value may be described by taking the logarithm of the numerical value.

  When the numerical information of the unit type is given, the appearance number update unit 3403 updates the appearance number of the unit type and stores it in the appearance number 2303 of the numerical value range distribution data storage unit 2300.

  The standard width calculation / update unit 3402 calculates a standard value of the width of the numerical range of the received numerical information (hereinafter, this is referred to as a standard width), and the calculated standard width is stored in the numerical range distribution data storage unit 2300. The data is stored in the standard width 2304.

  Here, the width of the numerical value range can be converted into a predetermined value according to the user operation so as to suit the user's intuition. For example, in the case of exponential numerical information, it is expressed by the logarithm of the numerical information. It is desirable to set the width of the range.

  There are various methods for updating the standard width of the range. For example, when new numerical information is given, the standard width 2304 already stored in the numerical range distribution data storage unit 2300 is assigned to each category. It can be obtained by multiplying the number of appearances, adding the newly obtained range of numerical information, and dividing again by the number of appearances of the unit type to which the numerical information is added.

  Further, when the numerical value information having no range is given, the standard value calculation / update unit 3402 can set a minimum constant width for the numerical value. For example, when the value width is 10% or less of the value, the value width can be regarded as 10%.

  The appearance document number update unit 3404 updates the number of documents in which numerical information appears, and stores it in the number of appearance documents 2305 of the numerical value range distribution data storage unit 2300.

  The appearance document expected value update unit 3405 extracts information on the minimum value 2301, the maximum value 2302, the standard width 2304, and the number of appearance documents 2305 in the numerical value range distribution data storage unit 2300, and all registered documents based on these information The estimated value DN of the number of documents including numerical information that conforms to a numerical range of a certain width is estimated. Further, the appearance document expected value update unit 3405 stores the expected value DN in the appearance document number expected value 2306 of the numerical range distribution data storage unit 2300.

  Here, various methods can be applied to obtain the expected value DN by the appearing document expected value update unit 3405. For example, {(standard width) × (number of appearing documents)} / {(maximum value) − (minimum). Value)}.

  The numerical value range distribution data storage unit 2300 includes, for each unit type, the total number (appearance number) 2303 of the numerical information of the unit type in all input documents, the number of appearance documents 2305 including the numerical information of the unit type, and the unit The standard width (standard width) 2304 of the numerical range of the numerical value information of the seed, the maximum value 2302 and the minimum value 2301 of the numerical values appearing in the document in the unit type, and the expected number of appearance documents 2306 are stored. In the present embodiment, the standard width 2304, the minimum value 2301, and the maximum value 2302 are expressed as logarithms of numerical information.

  When the numerical importance calculation unit 1300 receives the numerical search condition from the input unit 110, the numerical importance calculation unit 1300 sets the numerical value range with a certain width in all the registered documents regarding the numerical unit types used as the search condition from the numerical range distribution data storage unit 2300. An expected value DN of the number of documents including suitable numerical information is taken out, and an expected value of the number of documents in which a numerical value serving as a search condition appears is estimated using the expected value DN.

  Here, the expected value of the number of documents in which a numerical value as a search condition appears can be estimated by (numerical value search condition width) × DN.

  Also, when the search condition is a numerical value having no range width, the numerical importance calculation unit 1300, like the numerical range distribution statistical unit 3400, has a certain ratio to the numerical value that is the search condition. For example, 10% is treated as a lower limit of the width, and a width smaller than that is treated as having a width of 10% of the value.

  As described in the first embodiment, the expected value of the number of appearance documents of the search condition calculated by the numerical importance calculating unit 1300 is treated as equivalent to df as described above in the first embodiment. The idf which is the degree can be calculated and the processing can be continued.

(B-2) Operation of Second Embodiment Hereinafter, the operation of the numerical range distribution statistical unit 3400 according to the second embodiment will be described with reference to the functional configuration diagram of the numerical range distribution statistical unit 3400 of FIG. To do.

  Hereinafter, for example, among registered documents, the document D21 includes numerical information “1.5 to 25 Mbps”, and the document D22 includes 2 to “1 to 15 Mbps” and “50 Mbps”. It is assumed that two pieces of numerical information are included. Further, a specific processing example will be described on the assumption that there is no numerical information of the unit type “bps” in other documents.

  When the numerical value information extracted by the numerical value extraction unit 3300 is given to the numerical value range distribution statistical unit 3400, the number of appearances for each unit type and the number of appearance documents of documents including the unit type are updated.

  For example, when “1.5 to 25 Mbps” in the document D 21 is given to the numerical range distribution range statistical unit 3400, it is confirmed that there is numerical information regarding the unit “bps” and stored in the numerical range distribution data storage unit 2300. In addition, “1” is added to the number of appearing documents regarding the unit “bps”. Further, “1” is added to the number of occurrences of the unit type numerical value “bps”.

  Next, the numerical value range distribution statistical unit 3400 takes the logarithm of the minimum value 1.5M and the maximum value 25M of the numerical value range to obtain the minimum value 6.18 and the maximum value 7.40. The minimum value and the maximum value of the unit “bps” stored in the numerical value range distribution data storage unit 2300 with the minimum value and the maximum value are updated, and the width between the maximum value and the minimum value (7.40-6. 18) = 1.22 is obtained, and this width is updated as the standard width.

  If this information is the first for the unit “bps”, it is written as it is. As a result, for the unit type “bps” in the numerical range distribution data storage unit 2300, the number of appearance documents “1”, the number of appearances “1”, the minimum value “6.18”, the maximum value “7.40”, the standard width “ 1.22 ".

  Next, when the document D22 is input, it is assumed that the numerical value range distribution statistical unit 3400 is first executed for numerical information “1 to 15 Mbps”.

  Since the logarithmic width of “1 to 15M” is 1.18, from the standard width 1.22 and the number of appearances 1 for the unit “bps” of the numerical value range distribution data storage unit 2300 at this time, (1. The standard width is updated at 22 × 1 + 1.18) / (1 + 1) = 1.2.

  Others operate in the same manner as before, and the number of appearing documents is incremented by 1 to “2”, the number of appearances is incremented by “1”, and the minimum value is updated to a logarithm of “1M” to become “6”.

  Next, it is assumed that numerical information “50 Mbps” in the same document D22 is processed. In this case, since the value has no width, the width is set to a logarithm of 0.04 corresponding to 10%.

  Regarding this, from the standard width 1.2 and the number of appearances 2 at this time, (1.2 × 2 + 0.04) / (2 + 1) = 0.81, and the unit type “bps” of the numerical range distribution data storage unit 2300 Update the standard width for.

  Also, the maximum value is updated with “7.70” which is the logarithm of “50M”. Since the document is the same document D22 as before, the number of appearing documents is not updated, and the number of appearances for the numerical information “bps” is incremented by 1 and updated to “3”.

  As a result of the above operation, the numerical value range distribution data storage unit 2300 stores, in the unit type “bps”, the total number of appearances of the numerical information is 3, the number of documents in which the numerical information appears is 2, and the standard width of the numerical values is The logarithm is “0.81”, the minimum value is “6” logarithm, and the maximum value is “7.70” logarithm.

  Further, from these, the expected value DN of the number of documents including numerical information that conforms to a numerical range of a certain width in all documents registered for the unit “bps” is (0.81 × 2) / (7.70-6). ) = 0.95, which is also stored in the numerical range distribution data storage unit 2300.

  Next, the operation of the numerical importance calculation unit 1300 during document search will be described.

  For example, it is assumed that the numerical value search condition captured by the input unit 1100 is “1 to 10 Mbps”.

  First, the expected value DN obtained by referring to the numerical range distribution data storage unit 2300 for the unit type “bps” of this search condition is 0.95. Therefore, the estimated value of the number of appearing documents is calculated as (7−6) × 0.95 = 0.95 from the expected value DN, the logarithm of the search condition, “6”, and the maximum value “7”. .

  Using this, as in the first embodiment, the importance of the search condition is calculated by using the estimated value of the number of appearing documents as a value corresponding to idf, and the document search process is performed.

  In the case where the numerical search condition is “20 to 25 Mbps” and the search condition is in a range narrower than “1 to 10 Mbps” described above, the same operation is performed until the number of information appearances per document is calculated.

  Then, from the minimum value “7.30” of the logarithm in the search condition and the maximum value “7.40”, an estimated value of the number of appearing documents is (7.40−7.30) × 0.95 = 0.095. And calculate.

  It has been described that the registration by the numerical range distribution statistical unit 3400 of the numerical range distribution data storage unit 2300 and the reference by the numerical importance calculation unit 1300 are performed for each unit type.

  However, as in the first embodiment, the numerical values may be classified into unit types and further classified into certain intervals of numerical values such as digits.

  By such a combination, it is possible to estimate the number of appearance documents in consideration of both the value of the search condition by the numerical value and the width of the value, and weighting of the search condition by the numerical value based on it can be performed.

(B-3) Effects of the Second Embodiment As described above, according to the present embodiment, when the input search conditions are “20 to 25 Mbps” and “1 to 10 Mbps”, a narrower range is compared. The expected value of the number of appearing documents is reduced under the search conditions, that is, those that are more likely to narrow down documents. Based on this, the numerical importance calculation unit 1300 weights the search condition, so that a higher importance is calculated for the search condition based on a narrow range of numerical values.

  Further, according to the present embodiment, the statistic necessary for estimating the number of appearing documents is calculated in advance by the numerical range distribution statistical unit 3400 at the time of document input and stored in the numerical range distribution data storage unit 2300.

  In other words, in order to calculate the statistics necessary for estimation, the search of the documents that match the search conditions and the count are not performed at the time of the search, and the numerical search conditions are considered in consideration of the possibility of narrowing down by the range of numerical values as the search conditions. The importance can be calculated.

(C) Other Embodiments In the present invention, when a document search is performed using a numerical value as a search condition, when a plurality of numerical conditions, or a numerical condition and a word search condition are set, the weight of each search condition Can be used to improve the efficiency of the process of calculating.

  Specifically, when a document is registered in the database, it is possible to perform document registration so that a statistical tendency of numerical information in the document is calculated and stored in the database. Thereby, it is possible to eliminate the need to calculate the tendency in the numerical document set at the time of retrieval. In the case of a search condition using numerical values, a search process can be performed in which the weight is changed depending on the range of the numerical values.

  As described above, the present invention is used to calculate the appearance tendency of numerical values in a document at the time of registration and store it in a database and refer to it at the time of retrieval to suppress a decrease in processing performance.

  In the second embodiment described above, the numerical range distribution statistical unit 3400 has been described as calculating the standard width, the number of appearances, the number of appearance documents, the maximum value, the minimum value, and the expected value for each word type. In combination with this embodiment, the standard width, the number of appearances, the number of appearance documents, the maximum value, the minimum value, and the expected value may be calculated for each numerical section preset for each word type.

It is a functional block diagram which shows the function structure of the document search system of 1st and 2nd embodiment. It is an operation | movement flowchart which shows the operation | movement in the document input part of 1st Embodiment. It is explanatory drawing explaining the example of a document thrown into the document insertion part of 1st Embodiment. It is explanatory drawing which shows the example of the content of the DF data storage part of 1st Embodiment. It is explanatory drawing which shows the content of the numerical range distribution data storage part of 1st Embodiment. It is explanatory drawing which shows the example of the content of the document DB part of 1st Embodiment. It is a flowchart which shows the operation | movement in the document search part of 1st Embodiment. It is a block diagram which shows the function structure of the numerical range distribution statistics part of 2nd Embodiment, and the content example of a numerical range distribution data storage part.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Document search system, 1000 ... Document search part, 1100 ... Input part, 1200 ... Word importance calculation part, 1300 ... Numerical importance calculation part, 1400 ... Document evaluation value calculation part, 1500 ... Output part, 2000 ... Document storage 2200 ... DF data storage unit, 2300 ... Numerical range distribution data storage unit, 2400 ... Textile DB unit, 3000 ... Document input unit, 3100 ... Entry word extraction unit, 3200 ... Document number counting unit, 3300 ... Numerical value extraction unit 3400 ... Numerical range distribution statistics section.

Claims (9)

A document search system for searching for a document that meets the above search condition from a registered document registered with at least numerical information as a search condition,
Numerical information classification means for classifying each of one or more numerical information extracted from the input document for each unit type of numerical information;
A feature amount calculating means for calculating a feature amount for estimating a document including the numerical information with a predetermined numerical range for each unit type;
Feature quantity storage means for storing the feature quantity calculated by the feature quantity calculation means for each unit type;
Based on the unit type of the input numerical information input as the search condition, the feature quantity of the corresponding unit type is acquired from the feature quantity storage unit, and input as the search condition based on at least the acquired feature quantity Numerical importance calculation means for calculating the weight of the input numerical information,
A document retrieval system comprising: a document retrieval unit that retrieves a document from a registered document based on the weight of the input numerical information calculated by the numerical importance calculation unit. The feature amount calculating means is
A distribution statistics unit that classifies each numerical information extracted from the input document according to a predetermined numerical section preset according to the unit type, and obtains distribution statistics of the predetermined numerical section for each unit type ,
The numerical importance calculation means calculates a weight of the input numerical information input as the search condition based on a distribution statistical result of the predetermined numerical interval corresponding to a unit type of the input numerical information. The document search system according to claim 1.   The numerical importance calculation means calculates an estimated value of the number of documents matching the input numerical information input as a search condition based on the feature quantity stored in the feature quantity storage means, and the estimated value of the document The document search system according to claim 1, wherein the weight of the input numerical information input as the search condition is calculated based on the search condition. The feature amount calculating means is
Based on each numerical information classified for each unit type, an expected value calculation unit that calculates an expected value of the number of registered documents including numerical information that matches a numerical condition having a range of a certain width,
4. The numerical importance calculation means calculates the weight of input numerical information input as the search condition based on the expected value corresponding to the unit type of the input numerical information. The document search system described in any of the above.   The expectation value calculation unit, for the numerical information in all the registered documents, appears for each unit type and / or for each predetermined numerical section, a standard width of numerical values and an appearing document in which the numerical information appears. And a numerical range in which all numerical information appears, and is constant for each unit type and / or the predetermined numerical section based on the standard width, the number of appearing documents, and the numerical range. 5. The document search system according to claim 4, wherein the number of registered documents including a numerical value in the width range is calculated as an expected value. The distribution statistics unit is characterized by processing a logarithm of the numerical value of the numerical information extracted from the input document,
The document retrieval system according to claim 2, wherein the numerical importance calculation means processes a logarithm of the numerical value of the input numerical information. A document search method for searching a document that meets the above search condition from a registered document registered with at least numerical information as a search condition,
A numerical information classification step of classifying each of one or more pieces of numerical information extracted from the input document for each unit type of numerical information;
A feature amount calculating step for calculating a feature amount for estimating a document including the numerical information with a predetermined numerical range for each unit type; and
A feature amount storage means for storing the feature amount calculated in the feature amount calculation step for each unit type;
Based on the unit type of the input numerical information input as the search condition, the feature quantity of the corresponding unit type is acquired from the feature quantity storage unit, and input as the search condition based on at least the acquired feature quantity Numerical importance calculation step for calculating the weight of input numerical information,
A document search method comprising: a document search step of searching for a document from a registered document based on the weight of the input numerical information calculated by the numerical importance calculation step. A document search program for searching a document that meets the above search condition from a registered document registered with at least numerical information as a search condition,
On the computer,
Numerical information classification means for classifying each of one or more numerical information extracted from the input document for each unit type of numerical information,
A feature amount calculating means for calculating a feature amount for estimating a document including the numerical information with a predetermined numerical range for each unit type;
Feature quantity storage means for storing the feature quantity calculated by the feature quantity calculation means for each unit type;
Based on the unit type of the input numerical information input as the search condition, the feature quantity of the corresponding unit type is acquired from the feature quantity storage unit, and input as the search condition based on at least the acquired feature quantity Numerical importance calculation means for calculating the weight of input numerical information,
A document search program for functioning as a document search means for searching a document from a registered document based on the weight of input numerical information calculated by the numerical importance calculation means. A computer-readable recording medium storing a document search program for searching for a document that matches the search condition from a registered document registered with at least numerical information as a search condition, wherein the document search program is defined in claim 7. A computer-readable recording medium having a document search program recorded thereon, which corresponds to the document search program described above.

Images