JP7536558B2 - Document information extraction device and document information extraction method - Google Patents

Description

The present invention relates to a document information extraction device and a document information extraction method, and is suitable for use in a document information extraction device and a document information extraction method that extracts information appropriate for a user through document analysis of an input document.

Traditionally, corporate sales representatives have had to record sales activities with customers in documents such as daily sales reports, and also devise recommended information (e.g. products) to propose to customers in subsequent sales activities depending on the progress of sales activities and meeting situations. Sales results can vary greatly depending on the type of recommended information provided, so there was hope for technology that could efficiently select recommended information.

For example, Patent Document 1 describes an information processing device and method that, when a user creates a form, suggests a form document that may be useful as a reference based on user-input information such as the form layout. The information processing device described in Patent Document 1 compares the user-input information with attribute information of other forms registered in advance, and selects a recommended form, making it possible to suggest reference forms that the user has not imagined or that exceed the user's expectations.

JP 2013-025726 A

For example, in a company that provides products or services, when an employee in charge of selling the products or services (hereinafter referred to as a salesperson) selects a product or service suitable for proposal to a customer based on the results of sales activities up to that point, the product or service to be proposed to the customer generally changes in response to the history and progress of sales activities. Specifically, for example, if a meeting with a customer goes well, it is preferable to propose many products or services similar to those proposed in the previous sales activity, whereas if the meeting is not going well, there is a demand for proposing products or services that prioritize novelty. However, the conventional technology disclosed in the above-mentioned Patent Document 1 and the like has a problem in that it is not possible to determine output information to be recommended to the user in consideration of the progress of input information that changes over time.

The technology disclosed in Patent Document 1 is a technology for inputting form information and outputting reference form information, but it has the problem of being inapplicable when the document type of the input information and the document type of the output result (reference form) are different. For example, when a salesperson inputs information about his/her sales activities (hereinafter referred to as a sales report) and selects and outputs information about products or services suitable for proposing to a customer from a product catalog document, the attribute information in the sales report and the product catalog document are completely different, so that even if the attribute information of each is compared, it is not possible to select a recommended item.

The present invention has been made in consideration of the above points, and aims to propose a document information extraction device and document information extraction method that can extract suitable recommended information by taking into account the progress of input information that changes over time.

In order to solve such problems, the present invention provides a document information extraction device that proposes recommended items by performing document analysis of activity records recorded in chronological order for the same case, the document information extraction device including a storage unit that stores first document information in which the activity records are registered and second document information that is composed of a document type different from that of the first document information and in which document data relating to candidates for the recommended items are registered , a progress determination unit that classifies the progress status of the activity records into one of a plurality of progress categories based on keywords that indicate the progress status and are included in the activity records of the first document information, and a progress determination unit that extracts first related words that constitute related named entities from the first document information and generates first related word information that compiles the extracted first related words. a second document processing unit that extracts second related words that constitute related named entities from the second document information and generates second related word information that compiles the extracted second related words; and an item processing unit that selects and outputs recommended items having second related words that are directly or indirectly connected to first related words included in the activity record of the first document information from among candidates for recommended items based on a selection condition for recommended items corresponding to the progress category classified by the progress determination unit and a result of related word linkage that links the first document information and the second document information by connecting the first related word information and the second related word information with a common related word .

In order to solve the above problem, the present invention provides a document information extraction method for a document information extraction device that proposes recommended items by performing document analysis on activity records recorded in chronological order for the same case, the method including a first document information storage step of registering the activity records in first document information, a second document information storage step of registering document data related to the recommended item candidates in second document information composed of a document type different from that of the first document information, a progress determination step of classifying the progress status of the activity records into one of a plurality of progress categories based on a keyword that suggests the progress status and that is included in the activity records of the first document information, and a first document processing step of extracting first related words that constitute related named entities from the first document information and generating first related word information that compiles the extracted first related words. a second document processing step of extracting second related words constituting related named entities from the second document information and generating second related word information compiling the extracted second related words; and after the first document processing step and the second document processing step are executed , in an item processing step, selecting and outputting from among candidates for recommended items, recommended items having second related words directly or indirectly connected to first related words included in the activity record of the first document information, based on a selection condition for recommended items corresponding to the progress category classified by the progress determination step and a result of related word linkage in which the first document information and the second document information are linked by connecting the first related word information and the second related word information with a common related word .

According to the present invention, it is possible to extract suitable recommended information by taking into account the progress of input information that changes over time.

1 is a block diagram showing an example of a functional configuration of a document information extraction device 1 according to an embodiment of the present invention. 1 is a block diagram showing an example of a hardware configuration of an information processing device 10 used to realize a document information extraction device 1. FIG. 1 is an example of first document information 111. 10 is a diagram for explaining an image of a first related word table 116 included in first related word information 112. FIG. 13 is an example of a progress category table 117 included in first related word information 112. 13 is an example of second document information 113. 13 is a diagram for explaining an image of a second related word table 118 included in second related word information 114. FIG. FIG. 1 is a diagram for explaining an image of related word linkage between different documents. 9 is a diagram for explaining, in a table format, the related word linkage between the different kinds of documents shown in FIG. 8 . FIG. 2 is a diagram showing an example of a recommendation candidate calculation table 230. 13 is a flowchart illustrating an example of a processing procedure for a recommendation calculation process. 13 is a flowchart showing an example of a processing procedure for a relationship priority designation process. 13 is a flowchart showing an example of a processing procedure for evaluation value priority designation processing. 13 is a flowchart showing an example of a processing procedure for a past proposal prioritization process. 13 is an example of an activity report registration screen 240. 2 is an example of a recommendation confirmation screen 250.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Note that the following description and drawings are merely examples for explaining the present invention, and some parts have been omitted or simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

In the following explanation, the letter "S" before a symbol indicates a processing step. In addition, in the following explanation, various information (data) will be explained using the expressions "table" and "information", but the various information may be expressed using data structures other than these. In addition, when explaining identification information, expressions such as "identifier" and "ID" will be used as appropriate, but these are interchangeable.

In the following explanation, the term "document" refers to data that has been digitized from a document in a specified format, such as text data. In the following explanation, the term "word" refers to a named entity extracted by an information processing device (document information extraction device). In the following explanation, a named entity is a general term for words that refer to specific things, such as people's names or place names. For example, "engineer" and "Tokyo" are named entities for "people" and "places," respectively. In this case, the "people" and "places" into which named entities are classified are referred to as "categories."

(1) Configuration Fig. 1 is a block diagram showing an example of the functional configuration of a document information extraction device 1 according to an embodiment of the present invention. The document information extraction device 1 creates anonymity dictionaries storing related word sets from named entities extracted from two types of document information (first document information 111, second document information 113), and links the two types of document information by linking the anonymity dictionaries. The document information extraction device 1 also classifies document information including history, progress, etc., input in chronological order, based on preset progress category information (progress category table 117), calculates item suitability according to the progress category, and recommends optimal items.

As shown in FIG. 1, the document information extraction device 1 includes a memory unit 110, a first document processing unit 120, a second document processing unit 130, and an item processing unit 140.

The storage unit 110 stores first document information 111, first related word information 112, second document information 113, second related word information 114, and recommendation history information 115. The storage unit 110 stores this information (data), for example, as a database table provided by a DBMS (DataBase Management System) or as a file provided by a file system.

The first document information 111 and the second document information 113 include multiple documents from which named entities are extracted. The contents of the first document information 111 and the second document information 113 are not necessarily limited; for example, the first document information 111 and the second document information 113 may be information input by a user or information obtained via a communication network such as the Internet.

The first document information 111 stores input information (e.g., sales-related documents) entered by a sales representative, as will be described in detail later with reference to FIG. 3. The second document information 113 stores information (e.g., product catalog documents) from which output information is extracted and which is proposed to the user by the item processing unit 140 as a recommended item (recommended item), as will be described in detail later with reference to FIG. 6. Note that the document information extraction device 1 according to this embodiment has a feature of proposing appropriate output information based on named entities contained in the input information, even if the document types (document types) of the first document information 111 including the input information and the second document information 113 including the output information are different. However, the document types of the first document information 111 and the second document information 113 do not necessarily have to be different. For example, the first document information 111 and the second document information 113 may both be sales reports. However, the first document information 111 must at least be capable of acquiring information that indicates the progress of the activity; for example, it must contain multiple sales daily reports for multiple days for the same sales activity, and contain descriptions that can acquire information such as whether the sales activity is going well or difficult.

The first related word information 112 is progress classification information for classifying the related words extracted by the first document processing unit 120 and the documents included in the first document information 111 by activity progress, and includes a first related word table 116 and a progress category table 117.

The second related word information 114 is related word information extracted by the second document processing unit 130 and includes a second related word table 118.

The recommendation history information 115 stores the degree of access (an index of evaluation value) to items proposed by the item processing unit 140. The degree of access reflects, for example, the results of a user's actual selection or viewing of an item proposed by the item processing unit 140, and is stored in the recommendation history information 115 as the evaluation value of the item. In addition, the recommendation history information also stores a "recommendation proposal history" that accumulates the history of items proposed in previous sales activities (recommended items) for each case.

The first document processing unit 120 is configured to include a named entity feature extraction unit 121, a learning inference unit 122, an associated word table generation unit 123, and a progress category generation unit 124. The named entity feature extraction unit 121 performs morphological analysis on the document of the first document information 111 to extract named entities and features, and classifies the extracted named entities into categories. The learning inference unit 122 trains an inference model using a machine learning model. The associated word table generation unit 123 uses the inference model trained by the learning inference unit 122 to generate a first associated word table 116 that compiles associated named entities from among the named entities extracted by the named entity feature extraction unit 121. The progress category generation unit 124 classifies the document of the first document information 111 into multiple categories based on the progress category table 117 according to activity progress keywords contained in the document, thereby determining the progress status.

The second document processing unit 130 is composed of a named entity feature extraction unit 131, a learning inference unit 132, and a related word table generation unit 133. The named entity feature extraction unit 131 performs morphological analysis on the documents of the second document information 113 to extract named entities and features, and classifies the extracted named entities into categories. The learning inference unit 132 trains an inference model using a machine learning model. The related word table generation unit 133 uses the inference model to generate a second related word table 118 that compiles related named entities from among the named entities extracted by the named entity feature extraction unit 131.

The item processing unit 140 is configured to include a compatibility calculation unit 141, an item recommendation unit, and an item output unit. The compatibility calculation unit 141 calculates the compatibility between a specified document included in the first document information 111 and an item included in the second related word information 114 based on the first related word table 116, the second related word table 118, and the recommendation history information 115. The item recommendation unit 142 selects an item to recommend based on the compatibility of the above items, and proposes the selected item to the user via the item output unit 143, thereby realizing item recommendation. In addition, the item processing unit 140 obtains the user's usage history for the proposed item (such as the number of searches and the number of times a helpful button was pressed) and updates the recommendation history information 115.

Figure 2 is a block diagram showing an example of the hardware configuration of an information processing device 10 used to realize the document information extraction device 1.

The information processing device 10 illustrated in FIG. 2 includes a processor 11, a main memory device 12, an auxiliary memory device 13, an input device 14, an output device 15, and a communication device 16. The information processing device 10 used to realize the document information extraction device 1 may be realized in whole or in part 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. In addition, all or in part of the functions provided by the information processing device 10 may be realized by a service provided by the cloud system via an API (Application Programming Interface), for example. In addition, the document information extraction device 1 may be configured using multiple information processing devices 10 connected to each other so as to be able to communicate with each other.

In FIG. 2, the processor 11 is configured using, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), an AI (Artificial Intelligence) chip, etc.

The main memory device 12 is a device that stores programs and data, and is, for example, a read only memory (ROM), a random access memory (RAM), or a non-volatile memory (NVRAM (Non Volatile RAM)).

The auxiliary storage device 13 is, for example, a hard disk drive, an SSD (Solid State Drive), an optical storage device (CD (Compact Disc), DVD (Digital Versatile Disc), etc.), a storage system, an IC card, a read/write device for a recording medium such as an SD card or an optical recording medium, a storage area of a cloud server, etc. Programs and data can be read into the auxiliary storage device 13 via a recording medium reader or a communication device 16. The programs and data stored (memorized) in the auxiliary storage device 13 are read into the main storage device 12 as needed.

The input device 14 is an interface that accepts input 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, etc.

The output device 15 is an interface that outputs various information such as the progress of processing and the results of processing. 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.). As another example, the information processing device 10 may be configured to input and output information to and from other devices via the communication device 16.

The input device 14 and the output device 15 constitute a user interface that receives and presents information to and from the user.

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 a communication network (Internet, LAN (Local Area Network), WAN (Wide Area Network), dedicated line, public communication network building), and is, for example, a NIC (Network Interface Card), a wireless communication module, a USB module, etc.

In addition, 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.

Among the functional components of the document information extraction device 1, the functions of the first document processing unit 120, the second document processing unit 130, and the item processing unit 140 are realized by the processor 11 reading and executing a program stored in the main memory device 12 in the information processing device 10 shown in FIG. 2, or by the hardware (FPGA, ASIC, AI chip, etc.) that constitutes these devices. Also, among the functional components of the document information extraction device 1, the function of the memory unit 110 is realized by storing various target information (data) in the main memory device 12 or the auxiliary memory device 13 (which may be an external memory device) as, for example, a database table or a file managed by a file system.

(2) Various Information Next, various information stored in the storage unit 110 of the document information extraction device 1 will be described.

Figure 3 is an example of the first document information 111. The first document information 111 illustrated in Figure 3 is an internal document stored in a company's information processing system, and is a sales-related document regarding a case in which multiple sales activities were carried out on different activity days for the same sales case.

As shown in FIG. 3, the first document information 111 is information that manages information about each sales case (target case) (e.g., case name, demand source, order source, amount, case overview, etc.). In the first document information 111 in FIG. 3, sales daily reports 111a, 111b, and 111c, which are records of sales activities on each activity day, are managed as one group. In the following explanation, each sales daily report 111a to 111c is sometimes referred to as first document information 111a to 111c because each sales daily report 111a to 111c is the smallest unit document in the first document information 111. The first document processing unit 120 extracts specified named entities from these smallest unit documents in the first document information 111 (first document information 111a to 111c), and generates (creates) the first related word table 116 by combining related named entities from the extracted named entities.

Figure 4 is a diagram for explaining an image of the first related word table 116 included in the first related word information 112. As described above, the first related word table 116 is generated (created) by the first document processing unit 120. To explain in more detail, first, the named entity feature extraction unit 121 of the first document processing unit 120 extracts predetermined named entities (specifically, for example, "manual work", "worker", "work time", "automation", "semiconductor manufacturing", "start of review", "deep dive", "difficulty", "discrepancy", etc.) from the documents of the first document information 111 (first document information 111a to 111c). Then, the related word table generation unit 123 of the first document processing unit 120 uses the inference model learned by the learning inference unit 122 to compile related named entities from the extracted named entities, thereby generating the first related word table 116.

The first related word table 116 generated as described above corresponds to a "related word dictionary" which collectively refers to combinations of named entities that are used similarly (synonym relationships) and combinations of named entities that frequently appear in the same sentence (co-occurrence relationships) in the target first document information 111a to 111c. The network graph 1161 shown in FIG. 4 is a network graph in which the related words contained in the first related word table 116 (related word dictionary) in the first document information 111a are expanded. According to the network graph 1161, the relevance of the named entities used in the first document information 111a is represented in a form that is easy to visually recognize.

Figure 5 is an example of a progress category table 117 included in the first related word information 112. The progress category table 117 illustrated in Figure 5 is configured with items of category name 1171, recommendation priority rule 1172, and keyword 1173. Each of the above items and their contents are set in advance by the user according to arbitrary criteria in order to categorize the documents in the first document information 111 (first document information 111a to 111c).

For example, the name of each category into which the document will be classified is set in category name 1171, and in the case of FIG. 5, category names such as "Start", "Smooth", "Difficult", and "Digging deeper" are registered. Note that there is no limit to the number of categories or classification boundaries, but the previous document included in the first document information 111 must fall into one of the categories (category name 1171) in the progress category table 117.

The recommendation priority rules 1172 also specify rules for what attributes of items are to be given priority for recommendation in each category. As shown in FIG. 5, the recommendation priority rules 1172 corresponding to each category are not limited to one, and any type of rule with any content can be set, taking into account the characteristics of the corresponding category.

In addition, in the keyword 1173, a phrase (keyword) suggesting progress information corresponding to each category is set so that the first document information 111 (strictly speaking, the first document information 111a to 111c) can be automatically classified into each category by the progress category generation unit 124. As described above, the progress category generation unit 124 of the first document processing unit 120 classifies the progress status of each document (first document information 111a to 111c) of the first document information 111 based on the progress category table 117. At this time, if a keyword registered in the keyword 1173 is included in each document in the first document information 111, the progress category generation unit 124 automatically classifies the document into the corresponding category. Note that, as shown in FIG. 5, there is no limit to the number of keywords set for each category, but it is prohibited to set duplicate keywords in multiple categories.

Figure 6 is an example of second document information 113. The second document information 113 illustrated in Figure 6 is an internal document stored in a company's information processing system, and is a product catalog document that describes the detailed contents of each of multiple products sold (provided) by the company. Figure 6 illustrates second document information 113a and 113b as product catalog documents for each product.

The second document processing unit 130 extracts predetermined named entities from the smallest unit documents (second document information 113a, 113b) in the second document information 113, and generates (creates) the second related word table 118 by combining related named entities from among the extracted named entities. However, this method differs from the method in which the first document processing unit 120 generates the first related word table 116 from the first document information 111 in that the generation of the second related word table 118 by the second document processing unit 130 does not require the acquisition of progress information from the second document information 113a, 113b.

7 is a diagram for explaining an image of the second related word table 118 included in the second related word information 114. As described above, the second related word table 118 is generated (created) by the second document processing unit 130. To explain in more detail, first, the named entity feature quantity extraction unit 131 of the second document processing unit 130 extracts predetermined named entities (specifically, for example, "remote", "sensor", "operating state", "wireless communication", "site", "real time", "employee", "productivity", "office", etc.) from the documents of the second document information 113 (second document information 113a, 113b). Then, the related word table generation unit 133 of the second document processing unit 130 uses the inference model learned by the learning inference unit 132 to compile related named entities from the extracted named entities, thereby generating the second related word table 118.

The second related word table 118 generated as described above corresponds to a "related word dictionary" which collectively refers to combinations of named entities that are used similarly (synonym relationships) and combinations of named entities that frequently appear in the same sentence (co-occurrence relationships) in the target second document information 113a and 113b. The network graph 1181 shown in FIG. 7 is a network graph in which the related words contained in the second related word table 118 (related word dictionary) in the second document information 113a are expanded. According to the network graph 1181, the relevance of the named entities used in the second document information 113a is represented in a form that is easy to visually recognize.

(3) Linking Related Words Between Different Documents The item processing unit 140 of the document information extraction device 1 links related words between the first document information 111, which is input information, and the second document information 113, which is a candidate for output information, and selects candidates for recommended information (recommendation candidates) to be proposed to the user, taking into consideration the progress status included in the input information. Note that in this embodiment, even when the first document information 111 and the second document information 113 are different document types, related words between the documents can be linked. Below, an image of selecting recommendation candidates based on the linking of related words between different documents will be described.

Figure 8 is a diagram for explaining the concept of linking related words between different documents. The network graph 210 shown in Figure 8 is constructed by forming a network graph of related words in each of the first document information 111a and the second document information 113a, and connecting related words (common terms) common to both.

Specifically, in the network graph 210 shown in FIG. 8, the network graph 211 shown mainly on the left side is a network graph showing related words of the first document information 111a based on the first related word table 116 (similar to the network graph 1161 in FIG. 4), and the network graph 212 shown mainly on the right side is a network graph of related words of the second document information 113a based on the second related word table 118 (similar to the network graph 1181 in FIG. 7). According to the network graph 210 in FIG. 8, the words "work style" and "IoT" are related words (common terms) common to both the first document information 111a and the second document information 113a, and therefore the network graphs 211 and 212 can be connected by these common terms.

At this time, the phrase "remote work" can be reached from the phrase "manual work" contained in the sentence of the first document information 111a via two routes: "manual work - automation - IoT - remote work" and "manual work - worker - work style - work efficiency - centralized management - remote work". As a result, the item processing unit 140 can select the second document information 113a containing the phrase "remote work" in the sentence as a recommendation candidate. In this way, the item processing unit 140 of the document information extraction device 1 can link related words between the first document information 111a and the second document information 113a, which are heterogeneous documents, to select recommendation candidates.

Figure 9 is a diagram explaining the related word linkage between the different documents shown in Figure 8 in the form of a table. The related word linkage table 220 shown in Figure 9 is a table summarizing the process of related words from "manual work" to "remote work" in the first route of "manual work - automation - IoT - remote work" explained in Figure 8 and the second route of "manual work - worker - work style - work efficiency - centralized management - remote work". According to Figure 9, it can be seen that in the first route, a recommendation candidate for "remote work" can be selected from "manual work" written in the first document information 111a with three related words, while in the second route, five related words are required to select a recommendation candidate for "remote work" from "manual work".

Figure 10 is a diagram showing an example of a recommendation candidate calculation table 230. The recommendation candidate calculation table 230 is a table that shows the number of related words required to select a recommendation candidate in related word linkage between different documents, based on distance.

For example, when the first document information 111a is used as input information, the recommendation candidate calculation table 230 in FIG. 10 shows how many ways the recommendation candidates for multiple product names 231 included in the second document information 113 are linked to the term "manual work" included in the first document information 111a (number of connections 232) and the total value (score 233) of the product of the number of connections and the distance.

Generally speaking, the closer the link between the words in the input information (e.g., "manual work") and the words in the recommendation candidate (e.g., "remote operation") (i.e., the less related words required), the more related the recommendation candidate is to the input information. On the other hand, the farther the link between the words in the input information and the words in the recommendation candidate (i.e., the more related words required), the more the recommendation candidate appears to be a product that is not related to the input information at first glance. However, even if a recommendation candidate appears to be unrelated at first glance, if it is connected through many routes (connected through a large number of connections), it is highly worth recommending to users and is a product that is highly unexpected.

From the above viewpoint, in the recommendation candidate calculation table 230, by calculating the total value of the product of the number of connections and the distance as the score 233, even if the product has a low relevance, if the score 233 is high, it can be determined that the product is highly unexpected and worth recommending. For example, in the case of the recommendation candidate calculation table 230 of FIG. 10, a product called "energy saving" that can be connected at the shortest distance of "distance #1" can be determined to be a highly relevant recommendation candidate. On the other hand, a product called "AI introduction" that can only be connected at the shortest distance of "distance #4" is a recommendation candidate with a low relevance. However, since the product "AI introduction" can be connected at "distance #4" and "distance #5", it has a high score value of "9", and therefore can be determined to be a product with a high relevance and a high recommendation value, for example, compared to the product "remote operation" with a score value of "8".

As described above, in this embodiment, even in the case of related word linkage between different documents, the recommendation candidate calculation table 230 is used to take into consideration the number of items that can be connected through related word linkage between different documents and the distance between words required to connect them, making it possible to set priorities based on relevance (or unexpectedness) when selecting recommendation candidate items.

(4) Recommendation Calculation Process The items that a user wants to be recommended change from moment to moment depending on the progress of sales activities. Therefore, hereinafter, a recommendation calculation process in which the document information extraction device 1 according to the present embodiment proposes recommended items depending on the progress of sales activities will be described.

In this description, one of the documents (such as the first document information 111a to 111c) included in the first document information 111 is used as input information, and a recommendation calculation process is executed to select multiple recommendation candidates from each document (such as the second document information 113a, 113b, etc.) included in the second document information 113 with a priority order, thereby proposing suitable recommended items to the user. Note that it is assumed that the first related word information 112 based on the first document information 111 (including the first document information 111a) and the second related word information 114 based on the second document information 113 (including the second document information 113a) have already been generated when the following process begins.

FIG. 11 is a flowchart showing an example of the processing procedure of the recommendation calculation process. The recommendation calculation process is executed by the item processing unit 140. The recommendation calculation process is started, for example, when the document information extraction device 1 receives a user operation requesting the execution of information extraction from the first document information 111 (hereinafter, the first document information 111 from which information is extracted is referred to as the first document information 111 to be extracted). More specifically, in the recommendation calculation process, when the recommendation execution confirmation button 245 on the activity report registration screen 240 shown in FIG. 15 is pressed, the processing of step S110 is executed, and when the recommendation execution button 253 on the recommendation confirmation screen 250 shown in FIG. 16 is pressed, the processing from step S120 onward is executed.

According to FIG. 11, first, the compatibility calculation unit 141 of the item processing unit 140 creates the recommendation candidate calculation table 230 (step S110). The recommendation candidate calculation table 230 is as described in detail in FIG. 10. In addition, since the recommendation candidate calculation table 230 is used in the relevance priority designation process described below, the process of creating the recommendation candidate calculation table 230 may be executed during the relevance priority designation process shown in FIG. 12 (for example, between steps S112 and S113).

Next, the compatibility calculation unit 141 executes a relevance priority designation process using the progress category table 117 and the recommendation candidate calculation table 230 to select items to be recommendation candidates based on their relevance (or unexpectedness) to the first document information 111 (e.g., a sales daily report) that is the input information (step S120). The detailed processing procedure of the relevance priority designation process will be described later with reference to FIG. 12.

Next, the compatibility calculation unit 141 executes an evaluation value priority designation process that uses the progress category table 117 and the recommendation history information 115 to select items to be recommended candidates based on the access level (evaluation value) of the items (step S130). The detailed processing procedure of the evaluation value priority designation process will be described later with reference to FIG. 13.

Next, the compatibility calculation unit 141 executes a past proposal priority designation process that uses the progress category table 117 and the recommendation history information 115 to select items to be recommended candidates based on whether or not the items have been recommended in the sales activities of the same case (step S140). The detailed processing procedure of the past proposal priority designation process will be described later with reference to FIG. 14.

Note that the order in which steps S120 to S140 are executed is not limited to the example in FIG. 11 and may be changed as appropriate.

Then, the item recommendation unit 142 determines the priority order of the recommended items based on the execution results of each priority designation process of steps S120 to S140 (step S150). In detail, in step S150, the item recommendation unit 142 selects and prioritizes each item included in the second document information 113 based on the selection conditions for the recommended candidates specified in each priority designation process of steps S120 to S140, and determines, starting from the item with the highest priority, as many items as can be displayed in the recommendation result column 254 of the recommendation confirmation screen 250 described later in FIG. 16 as recommendation candidates to be proposed to the user.

Next, the item output unit 143 displays the recommendation candidates (recommended items) in the recommendation result column 254 of the recommendation confirmation screen 250 in a form that reflects the priority order determined in step S150 (step S160), and the recommendation calculation process ends.

Figure 12 is a flowchart showing an example of the processing procedure for the relatedness priority designation process. As described above, the relatedness priority designation process is the process of step S120 in Figure 11, and is executed by the compatibility calculation unit 141 of the item processing unit 140.

According to FIG. 12, first, the compatibility calculation unit 141 refers to the information in the progress category table 117 and obtains a record (progress category information) corresponding to the category of the first document information 111 to be extracted (step S121).

Next, the compatibility calculation unit 141 determines whether or not the recommendation priority rule 1172 of the progress category information acquired in step S121 specifies the relevance (or unexpectedness) between the first document information 111 to be extracted and the recommended item (step S122). Specifically, when the target recommendation priority rule 1172 describes a significant specification for the classification (A) illustrated in FIG. 5, the compatibility calculation unit 141 can determine that the relevance (or unexpectedness) between the first document information 111 to be extracted and the recommended item is specified. If the relevance (or unexpectedness) with the recommended item is specified (YES in step S122), proceed to step S123. On the other hand, if the relevance (or unexpectedness) with the recommended item is not specified (NO in step S122), the relevance priority specification process is terminated without specifying the selection conditions for the recommendation candidates based on relevance or unexpectedness.

In step S123, the compatibility calculation unit 141 determines whether the target recommendation priority rule 1172 specifies content equivalent to "prioritize relevance."

If a specification equivalent to "prioritize relevance" is made in step S123 (YES in step S123), the compatibility calculation unit 141 specifies a selection condition that prioritizes highly relevant items as recommendation candidates (step S124), and ends the relevance priority specification process. To explain the process of step S124 in detail, the compatibility calculation unit 141 can associate the degree of relevance with the selection conditions for recommendation candidates by, for example, referring to the recommendation candidate calculation table 230 and setting priorities for multiple items (e.g., products) that are to be recommended candidates, such as by attaching flags to items with high (large) relevance.

On the other hand, if no content equivalent to "prioritize relevance" is specified in step S123, in other words if a content equivalent to "prioritize unexpectedness" is specified (NO in step S123), the compatibility calculation unit 141 specifies a selection condition that prioritizes highly unexpected items as recommendation candidates (step S125), and ends the relevance priority specification process. To explain the process of step S125 in detail, the compatibility calculation unit 141 can associate the degree of unexpectedness with the selection conditions for recommendation candidates by, for example, referring to the recommendation candidate calculation table 230 and setting priorities for multiple items (e.g., products) that are to be recommended candidates, such as by attaching a flag to items that are more unexpected (larger).

Figure 13 is a flowchart showing an example of the processing procedure for the evaluation value priority designation process. As described above, the evaluation value priority designation process is the process of step S130 in Figure 11, and is executed by the compatibility calculation unit 141 of the item processing unit 140.

According to FIG. 13, first, the compatibility calculation unit 141 refers to the information in the progress category table 117 and acquires a record (progress category information) corresponding to the category of the first document information 111 to be extracted (step S131). Next, the compatibility calculation unit 141 acquires the recommendation history information 115 (step S132).

Next, the compatibility calculation unit 141 determines whether or not the evaluation value of the recommended item is specified in the recommendation priority rule 1172 of the progress category information acquired in step S131 (step S133). Specifically, the compatibility calculation unit 141 can determine that the evaluation value of the recommended item is specified when the target recommendation priority rule 1172 contains significant specification content for the category (B) illustrated in FIG. 5. If the evaluation value of the recommended item is specified (YES in step S133), the process proceeds to step S134. On the other hand, if the evaluation value of the recommended item is not specified (NO in step S133), the evaluation value priority specification process is terminated without specifying the selection conditions for the recommendation candidates based on the evaluation value.

In step S134, the compatibility calculation unit 141 determines whether the target recommendation priority rule 1172 specifies content equivalent to "give priority to highly rated items."

If a content equivalent to "give priority to high ratings" is specified in step S134 (YES in step S134), the compatibility calculation unit 141 specifies a selection condition that prioritizes items with high ratings as recommendation candidates based on the rating value of each item included in the recommendation history information 115 (step S135), and ends the rating value priority specification process. Here, the rating value of an item included in the recommendation history information 115 is a value that changes depending on the recommendation status of the item, and is determined based on, for example, the number of times the item has been searched for by a user or the number of times the user pressed the "helpful" button when the item was recommended in the past. In other words, an item with a high rating value means that it is a useful (highly rated) item when recommended. Therefore, in step S135, the compatibility calculation unit 141 can, for example, refer to the recommendation history information 115 and set priorities for multiple items (e.g., products) that are to be recommendation candidates by attaching flags to items with higher evaluation values, thereby associating the high evaluation values with the selection criteria for recommendation candidates.

On the other hand, if no specification equivalent to "give priority to high ratings" has been made in step S134, in other words, if a specification equivalent to "give priority to low ratings" has been made (NO in step S134), the compatibility calculation unit 141 specifies a selection condition that prioritizes items with low rating values as recommendation candidates (step S136), and ends the rating value priority specification process. A detailed explanation of the specific process of step S136 will be omitted, since it is sufficient to set priorities using the opposite criteria to those in step S135.

Figure 14 is a flowchart showing an example of the processing procedure for the past proposal priority designation process. As described above, the past proposal priority designation process is the process of step S140 in Figure 11, and is executed by the compatibility calculation unit 141 of the item processing unit 140.

According to FIG. 14, first, the compatibility calculation unit 141 refers to the information in the progress category table 117 and acquires a record (progress category information) corresponding to the category of the first document information 111 to be extracted (step S141). Next, the compatibility calculation unit 141 acquires the recommendation history information 115 (step S142).

Next, the compatibility calculation unit 141 determines whether or not the recommendation priority rule 1172 of the progress category information acquired in step S141 specifies the past proposal of the recommended item (step S143). Specifically, the compatibility calculation unit 141 can determine that the past proposal of the recommended item is specified when the target recommendation priority rule 1172 describes significant specification content for the category (C) illustrated in FIG. 5. If the past proposal of the recommended item is specified (YES in step S143), the process proceeds to step S144. On the other hand, if the past proposal of the recommended item is not specified (NO in step S143), the past proposal priority specification process is terminated without specifying the selection conditions for the recommendation candidates based on the past proposals.

In step S144, the compatibility calculation unit 141 determines whether the target recommendation priority rule 1172 specifies content equivalent to "give priority to unproposed items."

If a specification equivalent to "give priority to unproposed items" is made in step S144 (YES in step S144), the compatibility calculation unit 141 refers to the recommendation proposal history included in the recommendation history information 115, specifies a selection condition that prioritizes unproposed items that are not recorded in the recommendation proposal history for this case as recommendation candidates (step S145), and terminates the past proposal priority specification process. Here, the recommendation proposal history included in the recommendation history information 115 is information that accumulates the history of items (recommended items) proposed in previous sales activities for each case. In other words, items that are not recorded in the recommendation proposal history for this case are unproposed items that have never been proposed in past sales activities for this case. Therefore, in step S145, the compatibility calculation unit 141 can associate the presence or absence of past proposals with the selection conditions for recommendation candidates by, for example, setting a priority by flagging items that are not recorded in the recommendation proposal history of the subject case, which is included in the recommendation history information 115.

On the other hand, if a specification equivalent to "give priority to unproposed items" is not made in step S144, for example, if a specification equivalent to "give priority to proposed items" is made (NO in step S144), the compatibility calculation unit 141 refers to the recommendation proposal history included in the recommendation history information 115, specifies a selection condition that the proposed items recorded in the recommendation proposal history for this case are given priority as recommendation candidates (step S146), and ends the evaluation value priority specification process. A detailed explanation of the more specific process of step S146 is omitted, since it is sufficient to set the priority based on the opposite criteria to step S145.

(5) Example of Output Screen In the following, a specific example of the output screen that is displayed when registering a daily sales report or selecting recommended items for the registered daily sales report in the document information extraction device 1 according to this embodiment will be described.

Figure 15 is an example of an activity report registration screen 240. The activity report registration screen 240 is an example of a screen presented to the user via the output device 15 when the user registers the daily sales report, which is input information, in the first document information 111, and is realized, for example, by a GUI (Graphical User Interface) that can be operated by the user via the input device 14.

As shown in FIG. 15, the activity report registration screen 240 includes a selection field 241 in which the name of the sales activity case can be selected, an upload button 242 for uploading the daily sales report to the document information extraction device 1, a case registration button 243 for registering the case, an activity list information field 244 for displaying a list of sales activities for the case selected in the selection field 241, and a recommendation execution confirmation button 245 for selecting recommended items.

On the activity report registration screen 240, the user first selects in the selection field 241 the sales activity case to which the daily sales report to be registered belongs, and a list of sales activities for the selected case is displayed in the activity list information field 244. In the example of FIG. 15, the case "Smartification of XX Steel's XX Factory" has been selected.

Next, when the upload button 242 is pressed, for example, a file selection screen is displayed, and the daily sales report that the user has entered in advance can be selected as the item to be uploaded. By selecting the desired daily sales report file on the file selection screen, the daily sales report is uploaded to the document information extraction device 1.

When the case registration button 243 is pressed, the uploaded daily sales report is registered in the first document information 111 as an activity report for the selected case. In the example of FIG. 15, the daily sales report titled "XX Steel Works Case 6" is registered in the case "Smartification of XX Steel Works XX Factory."

As described above, the activity list information field 244 displays a list of sales activities for the case selected in the selection field 241. The "Activity History" field in the activity list information field 244 displays information about the sales daily reports registered for the case, and the user can select one of the sales daily reports. In the example of Figure 15, the sales daily report with the title "XX Steelworks Case 6" has been selected.

When the recommendation execution confirmation button 245 is pressed with any of the daily sales reports selected in the activity list information field 244, the item processing unit 140 is called with the selected daily sales report as input information. At this time, the called item processing unit 140 (item output unit 143) displays a recommendation confirmation screen 250 on the output device 15 as a confirmation screen for performing the recommendation calculation process described above and outputting an optimal recommendation from the daily sales report that has been used as input information. In the example of Figure 15, when the recommendation execution confirmation button 245 is pressed, the recommendation confirmation screen 250 is displayed regarding the execution of a recommendation for the daily sales report with the title "XX Steelmaking Project 6".

Figure 16 is an example of a recommendation confirmation screen 250. The recommendation confirmation screen 250 is an example of a screen presented to the user via the output device 15 when the recommendation execution confirmation button 245 is pressed on the activity report registration screen 240 illustrated in Figure 15, and is realized, for example, by a GUI (Graphical User Interface) that can be operated by the user via the input device 14.

As shown in FIG. 16, the recommendation confirmation screen 250 includes an information field 251 that displays information about the recommendations for the selected daily sales report, a rule change button 252 for manually switching the recommendation priority rules, a recommendation execution button 253 for executing the recommendation calculation process, a recommendation result field 254 that displays the recommendation results, and a recommendation item selection button 255 for selecting an item that the user likes (or is interested in) from among the recommended items displayed in the recommendation result field 254.

In the recommendation confirmation screen 250, the information field 251 automatically displays the results of the compatibility calculation unit 141 performing document analysis on the selected daily sales report by referring to the progress category table 117. Specifically, the information field 251 displays the title of the selected daily sales report, "XX Steelworks Project 6," the progress category of "Difficulty" to which the daily sales report falls (see category name 1171 in FIG. 5), and the recommendation priority rule corresponding to that progress category (see recommendation priority rule 1172 in FIG. 5).

The rule change button 252 is provided to manually change the recommendation priority rule displayed in the information column 251 according to the automatic classification of the progress category according to the user's wishes, and when the rule change button 252 displaying "Change" is pressed, the user can perform an operation to change the recommendation priority rule defined corresponding to the progress category to a recommendation priority rule of his/her choice. After the recommendation priority rule has been changed once, the rule change button 252 may display "Back" as shown in FIG. 16, for example, and when the rule change button 252 displaying "Back" is pressed, a process such as returning to the defined recommendation priority rule may be performed.

When the recommendation execution button 253 is pressed, the item processing unit 140 (suitability calculation unit 141, item recommendation unit 142, item output unit 143) executes a recommendation calculation process based on the recommendation priority rules displayed in the information field 251, and the recommended items selected in accordance with the recommendation priority rules are displayed in the recommendation result field 254. The item processing unit 140 updates the recommendation proposal history included in the recommendation history information 115 for each of the multiple recommended items displayed in the recommendation result field 254.

In the recommendation result column 254 of FIG. 16, multiple recommended items (product names in this example) selected according to the recommendation priority order that complies with the recommendation priority rules displayed in the information column 251 are visualized in a graph. In more detail, the vertical axis of the graph visualizes items with higher recommendation priority according to their "recommended level," and the horizontal axis of the graph displays items that are highly related to the selected daily sales report (input information) document (or items that are highly unexpected) in order of the degree of relatedness (or unexpectedness).

In the case of FIG. 16, the progress category to which the selected daily sales report of "XX Steel Project 6" corresponds is "Difficult", and generally, it is considered effective to propose a product that is highly unexpected. Here, the most unexpected item displayed in the recommendation result column 254 is "Work Style Reform Product", but its "Recommendation Level" is only in the top two. On the other hand, "Remote Product" is the third most unexpected item, but is the item with the highest "Recommendation Level", and is shown to be the most effective recommended item overall. The reason why the "Recommendation Level" of "Remote Product" is high is thought to be largely due to the influence of "Highly Rated Item Priority" set in the recommendation priority rule of "(B) How to Use History Information". In addition, since the above-mentioned "Work Style Reform Product" has the second highest "Recommendation Level", this item can also be considered to be an effective recommended item.

In addition, the recommended item selection button 255 provided for each recommended item displayed in the recommendation result column 254 is pressed when the user likes (or is interested in) the corresponding item, thereby contributing to an improvement in the evaluation value of the target item. Specifically, for example, each time the recommended item selection button 255 is pressed, the item processing unit 140 can increase the evaluation value of the target item included in the recommendation history information 115 by a predetermined number of points, or when the recommended item selection button 255 is pressed, another window is opened in which a button indicating "helpful" is displayed, and when the button indicating "helpful" is pressed, the item processing unit 140 can increase the evaluation value of the target item included in the recommendation history information 115 by a predetermined number of points. Also, for example, when the recommended item selection button 255 is pressed, detailed information on the target recommended item is displayed, and the item processing unit 140 may add to the evaluation value of the target item included in the recommendation history information 115 depending on the length of the display period (viewing time). Also, the evaluation value of each item included in the recommendation history information 115 is not limited to always increasing, and may be reduced by deducting points if the display history is added but not viewed.

In any case, the item processing unit 140 changes and updates the evaluation value of each item included in the recommendation history information 115 based on a user operation on the recommended item selection button 255, so that in the recommendation calculation process executed subsequently, the latest evaluation value can be reflected in the specification of the selection conditions for recommendation candidates by the evaluation value priority specification process (step S130 in FIG. 11, see FIG. 13). As a result, the document information extraction device 1 can add evaluation values reflecting the user's preferences and selection tendencies to the conditions and extract more suitable recommended information.

As described above, the document information extraction device 1 according to this embodiment enables information linkage between two or more types of documents (first document information 111, second document information 113) with different uses and content by using related word extraction technology. In the case of document data from which progress information on sales activities, etc. can be obtained from an input document (first document information 111) input in chronological order, it becomes possible to propose optimal recommended items according to the progress of activities in the input document.

Even if the first document information 111, which is the input document, and the second document information 113, which is the target of recommendation, are heterogeneous documents with completely different written contents and uses, the document information extraction device 1 according to this embodiment enables information linkage through related word extraction technology, so that it is possible to extract and propose suitable recommended information (recommended items) taking into account the progress status contained in the input information. Specifically, for example, it is possible to grasp the history and progress of sales activities from sales reports prepared by sales representatives, and easily search for optimal items from product catalog documents.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and it goes without saying that various modifications can be made without departing from the gist of the present invention. For example, the above embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to having all of the configurations described. In addition, it is possible to add, delete, or replace some of the configurations of the above embodiment with other configurations. Specifically, for example, in the above embodiment, the second document information 113 including candidate information of recommended items recommended to the user was described using a product catalog document as an example, but the second document information 113 of the present invention is not limited to a product catalog document, and can be applied to various document formats such as knowledge, patents, and papers.

Furthermore, the above-mentioned configurations, functional units, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits. Furthermore, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information on the programs, tables, files, etc. that realize each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), an IC card, an SD card, a DVD, or other recording media.

Furthermore, the layout of the various functional units, various processing units, and various databases of each information processing device described above is merely an example. The layout of the various functional units, various processing units, and various databases can be changed to an optimal layout in terms of the performance, processing efficiency, communication efficiency, etc. of the hardware and software equipped in these devices.

In addition, the configuration (schema, etc.) of the database that stores the various types of data mentioned above can be flexibly changed from the perspective of efficient use of resources, improved processing efficiency, improved access efficiency, improved search efficiency, etc.

In addition, the control lines and information lines shown in the drawings are those that are considered necessary for explanation, and not all control lines and information lines on the product are necessarily shown. In implementation, it can be assumed that almost all components are interconnected.

1 Document information extraction device 10 Information processing device 11 Processor 12 Main memory device 13 Auxiliary memory device 14 Input device 15 Output device 16 Communication device 110 Memory unit 111 First document information 112 First related word information 113 Second document information 114 Second related word information 115 Recommendation history information 116 First related word table 117 Progress category table 118 Second related word table 120 First document processing unit 121 Named entity feature extraction unit 122 Learning inference unit 123 Related word table generation unit 124 Progress category generation unit 130 Second document processing unit 131 Named entity feature extraction unit 132 Learning inference unit 133 Related word table generation unit 140 Item processing unit 141 Compatibility calculation unit 142 Item recommendation unit 143 Item output unit 210 Network graph 220 Related word link table 230 Recommendation candidate calculation table 240 Activity report registration screen 250 Recommendation confirmation screen

Claims (12)

A document information extraction device that proposes recommended items by performing document analysis on activity records that are recorded in chronological order for the same case,
a storage unit that stores first document information in which the activity record is registered, and second document information that is configured in a document type different from that of the first document information and in which document data related to the candidate recommended items is registered;
a progress determination unit that classifies the progress status of the activity record into one of a plurality of progress categories based on a keyword that indicates a progress status and is included in the activity record of the first document information;
a first document processing unit that extracts first related terms constituting related named entities from the first document information and generates first related term information that compiles the extracted first related terms;
a second document processing unit that extracts second related terms constituting related named entities from the second document information and generates second related term information that compiles the extracted second related terms;
an item processing unit that selects and outputs, from among the recommended item candidates, a recommended item having a second related word that is directly or indirectly connected to a first related word included in the activity record of the first document information, based on a selection condition for a recommended item according to the progress category classified by the progress determination unit and a result of related word linkage in which the first document information and the second document information are linked by connecting the first related word information and the second related word information with a common related word;
A document information extraction device comprising: a predetermined keyword corresponding to each of a plurality of progress categories representing the progress status is determined in advance;
The document information extraction device according to claim 1, characterized in that the progress determination unit extracts the keywords from each of the activity records included in the first document information, and classifies the progress status of each of the activity records into one of the multiple progress categories based on the extracted keywords. a predetermined priority rule is prepared in advance, which specifies what attributes of the recommended items should be preferentially recommended for each of the multiple stages of progress categories;
The document information extraction device according to claim 2 , wherein the item processing unit specifies a selection condition according to the priority rule corresponding to the progress category classified by the progress determination unit, and selects the recommended item from the candidate recommended items. The document information extraction device according to claim 3, characterized in that the priority rules can be changed in response to a user operation, and when the user operation is performed, the item processing unit specifies selection conditions in accordance with the changed priority rules and selects the recommended items from the candidate recommended items. The priority rule may include a relevance priority designation that specifies that the relevance to the activity record is reflected in the selection conditions of the recommended items,
When the priority rule includes the relevance priority designation, the item processing unit:
determining the relevance of the recommended item candidate having the second related term based on a distance in the related term linkage between the first related term included in the activity record and the second related term included in the second document information;
The document information extraction device according to claim 3 , further comprising: setting a priority of each of the candidates for the recommended items in the selection based on a result of the judgment. The priority rule may include a surprise priority designation that designates that the surprise of the activity record is reflected in the selection conditions of the recommended items,
When the priority rule includes the unexpectedness priority designation, the item processing unit:
determining the unexpectedness of the recommended item candidate having the second related term based on a distance and a number of connections in the related term linkage between the first related term included in the activity record and the second related term included in the second document information;
The document information extraction device according to claim 3 , further comprising: setting a priority of each of the candidates for the recommended items in the selection based on a result of the judgment. The storage unit further stores recommendation history information in which information about the past recommended items of the job selected by the item processing unit is recorded,
The priority rule may include a past proposal priority specification that specifies that the presence or absence of an output history of the item is reflected in the selection conditions of the recommended items for the item,
The document information extraction device according to claim 3, characterized in that, when the priority rule includes the past proposal priority designation, the item processing unit sets a priority of each candidate of the recommended items in the selection based on whether or not it is recorded in the recommendation history information. The recommendation history information holds an evaluation value calculated by a predetermined method based on a user's access results for each of the past recommended items of the case selected by the item processing unit, and
The priority rule may include an evaluation value priority specification that specifies that the evaluation value is to be reflected in a selection condition for the recommended item,
The document information extraction device according to claim 7, characterized in that, when the priority rule includes the evaluation value priority specification, the item processing unit sets a priority of each candidate of the recommended items in the selection based on the evaluation value in the recommendation history information. A document information extraction method for a document information extraction device that performs document analysis of activity records recorded in chronological order for the same case to propose recommended items, comprising:
a first document information storage step of registering the activity record in first document information;
a second document information storage step of registering document data related to the candidate recommended items in second document information composed of a document type different from that of the first document information;
a progress determination step of classifying the progress of the activity record into one of a plurality of progress categories based on a keyword suggesting the progress, the keyword being included in the activity record of the first document information;
a first document processing step of extracting first related terms constituting related named entities from the first document information and generating first related term information that compiles the extracted first related terms;
a second document processing step of extracting second related terms constituting related named entities from the second document information and generating second related term information that compiles the extracted second related terms;
After the first document processing step and the second document processing step are executed , in an item processing step, a step of selecting and outputting, from among the recommended item candidates, a recommended item having a second related word directly or indirectly connected to a first related word included in the activity record of the first document information, based on a selection condition for a recommended item corresponding to the progress category classified in the progress determination step and a result of related word linkage in which the first document information and the second document information are linked by connecting the first related word information and the second related word information with a common related word ;
A document information extraction method comprising: a predetermined keyword corresponding to each of a plurality of progress categories representing the progress status is determined in advance;
The document information extraction method according to claim 9, characterized in that in the progress determination step, the keywords are extracted from each of the activity records included in the first document information, and the progress status of each of the activity records is classified into one of the multiple progress categories based on the extracted keywords. a predetermined priority rule is prepared in advance, which specifies what attributes of the recommended items should be preferentially recommended for each of the multiple stages of progress categories;
The document information extraction method according to claim 10, characterized in that in the item processing step, a selection condition according to the priority rule corresponding to the progress category classified in the progress determination step is specified, and the recommended item is selected from the candidate recommended items. The document information extraction method according to claim 11, characterized in that the priority rules can be changed in response to a user operation, and when the user operation is performed, in the item processing step, selection conditions are specified in accordance with the changed priority rules, and the recommended items are selected from the candidate recommended items.

Images