Document analysis system, document analysis method, and program

Abstract

To accurately estimate the meaning of a word included in a document, to provide information useful for analyzing the document.SOLUTION: A document analysis apparatus is configured to use an information processing apparatus comprising a processor and a storage apparatus, the document analysis apparatus being configured to: store a target document which is a document to be analyzed; estimate one or more alternative words for a functional word which is a word modifying a phrase included in the target document; estimate the meaning of the functional word on the basis of the estimated alterative words; and specify the position of the phrase in the document on the basis of the estimated meaning of the functional word. The document analysis apparatus estimates the one or more alternative words for the functional word, through rule-based processing or processing using a machine learning model. The document analysis apparatus generates information on the specified position of the phrase.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a document analysis system, a document analysis method, and a program. [Background technology]

[0002] Patent Document 1 describes how to use polysemous words that have multiple meanings in documents such as proposals and specifications. The document describes a word sense identification device configured to identify the meanings of polysemous words included in text data when extracting them. The word sense identification device receives information including words with multiple meanings and explanations of each meaning of the words, generates co-occurrence information of other words with each meaning of the words included in the information based on the received information, and generates information that associates the generated co-occurrence information with the meanings of the words included in the information. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-12468 Summary of the Invention [Problem to be solved by the invention]

[0004] Words with multiple meanings, such as polysemous words, can have different meanings (functions, actions) depending on the document in which they are used. For example, the particle "yori" (polysemous word) means "cause" in the document "If you are unable to participate due to unavoidable reasons," "means" in the document "Procedures will be carried out by electronic application," and "starting point" in the document "Starting at 10:00 AM." In this case, if the meaning of the polysemous word "yori" can be accurately estimated, it will be possible to identify the position of the phrases "unavoidable reasons," "electronic application," and "10:00 AM" in each document, and analyze the structure of the document.

[0005] The system described in Patent Document 1 identifies the meaning of words contained in a document, but it cannot identify the meaning of the word in the document or the position of the phrase to which the word belongs. Furthermore, the system described in Patent Document 1 requires the preparation of a huge amount of information such as the meanings of sembling words and the relationships and co-occurrence rates of co-occurring words, which requires a great deal of effort and cost to implement.

[0006] The present invention aims to provide a document analysis system, a document analysis method, and a program that can accurately estimate the meaning of words written in a document and provide information useful for document analysis. [Means for solving the problem]

[0007] One aspect of the present invention for achieving the above object is a document analysis system configured using an information processing device having a processor and a storage device, wherein the information processing device stores a target document that is a document to be analyzed, an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document, and function word meaning estimation rules that are information that associates, for each of the function words, two or more meanings with estimated reference words that are alternative words corresponding to each of the two or more meanings, and estimates one or more alternative words for the function words using the alternative word estimation model, If any of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, the meaning associated with the inferred reference word is inferred to be the meaning of the function word; if none of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, the meaning of the function word is not inferred; The estimated meaning of the function word is identified as the position of the phrase in the document. Another aspect of the present invention is a document analysis system configured using an information processing device having a processor and a storage device, wherein the information processing device stores a target document that is a document to be analyzed, an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document, and a function word meaning estimation model that is a machine learning model that has learned data that associates each meaning with one or more of the alternative words as training data, and that inputs the alternative words and outputs their meanings, and estimates one or more alternative words for the function words using the alternative word estimation model, and One or moreThe meaning of the function word is inferred by inputting the alternative words into the function word meaning inference model, and the inferred meaning of the function word is identified as the position of the phrase in the document.

[0008] Other problems and solutions disclosed in the present application will be made clear in the detailed description and drawings. [Effects of the Invention]

[0009] According to the present invention, it is possible to accurately estimate the meaning of words written in a document and provide information useful for analyzing the document. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an outline of a document analysis system. [Figure 2] FIG. 10 is a diagram illustrating a function word selection process. [Figure 3] FIG. 10 is a diagram illustrating an alternative word estimation process. [Figure 4] FIG. 10 is a diagram illustrating a function word meaning estimation process. [Figure 5A] 1 is an example of a function word meaning inference rule used in a function word meaning inference process (rule base). [Figure 5B] This is an example of learning data used in function word meaning estimation processing (machine learning based). [Figure 5C] FIG. 10 is a diagram showing an example of inferring the meaning of a function word. [Figure 6] 10 is an example of phrase positioning management information. [Figure 7] 10 is a flowchart illustrating a document analysis process. [Figure 8A] 10 is a flowchart illustrating a function word meaning estimation process (rule-based). [Figure 8B] 10 is a flowchart illustrating a function word meaning estimation process (machine learning based). [Figure 9A] 10 is an example of a function word meaning estimation rule registration screen. [Figure 9B] 10 is an example of a function word meaning estimation rule management screen. [Figure 10A] 10 is an example of a learning data registration screen. [Figure 10B] 10 is an example of a learning data management screen. [Figure 11] 1 is an example of a hardware configuration of an information processing device used to realize a document analysis system. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, an embodiment of the present invention will be described with reference to the drawings as appropriate.

[0012] In the following description, duplicated descriptions of identical or similar components may be omitted. Also, in the following description, the letter "S" added before a reference numeral denotes a processing step.

[0013] In the following description, "document" refers to information written using characters (such as a sentence or document). A "document" is managed as data written in a predetermined data format that can be read by an information processing device (such as text data or various data formats that can be handled by word processing application software).

[0014] Also, "phrase" means a series of words (phrases, idioms, expressions, or a group of one or more words) that express a unified meaning.

[0015] FIG. 1 is a system flow diagram showing an outline of an information processing system (hereinafter referred to as a "document analysis system 1") according to one embodiment of the present invention.

[0016] The document analysis system 1 selects words related to a specific phrase (hereinafter referred to as a "target phrase 112") written in a document to be analyzed (hereinafter referred to as a "target document 111"), estimates the meaning (function, effect) of the selected word (hereinafter referred to as a "function word") in the target document 111, and identifies the position of the target phrase 112 in the target document 111 based on the estimated meaning.

[0017] The document analysis system 1 uses, for example, words that indicate grammatical functions (particles, auxiliary verbs, conjunctions, etc.). For example, if the target document 111 is "If you cannot participate due to unavoidable reasons" and the target phrase 112 is "unavoidable reasons," the document analysis system 1 selects "yori" as a function word and infers the meaning of the function word. Then, based on the inferred meaning of the function word, the document analysis system 1 identifies the position of the target phrase 112 "unavoidable reasons" in the target document 111 "If you cannot participate due to unavoidable reasons."

[0018] The document analysis system 1 is expected to be widely used for a variety of purposes and applications. For example, in a product manufacturing department, when researching laws and regulations related to a product to be manufactured, the meanings of function words contained in the document in the laws and regulations may be inferred, the positions of specific phrases in the document may be identified using the inferred meanings of the function words, structured data may be generated using the identified positions as clues, and relevant sections of the laws and regulations related to the product may be extracted. Another example is when the meanings of function words contained in a contract may be inferred, the positions of specific phrases in the contract may be identified using the inferred meanings of the function words, structured data may be generated using the identified positions as clues, and the data may be used to search and manage contracts.

[0019] As shown in the figure, the document analysis system 1 has the functions of a target selection unit 123, a function word selection unit 125, an alternative word estimation unit 130, a function word meaning estimation unit 135, a target phrase positioning management unit 140, and a data management unit 150. The document analysis system 1 also stores information (data) on a target document 111, a target phrase 112, a function word selected sentence 114, an alternative word estimation model 115, an alternative word estimation list 116, a function word meaning estimation rule 117, a function word meaning estimation model 118, and function word meanings 119.

[0020] Of the above functions, the target selection unit 123 selects the target phrase 112 from the target document 111. The target selection unit 123 selects the target phrase 112 from the target document 111 by, for example, performing natural language processing or rule-based processing. The target selection unit 123 also selects the target phrase 112 by presenting the target document 111 to the user via a user interface and accepting designation of the target phrase 112 from the user.

[0021] The function word selection unit 125 shown in FIG. 1 selects function words from the target document 111, and generates the target document 111 with information that can identify the selected function words added thereto as a function word selected sentence 114.

[0022] 2 is a diagram illustrating the process (hereinafter referred to as "function word selection process S211") performed by the function word selection unit 125. In the example shown in the figure, the function word selection unit 125 selects the particle "yori" (from) in the target phrase 112 "unavoidable reason" as a function word from the target document 111 "If you cannot participate due to unavoidable reasons," and generates the target document 111 with information that can identify the selected function word added as a function word-selected sentence 114.

[0023] In addition, the figure shows two examples of a function word selected sentence 114 in which the particle "from" which is a function word in the target document 111 is highlighted (displayed with the symbol "[ ]" in the figure), and a function word selected sentence 114 in which the particle "from" in the target document 111 is replaced with the symbol "*" (it may be replaced with a symbol other than "*" or a space). The latter function word selected sentence 114 is used, for example, when BERT (Bidirectional Encoder Representations from Transformers) is used as the alternative word estimation model 115 in the alternative word estimation process S212 described later. I can.

[0024] In the example shown in the figure, "Yameinaenai riyuu" is linked to "ni" and then linked to "ni" by "yori". When such multiple levels of dependency exist and function words are consecutive, the function word selection unit 125 selects the function word "yori" several levels ahead. Such dependency relationships (structures) can be obtained using existing software that performs morphological analysis and dependency analysis, for example. When the target document 111 is in Japanese, examples of software for performing morphological analysis include "MeCab" and "JUMAN," and examples of software for performing dependency analysis include open source software such as "GiNZA" and "CaboCha."

[0025] The alternative word estimation unit 130 shown in Fig. 1 estimates alternative words for the function words identified in the function word selected sentence 114, and generates an alternative word estimation list 116 listing the estimated alternative words. The alternative word estimation unit 130 performs the estimation by using, for example, a dictionary prepared in advance or a machine learning model (BERT, etc.). Note that when a function word is a polysemous word, it has been confirmed that there is a tendency for words estimated as alternative words to differ depending on the meaning of the polysemous word.

[0026] 3 is a diagram illustrating the process performed by the substitute word estimation unit 130 (hereinafter referred to as "substitute word estimation process S212"). The diagram illustrates a case where BERT is used as the substitute word estimation model 115. In this example, the substitute word estimation unit 130 inputs the function word selected sentence 114 "When you cannot participate due to unavoidable reasons" into the substitute word estimation model 115, and generates a substitute word estimation list 116 including the substitute words "'Therefore', 'Depending on', 'More than', 'About', 'Contrary to', 'Regarding', 'Necessary', 'With', 'Contrary to', 'Regarding', 'Keep', 'Appear', 'Concerning', 'Appear', 'Appear', 'Instead', 'Become', 'In addition', 'According to', 'By', 'Instead', ~" output by the substitute word estimation model 115.

[0027] The function word meaning inference unit 135 shown in FIG. 1 infers (determines, acquires) the meaning 119 of a function word based on the substitute word inference list 116 acquired by the substitute word inference unit 130.

[0028] 4 is a diagram illustrating the process (hereinafter referred to as "function word meaning estimation process S213") performed by the function word meaning estimation unit 135. The function word meaning estimation unit 135 estimates the meaning of a function word based on the alternative word estimation list 116, for example, by a rule-based mechanism using function word meaning estimation rules 117 or a machine learning-based mechanism using a function word meaning estimation model 118.

[0029] In the example shown in the figure, the function word meaning inference unit 135 selects the alternative word inference list 116 "'yorette', 'yo'" Based on the following words, the meaning of the functional word "yori" is presumed to be "cause."

[0030] 5A shows an example of a function word meaning estimation rule 117 used in the rule-based function word meaning estimation process S213. The illustrated function word meaning estimation rule 117 has items of a function word 1171, a meaning 1172, and an inferred reference word 1173.

[0031] Of the above items, the function word 1171 stores a function word. The meaning 1172 stores the meaning of the function word that is the estimation result. The estimated reference word 1173 stores one or more alternative words (hereinafter referred to as "estimated reference words") used to estimate whether or not the meaning corresponds. For example, if at least one estimated reference word is included in the alternative word estimation list 116, which is a list of alternative words estimated for the function word 1171, the function word meaning estimation unit 135 estimates the content of the meaning 1172 corresponding to the estimated reference word as the meaning of the function word. Note that the method of estimating the meaning of a function word is not limited to the example given here.

[0032] 5B shows an example of data (hereinafter referred to as "training data 118a") used to train the function word meaning estimation model 118 used in the machine learning-based function word meaning estimation process S213. As shown in the figure, the illustrated training data 118a has items of meaning 1181a and candidate words 1181b.

[0033] Among the above items, the meaning 1181a stores the meaning that is the estimation result. The candidate word 1181b stores one or more words (words that may be included in the alternative word estimation list 116; hereinafter, referred to as "candidate words") that are used to estimate whether or not the meaning corresponds. Note that the type of machine learning is not necessarily limited, but for example, DNN (Deep Neural Network) , an RNN (Recurrent Neural Network) can be used.

[0034] 5C shows several examples in which the function word meaning estimation unit 135 estimates the meaning 119 of a function word (the meaning of "from") based on the alternative word estimation list 116. For example, in the example of case ID "1" shown in the figure, the alternative word estimation list 116 includes estimated reference words 1173 such as "from," "afterward," and "onward," so the function word meaning estimation unit 135 estimates the meaning 119 of the function word to be "starting point." In addition, in the example of case ID "2," the alternative word estimation list 116 includes estimated reference words 1173 such as "from," "requirement," and "cause," so the function word meaning estimation unit 135 estimates the meaning 119 of the function word to be "cause." In addition, in the example of case ID "3," the alternative word estimation list 116 includes estimated reference words 1173 such as "from" and "use," so the function word meaning estimation unit 135 estimates the meaning 119 of the function word to be "means."

[0035] 1 identifies the position of the target phrase 112 based on the meaning 119 of the function word estimated by the function word meaning estimation unit 135. For example, the target phrase positioning management unit 140 identifies the meaning 119 of the function word estimated by the function word meaning estimation unit 135 as the position of the target phrase 112. The target phrase positioning management unit 140 manages (stores) the identified positioning and information generated based on the positioning as phrase positioning management information 120.

[0036] 6 is an example of phrase positioning management information 120. As shown in the figure, the illustrated phrase positioning management information 120 is made up of one or more records each having the following items: document ID 1121, position in document 1122, meaning of function word 1123, relation source phrase 1124, and relation destination phrase 1125. One record of the phrase positioning management information 120 corresponds to one of the relation source phrases 1124 (target phrases).

[0037] Among the above items, the document ID 1121 stores a document ID that is an identifier of the document (hereinafter referred to as the "extraction source document") from which the target phrase (related source phrase 1124) was extracted. The position in document 1122 stores information indicating the position of the target phrase in the extraction source document (information identified by a page number, line number, etc.). The function word meaning 1123 stores information indicating the position of the target phrase identified by the target phrase positioning management unit 140 (such as "cause," "means," "purpose," or "reason"). In this example, the function word meaning 1123 stores the function word meaning 119 estimated in the function word meaning estimation process S213. The related source phrase 1124 stores the target phrase. The related destination phrase 1125 stores another phrase (hereinafter referred to as the "related destination phrase") described in the extraction source document that has a relationship with the target phrase in the extraction source document according to the function word meaning 1123. The document analysis system 1 identifies the above relationship by, for example, natural language processing (such as syntax analysis).

[0038] The phrase positioning management information 120 shown in the figure can be used, for example, to structure the target document 111.

[0039] 1 manages various types of information (function word meaning estimation rules 117, function word meaning estimation models 118, etc.) handled in the document analysis system 1. Details of the data management unit 150 will be described later.

[0040] Next, the processing performed by the document analysis system 1 will be described with reference to a flowchart.

[0041] 7 is a flowchart for explaining the process (hereinafter referred to as "document analysis process S700") that the document analysis system 1 performs when analyzing the target document 111. The document analysis process S700 will be explained below with reference to this figure.

[0042] First, the target selection unit 123 selects one target phrase 112 from the phrases contained in the target document 111 (S711).

[0043] Next, the function word selection unit 125 selects a function word for the currently selected target phrase 112, and generates and stores a function word selected sentence 114 (S712).

[0044] Next, the document analysis system 1 determines whether the selected function word is polysemous (S713). The document analysis system 1 determines that the selected function word is polysemous by, for example, determining whether the selected function word is defined as a function word 1171 in the function word meaning estimation rules 117. If the document analysis system 1 determines that the selected function word is polysemous (S713: YES), the process proceeds to S714. If the document analysis system 1 determines that the selected function word is not polysemous (S713: NO), the process proceeds to S719.

[0045] In S714, the alternative word estimation unit 130 estimates alternative words for the selected function word and generates the alternative word estimation list 116.

[0046] Next, the function word meaning inference unit 135 performs a process (hereinafter referred to as "function word meaning inference process S715") to infer the meaning of the function word based on the substitute word inference list 116. The function word meaning inference process S715 will be described in detail later.

[0047] After executing the function word meaning estimation process S715, the document analysis system 1 determines whether the return value of the function word meaning estimation process S715 is "impossible to estimate" (S716). If the document analysis system 1 determines that the return value is "impossible to estimate" (S716: YES), the process proceeds to S719. In this case, the document analysis system 1 may output information indicating that the function word selected in S712 is impossible to estimate to notify the user, etc. If the document analysis system 1 determines that the return value is not "impossible to estimate" (S716: NO), the process proceeds to S717.

[0048] In S717, the target phrase positioning management unit 140 acquires the return value of the function word meaning estimation process S715 as the meaning of the function word selected in S712, and identifies the positioning of the target phrase 112 based on the acquired meaning.

[0049] Next, the target phrase positioning management unit 140 manages (stores) the positioning of the identified target phrase 112 and the information generated based on the positioning of the target phrase 112 as phrase positioning management information 120 (S718).

[0050] In S719, the document analysis system 1 determines whether all of the target phrases 112 have been selected from the target document 111. If the document analysis system 1 determines that all of the target phrases 112 have been selected from the target document 111 (S719: YES), the document analysis process S700 ends. If the document analysis system 1 determines that all of the target phrases 112 have not been selected from the target document 111 (there are unselected target phrases 112) (S719: NO), the process returns to S711.

[0051] FIG. 8A shows an example of the function word meaning estimation process S715 of FIG. 7 (rule-based estimation of the function word meaning). 1 is a flowchart for explaining the process of the above-mentioned method (when performed on a PC). The following explanation will be given with reference to this figure.

[0052] First, the function word meaning estimation unit 135 selects one word from the substitute word estimation list 116 (S811).

[0053] Next, the function word meaning estimation unit 135 determines whether or not a record that includes the word selected in S811 as the inferred reference word 1173 exists among the records of the function word selected in S712 of the function word meaning estimation rules 117 (S812). If the function word meaning estimation unit 135 determines that the record exists (S812: YES), the process proceeds to S813. If the function word meaning estimation unit 135 determines that the record does not exist (S812: NO), the process proceeds to S814.

[0054] In S813, the function word meaning inference unit 135 sets the content of the meaning 1172 of the record that includes the selected word as an inferred reference word as the meaning of the function word selected in S712 as a return value, and ends the function word meaning inference process S715. Thereafter, the process proceeds to S716 in FIG.

[0055] In S814, the function word meaning estimation unit 135 determines whether or not all the words listed in the alternative word estimation list 116 have been selected. If the function word meaning estimation unit 135 determines that all the words in the alternative word estimation list 116 have been selected (S814: YES), the process proceeds to S815. If the function word meaning estimation unit 135 determines that all the words in the alternative word estimation list 116 have not been selected (S814: NO), the process returns to S811.

[0056] In S815, the function word meaning inference unit 135 sets "inference impossible" as a return value and ends the function word meaning inference process S715. After that, the process proceeds to S716 in FIG.

[0057] 8B is a flowchart illustrating another example of the function word meaning estimation process S715 in FIG. 7 (when the meaning of the function word is estimated based on machine learning). The following description will be given with reference to this figure.

[0058] First, the function word meaning estimation unit 135 inputs the contents of the substitute word estimation list 116 to the function word meaning estimation model 118, and obtains the output of the function word meaning estimation model 118 as the meaning of the function word selected in S712 (S851).

[0059] Next, the function word meaning estimation unit 135 sets the meaning acquired in S851 as the meaning of the function word selected in S712 as a return value (S852). After that, the process proceeds to S716 in Fig. 7. Note that, for example, if the estimation accuracy by the function word meaning estimation model 118 is lower than a preset threshold, the function word meaning estimation unit 135 may set the return value to "impossible to estimate."

[0060] <Data management function> The data management unit 150 provides a user interface for managing (registering, editing, deleting, searching, etc.) various information handled in the document analysis system 1 (function word meaning estimation rules 117, function word meaning estimation models 118, etc.).

[0061] 9A is an example of a screen (hereinafter referred to as a "function word meaning estimation rule registration screen 910") that is displayed when the data management unit 150 receives from the user the contents of the function word meaning estimation rule 117. As shown in the figure, the function word meaning estimation rule registration screen 910 has a function word input field 911, a meaning input field 912, an estimation reference word input field 913, a register button 915, and a cancel button 916.

[0062] The user inputs a function word input field 911, a meaning input field 912, and an estimated reference word input field 91 When the user inputs the content into each input field and operates the register button 915, the data management unit 150 registers the content input into each input field as the function word meaning estimation rule 117.

[0063] 9B is an example of a screen (hereinafter referred to as a "function word meaning estimation rule management screen 920") that the data management unit 150 displays when the user manages the contents of the function word meaning estimation rules 117. As shown in the figure, the function word meaning estimation rule management screen 920 has a function word input field 921, a display button 922, a content display field 930, an add button 941, and a back button 942. Furthermore, the content display field 930 has a meaning display field 931, an inferred reference word display field 932, and a delete button 933.

[0064] When the user inputs a function word in the function word input field 921 and operates the display button 922, the data management unit 150 displays the registered contents of the function word meaning estimation rule 117 corresponding to the input function word in the content display field 930. By referring to the content display field 930, the user can easily check the contents of the function word meaning estimation rule 117.

[0065] When the user operates the delete button 933, the data management unit 150 deletes (or invalidates) the registration contents that are no longer necessary from the function word meaning estimation rules 117.

[0066] When the user operates the add button 941, the data management unit 150 displays a function word meaning estimation rule registration screen 910 shown in Fig. 9A (a function word meaning estimation rule registration screen 910 in which the content of the function word input field 921 is displayed in a function word input field 911). By using the displayed function word meaning estimation rule registration screen 910, the user can efficiently register a new function word meaning estimation rule to the function word meaning estimation rules 117.

[0067] 10A shows an example of a screen (hereinafter referred to as a "learning data registration screen 1010") that is displayed when the data management unit 150 receives the contents of the learning data 118a from the user. As shown in the figure, the learning data registration screen 1010 has a meaning input field 1011, a candidate word input field 1012, a registration button 1015, and a cancel button 1016.

[0068] When the user inputs content into the meaning input field 1011 and the candidate word input field 1012 and operates the register button 1015, the data management unit 150 registers the input content as learning data 118a.

[0069] 10B shows an example of a screen (hereinafter referred to as a "learning data management screen 1020") that the data management unit 150 displays when the user manages the contents of the learning data 118a. As shown in the figure, the learning data management screen 1020 has a learning data display field 1030, an add button 1041, and a back button 1042. The learning data display field 1030 also has a meaning display field 1031, a candidate word display field 1032, an edit button 1033, and a delete button 1034, which are provided for each learning data 118a (each record shown in the figure).

[0070] When the user operates the edit button 1033 of a record in the learning data display field 1030, the data management unit 150 displays a learning data registration screen 1010 that describes the contents of the record. The user can efficiently edit the contents of the learning data 118a using the displayed learning data registration screen 1010. For example, if there is an error in estimating the meaning of a function word, the user can use this function to edit the learning data 118a (e.g., add new learning data that enables the correct meaning to be estimated). In addition, the user can get an overview of past candidate words and their selected meanings, which can be useful for considering adding new meanings, etc.

[0071] When the user operates the delete button 1034 of a record in the display field 1030 of the learning data, the data management unit 150 deletes (or invalidates) the learning data of that record from the learning data 118a.

[0072] When the user operates the add button 1041, the data management unit 150 displays a learning data registration screen 1010. The user can use the displayed learning data registration screen 1010 to efficiently register (add) learning data to the learning data 118a.

[0073] <Summary> As described above in detail, the document analysis system 1 of this embodiment can infer alternative words for function words and accurately infer the meaning of the function words based on the inferred alternative words (alternative word estimation list 116) using rules or machine learning. Furthermore, the document analysis system 1 identifies the position of the target phrase 112 in a document and provides the user with the identified position and information based on that position, allowing the user to efficiently obtain information useful for document analysis. Furthermore, there is no need to prepare a vast amount of information, such as a dictionary of synonyms, in advance to infer the meaning of a function word, and the document analysis system 1 of this embodiment reduces the workload and cost burden on the user when implemented.

[0074] <Hardware> FIG. 11 shows an example of the hardware configuration of an information processing device (computer) used to realize the document analysis system 1 described above.

[0075] The illustrated information processing device 10 includes a processor 11, a main storage device 12 (memory), an auxiliary storage device 13 (external storage device), an input device 14, an output device 15, and a communication device 16. These are communicatively connected via a bus, a communication cable, etc. Examples of the information processing device 10 include a personal computer, a server device, a smartphone, a tablet, an office computer, a general-purpose machine (mainframe), etc.

[0076] The information processing device 10 may be realized, in whole or in part, by using virtual information processing resources provided using virtualization technology, process space separation technology, or the like, such as a virtual server provided by a cloud system. Also, all or in part of the functions provided by the information processing device 10 may be realized, for example, by a service provided by a cloud system via an API (Application Programming Interface), or the like. All or part of the functions provided by the information processing device 10 may be provided as, for example, SaaS (Software as a Service), PaaS (Platform as a Service), IaaS (Infrastructure as a Service), etc. It may be realized by using the Service as a Service.

[0077] The processor 11 may be, for example, a central processing unit (CPU), a micro processing unit (MPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), or the like. It is composed of chips such as Gate Array (Gate Array), ASIC (Application Specific Integrated Circuit), and AI (Artificial Intelligence) chips.

[0078] The main memory device 12 is a device used by the processor 11 when executing a program, and is, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), or a non-volatile memory (NVRAM (Non Volatile RAM)). The various functions realized in the document analysis system 1 are realized by each processor 11 reading out programs and data stored in the auxiliary memory device 13 into the main memory device 12 and executing them.

[0079] The auxiliary storage device 13 is a device for storing programs and data, and may be, for example, an SSD (Solid State Drive), a hard disk drive, or an optical storage device (CD (Compact Disc) The auxiliary storage device 13 can be configured with a non-transitory storage medium read / write device such as a digital video card (Digital Versatile Disc), a DVD (Digital Versatile Disc), a storage system, an IC card, an SD card, or an optical storage medium, a non-transitory storage area of a cloud server, etc. Programs and data can be read into the auxiliary storage device 13 from other information processing devices equipped with non-transitory storage media or non-transitory storage devices via a storage medium reader or a communication device 16. The programs and data stored in the auxiliary storage device 13 are read into the main storage device 12 as needed.

[0080] The input device 14 is an interface that accepts input of information from the outside, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a pen-input tablet, a voice input device, or the like.

[0081] The output device 15 is an interface that outputs various information such as the progress of processing and the results of processing to the outside. The output device 15 is, for example, a display device (liquid crystal monitor, LCD (Liquid Crystal Display), graphic card, etc.) that visualizes the various information described above, a device that converts the various information described above into audio (audio output device (speaker, etc.)), or a device that converts the various information described above into text (printer, etc.). Note that, for example, the information processing device 10 may be configured to input and output information to and from other devices via the communication device 16.

[0082] The input device 14 and the output device 15 constitute a user interface that realizes interactive processing with the user (receiving information, providing information, etc.).

[0083] The communication device 16 is a device that realizes communication with other devices. The communication device 16 is a wired or wireless communication interface that realizes communication with other devices via the communication medium 5, and is, for example, a NIC (Network Interface Card), a wireless communication module, a USB module, or the like.

[0084] The information processing device 10 may be equipped with, for example, an operating system, a file system, a DBMS (DataBase Management System) (relational database, NoSQL, etc.), a KVS (Key-Value Store), etc.

[0085] Although the embodiments have been described above, the present invention is not limited to the above-described embodiments, and various modifications are included, and the present invention is not necessarily limited to those including all of the configurations described. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0086] For example, the type of written language of the target document 111 is not necessarily limited, and the target document 111 may be written in another language such as English. If the written language of the target document 111 is, for example, English, words are already separated by spaces, so the process of detecting word boundaries can be omitted. Furthermore, software exists for performing morphological analysis and dependency analysis on documents written in English, just as it does for Japanese, so the mechanism of the present invention can be realized by utilizing such software. [Explanation of symbols]

[0087] 1 Document analysis system, 10 Information processing device, 11 Processor, 12 Main memory device, 111 Target document, 112 Target phrase, 114 Function word selected sentence, 115 Alternative word estimation model, 116 Alternative word estimation list, 117 Function word meaning estimation rule, 118 Function word meaning estimation model, 118a Training data, 119 Function word meaning, 120 Phrase positioning management information, 123 Target selection unit, 125 Function word selection unit, 130 Alternative word estimation unit, 135 Function word meaning estimation unit, 140 Target phrase positioning management unit, 150 Data management unit, S211 function word selection processing, S212 alternative word inference processing, S213 function word meaning inference processing, S700 document analysis processing, S715 function word meaning inference processing, 910 function word meaning inference rule registration screen, 920 function word meaning inference rule management screen, 1010 learning data registration screen, 1020 learning data management screen

Claims

1. The information processing device includes a processor and a storage device, The information processing device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation rule, which is information associating two or more meanings with estimation reference words that are alternative words corresponding to each of the two or more meanings, for each of the function words; Remember, predicting one or more alternative words for the function words using the alternative word prediction model; If any of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, the meaning associated with the inferred reference word is inferred to be the meaning of the function word; if none of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, the meaning of the function word is not inferred; Identifying the estimated meaning of the function word as the position of the phrase in the document; Document analysis system.

2. The information processing device includes a processor and a storage device, The information processing device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation model that is trained using data that associates each meaning with one or more of the alternative words as training data, and that inputs the alternative words and outputs the meaning; Remember, predicting one or more alternative words for the function words using the alternative word prediction model; Inferring the meaning of the function word by inputting the one or more estimated alternative words into the function word meaning inference model; Identifying the estimated meaning of the function word as the position of the phrase in the document; Document analysis system.

3. 3. The document analysis system according to claim 1, generating information regarding the identified positioning of the phrase; Document analysis system.

4. 4. The document analysis system according to claim 3, The information regarding the positioning is information indicating other phrases in the target document that have a relationship with the phrase in terms of the meaning of the function word. Document analysis system.

5. 2. The document analysis system of claim 1, providing a user interface for editing the function word meaning inference rules; Document analysis system.

6. 3. The document analysis system according to claim 2, providing a user interface for editing the training data; Document analysis system.

7. 3. The document analysis system according to claim 1, The alternative word prediction model is BERT (Bidirectional Encoder Representations from Transformers), Document analysis system.

8. An information processing device having a processor and a storage device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation rule, which is information associating two or more meanings with estimation reference words that are alternative words corresponding to each of the two or more meanings, for each of the function words; storing the predicting one or more alternative words for the function words using the alternative word prediction model; If any of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, inferring that the meaning associated with the inferred reference word is the meaning of the function word, and if none of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, not inferring the meaning of the function word; and identifying the estimated meaning of the function word as the position of the phrase in the document; ,document analysis methods.

9. An information processing device having a processor and a storage device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation model that is trained using data that associates each meaning with one or more of the alternative words as training data, and that inputs the alternative words and outputs the meaning; storing the predicting one or more alternative words for the function words using the alternative word prediction model; Inferring the meaning of the function word by inputting the one or more estimated alternative words into the function word meaning inference model; and identifying the estimated meaning of the function word as the position of the phrase in the document; ,document analysis methods.

10. 10. A document analysis method according to claim 8 or 9, generating information about the identified position of the phrase by the information processing device; The document analysis method further comprises:

11. 11. A document analysis method according to claim 10, The information regarding the positioning is information indicating other phrases in the target document that have a relationship with the phrase in terms of the meaning of the function word. Document analysis methods.

12. An information processing device having a processor and a storage device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation rule, which is information associating two or more meanings with estimation reference words that are alternative words corresponding to each of the two or more meanings, for each of the function words; The function to remember a function of estimating one or more alternative words for the function word using the alternative word estimation model; a function of inferring the meaning of the function word when any of the one or more alternative words inferred for the function word exists in the function word meaning inference rule as the inferred reference word for the function word, and not inferring the meaning of the function word when any of the one or more alternative words inferred for the function word does not exist in the function word meaning inference rule as the inferred reference word for the function word; and a function of identifying the estimated meaning of the function word as the position of the phrase in the document; A program to achieve this.

13. An information processing device having a processor and a storage device, a target document, which is the document to be analyzed; an alternative word estimation model that is a machine learning model that estimates alternative words for function words that are polysemous words related to phrases described in the target document; and a function word meaning estimation model that is trained using data that associates each meaning with one or more of the alternative words as training data, and that inputs the alternative words and outputs the meaning; The function to remember a function of estimating one or more alternative words for the function word using the alternative word estimation model; a function of inferring the meaning of the function word by inputting the one or more inferred alternative words into the function word meaning inference model; and a function of identifying the estimated meaning of the function word as the position of the phrase in the document; A program to achieve this.

Images