Information processing device and information processing method

Abstract

To decrease manhours of definition of word pairs.SOLUTION: An information processing device has: a word pair generation part which generates a word pair between categories meeting predetermined conditions from extracted words; a relationship estimation part which accumulates an input history of search keywords, and estimates relationship between words to be acquired from the input history of search keywords; a filtering part which filters a candidate for a word pair to be registered in a word pair initial dictionary based upon the relationship; and an initial dictionary registration part which presents the candidate for the word pair to the outside, and registers, based on a result of a determination on whether to register in the word pair initial dictionary, a candidate for a word pair determined to be registered in the word pair initial dictionary.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present invention relates to an information processing apparatus and an information processing method, and is suitably applicable to, for example, an information processing apparatus and an information processing method related to a technique for assisting the creation of a dictionary used in natural language processing.

Background Art

[0002] In recent years, the use of machine learning has been progressing in various fields. One of such usage forms of machine learning is a document analysis technique that extracts various words from a large number of documents, learns the relevance between the extracted words, and analyzes the structure of the documents. For example, Patent Document 1 discloses a method of parsing the syntax of an input document in sentence units and extracting word pairs according to rules that define the relationship between words belonging to two categories.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the technique described in Patent Document 1, the rules for defining the relationships between all categories to which the words included in the document belong are defined by associating the distance between the nodes of the syntax tree and the categories to which the words belong. Therefore, with the technique described in Patent Document 1, it is not possible to extract word pairs based on the relationships between desired categories and create an initial dictionary of word pairs. To achieve this, it was necessary to manually define all word pairs.

[0005] The present invention has been made in consideration of the above points, and aims to propose an information processing apparatus and an information processing method capable of reducing the man-hours for defining word pairs.

Means for Solving the Problems

[0006] In order to solve such problems, the present invention provides an information processing apparatus for assisting in creating an initial dictionary of word pairs based on the relationship between words belonging to two categories from document information composed of one or more sentences. The information processing apparatus includes: a word extraction unit that extracts words belonging to the two categories from the document information; a search unit that determines whether the document information includes a word belonging to at least one of the two categories and corresponding to an input search keyword, and presents the document information determined to include the word to the outside; a word pair generation unit that generates word pairs between categories that satisfy predetermined conditions from the extracted words; a relationship estimation unit that accumulates the input history of the search keyword and estimates the relationship between desired words obtained from the input history of the search keyword; a filtering unit that filters candidates for word pairs to be registered in the initial dictionary of word pairs based on the relationship; an initial dictionary registration unit that presents the candidates for word pairs to the outside and, based on the result of determining whether to register the candidates for word pairs in the initial dictionary of word pairs, registers the candidates for word pairs determined to be registered in the initial dictionary of word pairs in the initial dictionary of word pairs.

[0007] Further, in the present invention, there is provided an information processing method for an information processing apparatus that supports creation of an initial dictionary of word pairs based on the relationship between words belonging to two categories from document information composed of one or more sentences. The method includes: a word extraction step in which a word extraction unit extracts words belonging to the two categories from the document information; a search step in which a search unit determines whether the document information includes a word belonging to at least one of the two categories and corresponding to an input search keyword, and presents the document information determined to include the word to the outside; a word pair generation step in which a word pair generation unit generates word pairs between categories that satisfy a predetermined condition from the extracted words; a relationship estimation step in which a filtering unit accumulates the input history of the search keyword and estimates the relationship between words desired to be obtained from the input history of the search keyword; a filtering step in which candidates for word pairs to be registered in the initial dictionary of word pairs are filtered based on the relationship; and an initial dictionary registration step in which an initial dictionary registration unit presents the candidates for word pairs to the outside and registers the candidates for word pairs determined to be registered in the initial dictionary of word pairs in the initial dictionary of word pairs.

Effect of the Invention

[0008] According to the present invention, the man-hours for defining word pairs can be reduced.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Embodiments for Carrying Out the Invention

[0010] Hereinafter, based on the drawings, an embodiment of the present invention will be described in detail. However, the present invention is not to be construed as being limited to the description of the embodiments shown below. Those skilled in the art can easily understand that the specific configuration can be changed without departing from the spirit or gist of the present invention.

[0011] In the configuration of the invention described below, the same or similar configurations or functions are denoted by the same reference numerals, and redundant descriptions are omitted. Further, in the following description, various processes may be described using expressions such as "table" and "information", but the information may be represented by data structures other than these. Also, not all of the elements and combinations thereof described in the embodiments are essential for the solution means of the invention.

[0012] FIG. 1 is a block diagram showing a configuration example of an information processing system including an information processing apparatus 1 according to the present embodiment. The information processing apparatus 1 has a function of assisting in creating a word pair initial dictionary 47 based on the relationship between words belonging to two categories from document data 44 (hereinafter, may also be simply referred to as "document information" or "document") as an example of document information composed of one or more sentences. This will be specifically described below.

[0013] The information processing system includes the information processing apparatus 1 and the client terminal 2. The information processing apparatus 1 and the client terminal 2 are communicably connected via a communication network 3. The communication network 3 is a communication means of a wired or wireless system, and is, for example, a communication means conforming to standards such as LAN (Local Area Network), WAN (Wide Area Network), WiFi (registered trademark), the Internet, a dedicated line, and a public communication network.

[0014] FIG. 2 is a block diagram showing a configuration example of a computer constituting the information processing system shown in FIG. 1. The information processing apparatus 1 and the client terminal 2 are examples of a computer, and have hardware elements such as a processor 10, a main storage device 11, an input device 12, an output device 13, an auxiliary storage device 14, and a network interface 15 (the "NW I / F" shown in the figure). Each hardware element is connected to each other via a bus 16.

[0015] Processor 10 executes a program stored in main memory device 11. Processor 10 executes processing according to the program. Main memory device 11 is a memory or the like, and stores the program executed by processor 10 and the information used by this program. Further, main memory device 11 has a work area temporarily used by the program.

[0016] Auxiliary storage device 14 is at least one of, for example, an HDD (Hard Disk Drive) and an SSD (Solid State Drive), and stores a large amount of data permanently. The program and information stored in main memory device 11 may be stored in auxiliary storage device 14. In this case, processor 10 reads the program and information from auxiliary storage device 14 and loads them into main memory device 11.

[0017] Input device 12 is an interface that receives external input, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a pen-input type tablet, a voice input device, or the like.

[0018] Output device 13 is an interface that outputs various information such as the progress of processing and the processing results. Output device 13 is, for example, a display device (LCD (Liquid Crystal Display), graphics card, etc.) that visualizes the above various information, a device (voice output device (speaker, etc.)) that converts the above various information into voice, and a device (printing device, etc.) that converts the above various information into characters. Note that, for example, information processing device 1 may be configured to input and output information to and from other devices via communication network 3.

[0019] Client terminal 2 is a terminal operated by a user when acquiring a document including necessary information from a plurality of document information. Client terminal 2 displays the operation screen presented by information processing device 1, receives the user's input, and displays the information processed by information processing device 1. Details of this operation screen will be described later.

[0020] The information processing apparatus 1 expands the word extraction processing program 20 in the main memory device 11, and the processor 10 executes the word extraction processing program 20.

[0021] FIG. 3 is a diagram showing an example of the word extraction processing program 20 and information stored in the main memory device 11. The word extraction processing program 20 includes an input reception unit 31, a category word extraction model generation unit 33, a category word extraction unit 34, an inter-category word pair extraction model generation unit 35, an inter-category word pair extraction unit 36, a search unit 37, and an initial dictionary generation unit 38 for inter-category word pairs.

[0022] The main memory device 11 further stores document data 44, a category dictionary 45, a related word dictionary 46, a word pair initial dictionary 47, an inflected word management table 48, a model management table 49, a candidate filtering pattern 50, a relevance information candidate list 51, a filtering process list 52, an extracted word / word pair database (hereinafter abbreviated as "DB") 53, and a search keyword history management table 54. These data and the like are used by the word extraction processing program 20.

[0023] The input reception unit 31 receives inputs of, for example, two categories. The input reception unit 31 acquires information input by the user on the operation screen presented by the information processing apparatus 1 via the client terminal 2. The operation screen will be described later.

[0024] The category word extraction model generation unit 33 generates a model for extracting words belonging to each category from a document using a category dictionary 45 that associates words belonging to a plurality of categories, for example, two categories, and a related word dictionary 46 that manages synonyms, antonyms, and the like.

[0025] The category word extraction unit 34 is an example of a word extraction unit, and extracts words belonging to each category from the top of the document. The category word extraction unit 34 extracts words belonging to each category from the document using the model generated by the category word extraction model generation unit 33.

[0026] The inter-category word pair extraction model generation unit 35 generates a model for extracting word pairs that belong to two categories defined in the inter-category word pair initial dictionary from the top of the document and are estimated to correspond to the relationship, using the inter-category word pair initial dictionary generated by the initial dictionary generation unit 38 described later.

[0027] The inter-category word pair extraction unit 36 extracts inter-category word pairs from the top of the document using the inter-category word pair extraction model.

[0028] The words and word pairs extracted by the category word extraction unit 34 and the inter-category word pair extraction unit 36 are stored in the extracted word and word pair DB 53 and used by other applications and the like.

[0029] The search unit 37, for example, receives an input of a search keyword, determines whether a word belonging to at least one of two categories in the document information and corresponding to the input search keyword is included, and presents the document information to the outside when it is determined that the word is included. In the above-described determination, the search unit 37 acquires two categories, a period, and a keyword for search input by the user, and performs a process (hereinafter also referred to as "search process") of extracting words that belong to the two categories respectively and are estimated to correspond to the keyword for search from the document data 44.

[0030] The initial dictionary generation unit 38 includes a category word extraction unit 34, a word pair candidate extraction unit 40, a word pair relationship acquisition unit 41, a word pair initial dictionary determination unit 42, and a candidate filtering generation registration processing unit 43. The initial dictionary generation unit 38 performs a process of generating an initial dictionary of word pairs between categories.

[0031] The category word extraction unit 34 performs named entity extraction on each of the divided elements to determine whether there is a word belonging to two categories specified by the user (hereinafter also referred to as the first category and the second category), and determines the category of the extracted word.

[0032] The word pair candidate extraction unit 40 is an example of a word pair generation unit, has a function of extracting word pairs between categories, and generates word pairs between categories that satisfy predetermined conditions from the extracted words. The word pair candidate extraction unit 40 generates candidates for word pairs between categories (hereinafter also referred to as "word pair candidates") from the words extracted according to predetermined conditions.

[0033] The word pair relationship acquisition unit 41 is an example of a relationship estimation unit, accumulates the input history of search keywords, estimates the relationship between words to be obtained from the input history of search keywords, and extracts this relationship between words as relationship candidate information. The word pair relationship acquisition unit 41 executes various processes as will be described later. For example, it acquires the keyword history information searched by the logged-in user in the past from the search keyword history management table 54 and calculates the search frequency for each pair of search keywords. Further, when there is a pair of search keywords with a frequency equal to or higher than a predetermined threshold among the frequencies of pairs of search keywords, the word pair relationship acquisition unit 41 calculates the similarity of the relationship information 135 extracted by the model name 132 stored in the model management table 49. Details of the word pair relationship acquisition unit 41 will be described later.

[0034] The candidate filtering generation registration processing unit 43 is an example of a filtering unit, and filters candidates for word pairs to be registered in the word pair initial dictionary 47 based on the above-described relationship. The candidate filtering generation registration processing unit 43 extracts elements of a sentence including variant words from the acquired document information, performs an appearance trend analysis, generates and updates a candidate filtering pattern 50.

[0035] The candidate filtering generation registration processing unit 43 has one or more candidate filtering patterns and a filtering verification process for verifying whether each candidate filtering pattern matches. The candidate filtering generation registration processing unit 43 determines whether the word pair candidate is included in the candidates of the relevant information 135. If the word pair candidate is included in the candidates of the relevant information 135, it performs the filtering verification process associated with each candidate filtering pattern. If there is a match in this filtering verification process, it adds the identification information of the candidate filtering pattern to the word pair candidate and adds it to the initial dictionary candidate list (corresponding to the initial dictionary candidate list 226 in FIG. 17 described later).

[0036] This candidate filtering pattern includes, in the document information, at least a phrase that changes the meaning of a word (for example, a phrase that constitutes a negative word, irony, etc., such as "negative", "wouldn't it", etc.).

[0037] Furthermore, this candidate filtering pattern includes the word extracted as a result of verification by the filtering verification process and the category to which it belongs.

[0038] The word pair initial dictionary determination unit 42 is an example of the initial dictionary registration unit. It presents the word pair candidate to the outside (for example, the user) and registers the word pair candidate determined to be registered in the word pair initial dictionary 47 (for example, by the user) in the word pair initial dictionary 47. Also, the word pair initial dictionary determination unit 42 acquires one word pair candidate from the word pair candidates and determines whether each word of the word pair candidate is included in the relational candidate information extracted by the word pair relational acquisition unit 41.

[0039] Next, the data group used in this embodiment will be described. The document data 44 is the document information to be processed. The document data 44 has an identifier that uniquely identifies the document data, a title of the document data 44, storage location information of the document data 44, date and time information when the document data was stored, and the text of the document data 44.

[0040] As shown in FIG. 4, the category dictionary 45 has a category name (e.g., field information, etc.) 101 designed by the user, words 102 handled in each category, and a category-specific tag 103 assigned to words that the user who designed the category determined to be category-specific.

[0041] As shown in FIG. 5, the related word dictionary 46 has a category 111, words 112 handled in each category, synonyms 113 of the words 112, antonyms of the words 112, and related words 115 of the words 112. The related words 115 belong to a category different from the category 111 to which the words 112 belong, and are, for example, words associated from co-occurrence indices, the strength of vector values obtained by vectorizing words, etc.

[0042] As shown in FIG. 6, the inflected word management table 48 has a type of change 121 and inflected words 122 belonging to the type of change 121. The type of change 121 indicates the classification of word changes, for example, "negative words". The inflected words 122 are words such as "negation" and "not".

[0043] The model management table 49 manages models for extracting word pairs. As shown in FIG. 7, the model management table 49 has an identifier 131 that uniquely identifies the model, a model name 132 that is the name of the model, a storage directory 133 as an example of information specifying the document data 44 used when generating the model name 132, the creator 134 who generated the model name 132, relationship information 135 regarding the relationship between words considered by the creator 134 when creating the model name 132, and initial dictionary information 136 representing an initial dictionary name as an example of information regarding the categories and the initial dictionary of word pairs between categories used when generating the model name 132. The initial dictionary information 136 will be described later.

[0044] The above-mentioned word pair relationship acquisition unit 41 is an example of a relationship estimation unit. It acquires search history records from the search keyword history (for example, the search history of the logged-in user), and calculates the appearance frequency of the combination of the above-mentioned search keywords for each search history record. When there is a search keyword with a frequency equal to or higher than a predetermined threshold, the word pair relationship acquisition unit 41 calculates the similarity between the relationship information of each record in the model management table 49 and the search keyword with a frequency equal to or higher than the threshold. When there is a record with a similarity equal to or higher than a predetermined threshold, and the two category names of the record are not two categories (for example, input by the user), the search keyword of the record is used as a candidate for the relationship information.

[0045] The word pair relationship acquisition unit 41 is an example of a relationship estimation unit. For example, when the search process of a document is being performed, a pair of two categories and search keywords input at that time is used as the object of calculating the similarity with the records in the model management table 49. When there is no record with the pair of the above-mentioned search keywords as the relationship, the pair of the above-mentioned search keywords is used as a candidate for the relationship information, and a message is created and presented externally recommending the creation of a word pair extraction model including the pair of the above-mentioned search keywords.

[0046] Since the model management table 49 is a table for managing the document data 44 used for generating (learning) the model, it manages a plurality of document information. Therefore, the model management table 49 also manages the storage directory 133 storing a plurality of documents used for learning.

[0047] Here, as shown in FIG. 8, the document data 44 manages a document ID 202 for identifying a plurality of document data 44 from each other, a title 301 of the document data 44, a text 302 of the document data 44, a generation date and time 303 of the document data 44, a source URL (Uniform Resource Locator) 304, and a storage directory 305.

[0048] Document ID 202 is the identification information of a document recorded in another table. Source URL 304 indicates the URL of the acquisition source when, for example, document data 44 is acquired from the web. Storage directory 305 indicates the storage location of document data 44.

[0049] Initial dictionary information 136 has a first category 137, a second category 140, version information 138, 141 of each category dictionary, NG dictionary names 139, 142 for registering words to be excluded from the extraction target in each category, and an initial dictionary name 143 for relational word pairs between the first category and the second category.

[0050] As shown in FIG. 9, the search keyword history management table 55 has an identifier 151 that uniquely identifies a search keyword history record, date and time information 152 when the search keyword was input, the user name 153 who performed the search, the search keyword 154 input by the user name 153, a first category 155, and a second category 156.

[0051] As shown in FIG. 10, the relational information candidate list 51 has an identifier 161 that uniquely identifies a relational information candidate, a first category 162, a second category 163, and relational information 164.

[0052] As shown in FIG. 11, the candidate filtering pattern 50 has an identifier 171 that uniquely identifies a candidate filtering pattern and a candidate filtering pattern 172.

[0053] As shown in FIG. 12, the filtering process list 52 has an identifier 181 that uniquely identifies candidate filtering pattern processing, a candidate filtering pattern 182, a process name 183, and a plurality of parameter groups 184 at the time of execution. The parameter groups 184 may differ in parameter names and the number of parameters for each process name 183.

[0054] As shown in FIG. 13, the word pair initial dictionary 47 output according to this embodiment stores a word 191 belonging to a first category and a word 192 belonging to a second category.

[0055] As shown in FIG. 14, the extracted word / word pair DB 53 has an identifier 201 that uniquely identifies an extracted word or word pair, an identifier 202 of document information including the extracted word or word pair, a title 203 of the document information, the document text 204, and a word or word pair 205 extracted from the document.

[0056] The extracted word / word pair 205 has at least a model name 206 used for extracting the word / word pair, a category 207-n of the extracted word, an extracted word 208-n, and an extraction accuracy (score) 209.

[0057] The extracted word / word pair DB 53 is utilized when the user searches for a document from the document data 44 using an operation screen (see FIG. 15) described later.

[0058] The search keyword history management table 54 manages the history of search keywords.

[0059] FIGS. 15 and 16 are diagrams showing an example of a proper name extraction screen. The proper name extraction screen has a category design tab 222-1, an initial dictionary creation tab 222-2, and a search tab 222-3. FIG. 15 is a diagram showing an example of the initial dictionary creation tab 221-2, and FIG. 16 is a diagram showing an example of the search tab 221-3.

[0060] The operation screen shown in FIG. 15 is a screen displayed when a user (for example, UserA) selects the initial dictionary creation tab 221-2 after logging in to the information processing apparatus 1. The user designates desired category names in a category input area 222 for inputting a first category to which a word pair belongs and a category input area 223 for inputting a second category, respectively.

[0061] The category input areas 222, 223 may allow a user to input a category name using the input device 12, or may extract and list category names registered in the category dictionary 45, display a list of the registered category names in the category input areas 222, 223, and allow selection.

[0062] The relationship input area 224 is an area where the relationship between categories is input, and is an area for specifying the relationship between the first category in the category input area 222 and the second category in the category input area 223. A list of frequently searched keywords is displayed in the relationship input area 224 based on the search history of the logged-in user. Specifically, in the relationship input area 224, a list of frequently searched keyword pairs is displayed from the search history that lists pairs of search keywords that the user searched using the search screen of the search tab 221-3 shown in FIG. 16, and the user selects them. The relationship input area 224 may allow the user to enumerate keywords regarding the relationship between categories desired by the user. If a pair of keywords regarding the relationship between categories that the user wants to create is not displayed, the user may directly input it into the relationship input area 224.

[0063] When the relationship input area 224 is selected or input, the initial dictionary generation unit 38 acquires the document information stored in the document data 44, and extracts word pairs that are candidates for the initial dictionary from this document information and the information input in the category input areas 222, 223, and the relationship input area 224, and presents them as an initial dictionary candidate list 226.

[0064] The initial dictionary candidate list 226 includes a candidate ID 227, a document ID 228 including the word pair of the initial dictionary candidate, a document text 229, a word pair candidate 230 including a word 230-1 belonging to the first category and a word 230-2 belonging to the second category, a confirmation checkbox 231 indicating whether the user checks the details, and a registration checkbox 232 indicating whether the user registers the word pair candidate 230 in the initial dictionary.

[0065] In this embodiment, after an initial dictionary name for registering word pairs in which the registration check box 232 is checked is entered in the dictionary name input area 233 and then the OK button 234 is pressed, the initial dictionary generation unit 38 creates an initial dictionary. When the confirmation check box 231 is checked, information on the candidate ID 227 selected as "confirmed" is displayed in the document display area 235 that displays the document text including the word pair. Specifically, in the document display area 235, for example, the document text "The NY Dow continues to fall by $270. The FRB chairman is negative about interest rate cuts this year" is displayed with the notation colors of some of its word pairs (for example, "FRB" and "interest rate cut") changed.

[0066] Furthermore, when a candidate filtering pattern ID is added to the word pair, content information of the candidate filtering pattern ID (for example, "including a negative word among the peripheral words of 'interest rate cut'") is displayed in the corresponding filtering candidate display area 236.

[0067] The user checks the information displayed in the filtering candidate display area 236 and decides whether to register the word pair in the initial dictionary, edit it, or cancel it, and presses any one of the OK button 238, the edit (REVISE) button 239, and the cancel button 240. When the edit button 239 is pressed to select the editing of the words in the word pair, the user enters the edited words in the word editing area 237. When the OK button 238 or the edit button 239 is pressed, the initial dictionary generation unit 38 registers the candidates in which the registration check box 232 is checked in the initialization dictionary.

[0068] The operation screen of the search tab 221-3 shown in FIG. 16 is a search screen for the user to specify two categories after logging in to the information processing apparatus 1 and search for words belonging to each specified category from the document data 44. The user specifies desired category names in the category input area 250 for entering the first category to which the word pair belongs and the category input area 251 for entering the second category, respectively.

[0069] The category input areas 250 and 251 allow the user to input category names using the input device 12, or to extract and list the category names registered in the category dictionary 45, display the list in the category input areas 250 and 251, and make it selectable.

[0070] The period input area 252 is an area for specifying the period for acquisition based on the generation date or storage date of the data stored in the document data 44. When the user inputs one or more search keywords in the keyword input area 253, the category word extraction unit 34 of the initial dictionary generation unit 38 extracts the documents containing the words estimated to belong to the two categories specified by the user from the document data 44, further extracts the documents containing the search keywords specified by the user, lists them, and presents them in the document list 254.

[0071] The document list 254 stores the document title 255, the document body 256, the keyword of interest 257, the site 258, the detailed information display button 259 to be pressed when detailed information is to be displayed, etc. The site 258 is the storage location of the document, for example, the directory of the storage or the URL (Uniform Resource Locator) of the website.

[0072] The keyword of interest 257 includes, for example, the keywords with high search frequency contained in the document. When the user presses the detailed information display button 259, the document body is displayed in the same manner as the document display area 235 of FIG. 15 described above.

[0073] Furthermore, the search screen shown in FIG. 16 has a notification area 260. When the information processing device 1 determines that it is better to create an initial dictionary of word pairs that span two categories from the user's search keyword history information and generate a model for extracting the relationship of the word pairs, the notification area 260 creates a sentence recommending to generate a model including the set of search keywords with high search frequency as relationship information and presents it to the user to prompt a decision on whether to create an initial dictionary of relational word pairs.

[0074] When the user presses the OK button 261, the information processing apparatus 1 executes a process of using the set of search keywords as related information and creating candidate relational word pairs by the word pair relational acquisition unit 41.

[0075] The outline of the information processing system including the information processing apparatus 1 according to the present embodiment is as described above. Next, an operation example of the information processing system including the information processing apparatus 1 will be described. First, an outline of the information processing method of the information processing apparatus 1 according to the present embodiment will be described. The information processing method is an information processing method of the information processing apparatus 1 that supports the creation of the word pair initial dictionary 47 based on the relationship between words belonging to two categories from document information composed of one or more sentences. The category word extraction unit 34 as an example of the word extraction unit extracts words belonging to two categories from the document information in a word extraction step. The search unit 37 determines whether a word belonging to at least one of the two categories from the document information and corresponding to the input search keyword is included, and presents the document information determined to include the word to the outside in a search step. The word pair candidate extraction unit 40 as an example of the word pair generation unit generates word pairs between categories that satisfy predetermined conditions from the extracted words in a word pair generation step. The candidate filtering generation registration processing unit 43 as an example of the filtering unit accumulates the input history of the search keyword and estimates the relationship between words desired to be obtained from the input history of the search keyword in a relationship estimation step. Based on the relationship, it filters the candidates of the word pairs to be registered in the word pair initial dictionary 47 in a filtering step. The word pair initial dictionary determination unit 42 as an example of the initial dictionary registration unit presents the candidates of the word pairs to the outside and, as a result of determining whether to register them in the word pair initial dictionary 47, registers the candidates of the word pairs determined to be registered in the word pair initial dictionary 47 in the word pair initial dictionary 47 in an initial dictionary registration step. This will be described more specifically below.

[0076] FIG. 17 is a flowchart showing an example of the procedure of initial dictionary creation support processing. The initial dictionary creation support processing is a process executed by the word extraction processing program 20 when the user selects the initial dictionary creation tab 221-2 of the relational word pair via the operation screen described later.

[0077] When the user logs in via the client terminal 2 (step S62), if the selected screen is the word pair initial dictionary creation screen (FIG. 15) (step S60, y), the word extraction processing program 20 calls the initial dictionary generation unit 38.

[0078] The initial dictionary generation unit 38 first calls the word pair relationship acquisition unit 41 to acquire relationship information by the word pair relationship acquisition unit 41 (step S61). Next, the initial dictionary generation unit 38 acquires document information to be the information extraction target from the document data 44 (step S62). Note that the document information to be acquired does not necessarily have to be all the document information stored in the document data 44, and it may be document information newly stored in the document data 44 by the user for model creation.

[0079] The initial dictionary generation unit 38 acquires one sentence from the document information acquired in step S62 (step S64), performs morphological analysis, and divides it into elements (step S65). The morphological analysis technique is realized by using natural language processing methods such as MeCab or NLTK (Natural Language Tool Kit), for example. For example, if the document acquired in step S64 is "The NY Dow has fallen another 270 dollars, and the Fed chairman is negative about interest rate cuts this year.", it is assumed that it is divided into "NY Dow / has / fallen further / 270 dollars / lower / to / become / Fed chairman / has / this year / interest rate cuts / is / negative / about" by morphological analysis.

[0080] The category word extraction unit 34 performs proper name extraction on each of the divided elements to determine whether there are words belonging to the first category and the second category specified by the user, and determines the category of the extracted words (step S66).

[0081] For the extraction of inherent expressions, natural language processing techniques such as sspaCy may be used, for example. In the previous example, for the categories specified by the user, if the first category is "FOMC" and the second category is "stock price fluctuation", for example, the category word extraction unit 34 determines that "NY Dow", "continuous decline", and "interest rate cut" belong to the "stock price fluctuation" category, and "FRB" belongs to the "FOMC" category.

[0082] Here, a plurality of words belonging to each category are managed as a dictionary, and category-specific tags (103) are assigned to the words specific to the category. The word pair candidate extraction unit 40 generates word pair candidates between categories that satisfy predetermined conditions from the words extracted in step S66 according to the conditions (step S67). The conditions are, for example, that at least one sentence contains one or more words belonging to the first category and the second category, and that at least one sentence contains a word to which a category-specific tag 103 is assigned. Note that the conditions may be further added, for example, when improving the accuracy of word extraction.

[0083] The word pair candidates generated in step S67 are, for example, three pairs: (FRB, NY Dow), (FRB, continuous decline), and (FRB, interest rate cut). Next, the word pair initial dictionary determination unit 42 executes a word pair candidate filtering process. In the word pair candidate filtering process, the word pair initial dictionary determination unit 42 filters out word pairs that correspond to the relationship information specified by the user and extracted by the word pair relationship acquisition unit 41 from the word pair candidates generated in step S67 (step S68). Details of the word pair candidate filtering process will be described later.

[0084] In this embodiment, the processes from step S64 to step S68 are repeated for the number of acquired documents, and the candidate word pairs for initial dictionary registration are listed together with the document information (see the initial dictionary candidate list 226 in FIG. 15).

[0085] Also, in the present embodiment, for the list of word pairs extracted as a result of filtering, that is, the initial dictionary candidate list 226, a determination result as to whether the user can register the initial dictionary is obtained (step S69), and the word pairs determined by the user to be registered in the initial dictionary are registered in the word pair initial dictionary (step S70).

[0086] On the other hand, when the screen selected by the user after login is the search screen of FIG. 16 (step S71, y), the search unit 37 acquires the two category input areas 250 and 251, the period input area 252, and the search keyword input area 253 input by the user, and performs a search process of extracting words that belong to the two categories respectively and are estimated to correspond to the search keyword input area 253 from the document data 44 (step S72).

[0087] The search unit 37 accumulates the history of search keywords in the search keyword history management table 54 every time the user performs a search. Here, the word pair relationship acquisition unit 41 determines whether the keyword input area 253 input by the user on the search screen should recommend the creation of a relational word pair initial dictionary.

[0088] FIG. 18 is a flowchart showing an example of the procedure of the word pair relationship acquisition process. The word pair relationship acquisition process is executed by the word pair relationship acquisition unit 41. The word pair relationship acquisition unit 41 acquires the keyword history information searched by the logged-in user in the past from the search keyword history management table 54, and calculates the search frequency for each set of search keywords (step S80).

[0089] If there is a set of search keywords with a frequency equal to or higher than a predetermined threshold among the frequencies of the sets of search keywords (step S81, y), the word pair relationship acquisition unit 41 calculates the similarity of the relationship information 135 extracted by the model name 132 stored in the model management table 49 (step S82). Specifically, the word pair relationship acquisition unit 41 calculates the similarity between the relationship information 135 in the model management table 49 and the corresponding search keyword history or the search keyword input by the user.

[0090] For example, when the set of search keywords with high user search frequency is (FOMC, stock price, factors for increase), the word pair relationship acquisition unit 41 calculates the similarity between this set of search keywords and all the records stored in the model management table 49.

[0091] When there is a record with a similarity above a predetermined threshold (step S83, y), if the category used to generate the model name 132 from which the record was extracted is different from the two categories entered by the user on the search screen or the two categories 155 and 156 to which the search keyword 154 belongs and is stored in the search keyword history management table 55 (step S84, y), it is regarded as a candidate for relationship information and recorded in the relationship information candidate list 51.

[0092] Here, when the operation screen selected by the user is the search screen (Figure 16), that is, when in the middle of a search process (step S85, y), the word pair relationship acquisition unit 41 generates a recommendation sentence for creating a word pair extraction model including the set of user input search keywords, and presents it to the notification area 260 of the search screen in Figure 16 (step S86). That is, the word pair relationship acquisition unit 41 generates an inquiry sentence for creating a word pair extraction model including the input search keywords and presents it to the user.

[0093] When the user presses the OK button 261 indicating that they want to generate a model (step S87, y), the word pair relationship acquisition unit 41 obtains synonyms and related words of each word in the relationship candidate information using the category dictionary 45 and the related word dictionary 46 to expand the relationship candidate information (step S88).

[0094] For example, when the relationship candidate information includes "FOMC, stock price, factors for decrease" as search keywords, "FRB" is added as a related word of "FOMC", "interest rate cut" is added as a related word of "stock price", and "decline" and "continuous decline" are added as synonyms of "decrease" to the relationship candidates.

[0095] When the screen selected by the user is the initial dictionary creation screen (Figure 15) (step S85, n), the word pair relationship acquisition unit 41 lists the history of pairs of search keywords whose search frequency is equal to or higher than the threshold, and presents it as relationship candidate information to the relationship input area 224 (step S89). When the user selects one from the relationship input area 224 (step S100, y), step S88 is executed. In step S88, the word pair relationship acquisition unit 41 determines, for example, the category and synonyms of each word in the relationship candidate information, and adds the relationship candidate information.

[0096] Figure 19 is a flowchart showing an example of the procedure of the word pair candidate filtering process (step S68) shown in Figure 17. In the word pair candidate filtering process, the word pair initial dictionary determination unit 42 repeats the process for the number of word pair candidates for each word pair candidate generated in step S67 shown in Figure 17 (step S90).

[0097] The word pair initial dictionary determination unit 42 acquires one word pair candidate from the word pair candidates (step S91). The word pair initial dictionary determination unit 42 determines whether each word of the word pair candidate is included in the relationship candidate information extracted by the word pair relationship acquisition unit 41 (step S92).

[0098] When the word of the word pair candidate is not included in the relationship candidate information (step S92, n), the word pair initial dictionary determination unit 42 excludes the word pair candidate from the initial dictionary candidates, returns to step S1, and acquires the next word pair candidate. On the other hand, when the word of the word pair candidate is included in the relationship candidate information (step S92, y), the word pair initial dictionary determination unit 42 further determines whether it corresponds to the candidate filtering pattern for the number of candidate filtering patterns (step S93). That is, the word pair initial dictionary determination unit 42 determines whether, for example, a word that changes the meaning of a word (hereinafter, a changing word) is included as the candidate filtering pattern.

[0099] As shown in the above-described change word management table 48 (see FIG. 6), when the type of change 121 is "negative word", words such as "negative", "not", and "zu" are associated with the change word 122. When the type of change 121 is "irony", words such as "wouldn't it be" are associated with the change word 122.

[0100] The candidate filtering generation registration processing unit 43 extracts elements of a sentence including a change word from the acquired document information, performs an appearance tendency analysis, generates, and updates a candidate filtering pattern 50. The candidate filtering generation registration processing unit 43 acquires one filtering pattern record from the candidate filtering pattern 50 (step S94).

[0101] The candidate filtering generation registration processing unit 43 acquires, from the filtering processing list 52, the verification processing (the processing of the processing name 183 in the filtering processing list 52 shown in FIG. 12) associated with the filtering pattern 182 of the acquired candidate filtering pattern record (step S95).

[0102] The candidate filtering generation registration processing unit 43 acquires the main text of the original document information from which the word pair candidates are extracted, and verifies whether it corresponds to the content of the record of the candidate filtering pattern 172 of the candidate filtering pattern 50 shown in FIG. 11 (the same as the candidate filtering pattern 182 of the filtering processing list 52 shown in FIG. 12) in the verification processing which is the processing of the above processing name 183. That is, the candidate filtering generation registration processing unit 43 verifies, for example, whether a word pair candidate (FRB, interest rate cut) matches "including a negative word in the peripheral words" where the ID 171 of the candidate filtering pattern 50 is "1".

[0103] When it matches (step S96, y), the candidate filtering generation registration processing unit 43 adds the word pair candidate to the initial dictionary candidate list 226. On the other hand, when it does not match (step S96, n), the candidate filtering generation registration processing unit 43 does not add the word pair candidate to the initial dictionary candidate list 226, returns to step S94 to obtain the next filtering pattern, and executes the subsequent steps.

[0104] Since the verification process where the candidate filtering pattern ID 171 is "1" is "check_negative_word(document text, verification target word)" where the ID 181 in the filtering process list 52 is "1", the document information "The NY Dow has fallen another 270 dollars, and the FRB chairman is negative about interest rate cuts this year." from which the word pair candidate (FRB, interest rate cut) is extracted and "interest rate cut" are used as parameters to execute the verification process.

[0105] That is, in this embodiment, it is verified by a natural language analysis method whether negative words are included in the words around "interest rate cut", and when the word "negative" is detected, it is determined that it matches the candidate filtering pattern ID "1" (step S96, y).

[0106] In this case, the candidate filtering generation registration processing unit 43 adds the candidate filtering pattern ID "1" to the word pair candidate (step S97) and adds it to the initial dictionary candidate list (step S98). When it does not match the candidate filtering pattern (step S96, n), the candidate filtering generation registration processing unit 43 adds the word pair candidate to the initial dictionary candidate list (step S98). Further, the candidate filtering generation registration processing unit 43 performs a series of processes from step S94 to step S97 for the number of candidate filtering patterns, and finishes the word pair candidate filtering process.

[0107] The information processing apparatus 1 according to the present embodiment is an information processing apparatus 1 that supports the creation of a word pair initial dictionary based on the relationship between words belonging to two categories from document information composed of one or more sentences. The information processing apparatus 1 includes a category word extraction unit 34 that extracts words belonging to two categories from the document information, a search unit 37 that determines whether a word belonging to at least one of the two categories from the document information includes a word corresponding to an input search keyword, and presents the document information determined to include the word to the outside, a word pair candidate extraction unit 40 that generates a word pair between categories that satisfies a predetermined condition from the extracted words, a word pair relationship acquisition unit 41 that accumulates the input history of the search keyword and estimates the relationship between words desired to be obtained from the input history of the search keyword, a candidate filtering generation registration processing unit 43 that filters candidates for word pairs to be registered in the word pair initial dictionary 47 based on the relationship, and a word pair initial dictionary determination unit 42 that presents candidates for word pairs to the outside and, as a result of determining whether to register them in the word pair initial dictionary 47, registers the candidates for word pairs determined to be registered in the word pair initial dictionary 47 in the word pair initial dictionary 47.

[0108] By doing so, since the relationship between categories that the user wants to obtain is estimated from the input history of the search keyword and word pairs based on the relationship between categories are generated, the man-hour for the user to manually define word pairs can be reduced.

[0109] Note that the present invention is not limited to the above-described embodiment, and includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiment has been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to an apparatus having all the configurations described. Also, each element described in parallel in the present embodiment may be in a mode in which at least one of each element is connected in series to another element.

Industrial Applicability

[0110] The present invention can be applied to an information processing apparatus related to a technique for assisting the creation of a dictionary used in, for example, natural language processing.

Explanation of Signs

[0111] 1……Information processing system, 2……User device, 20……Word extraction processing unit, 37……Search unit, 38……Initial dictionary generation unit, 45……Category dictionary, 46……Related word dictionary, 47……Word pair initial dictionary, 48……Variant word management table, 49……Model management table, 50……Candidate filtering pattern, 51……Filtering processing list

Claims

1. An information processing apparatus for assisting in creating an initial dictionary of word pairs based on the relationship between words belonging to two categories from document information composed of one or more sentences, comprising: a word extraction unit that extracts words belonging to the two categories from the document information; a search unit that determines whether a word belonging to at least one of the two categories from the document information includes a word corresponding to an input search keyword, and presents the document information determined to include the word to the outside; a word pair generation unit that generates word pairs between categories that satisfy a predetermined condition from the extracted words; a relationship estimation unit that accumulates the input history of the search keyword and estimates the relationship between words desired to be obtained from the input history of the search keyword; a filtering unit that filters candidates for word pairs to be registered in the initial dictionary of word pairs based on the relationship; an initial dictionary registration unit that presents the candidates for word pairs to the outside and, as a result of determining whether to register them in the initial dictionary of word pairs, registers the candidates for word pairs determined to be registered in the initial dictionary of word pairs in the initial dictionary of word pairs; An information processing apparatus characterized by comprising the above.

2. A plurality of words belonging to each category are managed as a dictionary, and a category-specific tag is assigned to a word specific to the category. The word pair generation unit, as the predetermined condition, includes at least one word belonging to the category in one sentence and a word to which the category-specific tag of at least one of the categories is assigned in one sentence. The information processing apparatus according to claim 1, characterized in that.

3. It has a model management table for managing a model for extracting the word pair. The model management table has a model name that is the name of the model, information for specifying the document information used when generating the model, relationship information regarding the relationship between the words, two category names, and an initial dictionary name of word pairs between categories regarding the initial dictionary of word pairs. The relationship estimation unit acquires a search history record of the logged-in user from the history of the search keyword, calculates the appearance frequency of the combination of the search keywords for each search history record, and when there is a search keyword with a frequency equal to or higher than a predetermined threshold, calculates the similarity between the relationship information of each record in the model management table and the search keyword with a frequency equal to or higher than the threshold. When there is a record whose similarity is equal to or greater than a predetermined threshold value, and when the two category names of the record are not the two categories, the search keyword of the record is used as a candidate for related information The information processing apparatus according to claim 1, characterized in that

4. The relationship estimation unit regards the combination of the two categories and the search keyword as an object for calculating similarity with records in the model management table. When there is no record having the combination of the search keywords as a relationship, the combination of the search keywords is used as a candidate for related information, creates a recommendation to recommend creating a word pair extraction model including the combination of the search keywords, and presents it to the outside The information processing apparatus according to claim 3, characterized in that

5. The filtering unit has one or more candidate filtering patterns and a filtering verification process for verifying whether or not it matches each candidate filtering pattern, determines whether or not the candidate of the word pair is included in the candidate of the related information, and when the candidate of the word pair is included in the candidate of the related information, performs a filtering verification process associated with each candidate filtering pattern. If it matches in the filtering verification process, adds the identification information of the candidate filtering pattern to the candidate word pair and adds it to the initial dictionary candidate list The information processing apparatus according to claim 4, characterized in that

6. The candidate filtering pattern includes at least a phrase that changes the meaning of a word in the document information The information processing apparatus according to claim 5, characterized in that

7. The candidate filtering pattern includes the word extracted as a result of verification by the filtering verification process and the category to which the word belongs The information processing apparatus according to claim 5, characterized in that

8. An information processing method of an information processing apparatus for assisting in creating an initial dictionary of word pairs based on the relationship between words belonging to two categories from document information composed of one or more sentences, comprising a word extraction step in which a word extraction unit extracts words belonging to the two categories from the document information a search step in which a search unit determines whether or not the document information includes a word belonging to at least one of the two categories and corresponding to an input search keyword, and presents the document information determined to include the word to the outside ​ A word pair generation step in which a word pair generation unit generates word pairs between categories that satisfy a predetermined condition from the extracted words; A relationship estimation step in which a filtering unit accumulates the input history of the search keyword and estimates the relationship between words desired to be obtained from the input history of the search keyword; A filtering step of filtering candidates for word pairs to be registered in the initial word pair dictionary based on the relationship; An initial dictionary registration step in which an initial dictionary registration unit presents candidates for the word pairs to the outside and, as a result of determining whether to register them in the initial word pair dictionary, registers the candidates for the word pairs determined to be registered in the initial word pair dictionary in the initial word pair dictionary; An information processing method characterized by comprising the above.

Images