Data management device, data management method, and program

Abstract

We provide a data management device that can improve the usability of FAQ systems. [Solution] A data management device for managing question sentences and answer sentences to question sentences, comprising a control unit 110 and a communication unit 140. The control unit acquires a plurality of text data, vectorizes the plurality of text data to create a plurality of text vectors, extracts a plurality of question sentences from the plurality of text data based on the plurality of text vectors, clusters the extracted plurality of question sentences into question clusters containing mutually similar question sentences based on the plurality of question sentence vectors which are the plurality of text vectors corresponding to the plurality of question sentences, determines a representative question sentence for each question cluster that represents the question sentences belonging to the question cluster corresponding to the center of the question cluster, and the communication unit transmits registration recommendation information recommending that the determined representative question sentence be registered in the storage device to an administrator terminal used by the administrator who manages the data management device.

Description

Technical Field

[0001] This disclosure relates to a data management device, a data management method, and a program.

Background Art

[0002] Conventionally, there has been known a FAQ system that creates Frequently Asked Questions (FAQ) by pattern recognition of question candidates (Q candidates) and answer candidates (A candidates) from dialogue logs. As a pattern recognition model for pattern recognition here, for a morphological analysis of a dialogue log in which a series of conversations between a questioner and an answerer are taken as text, the number of morphemes per utterance, which is the number of morphemes per utterance calculated for each record, which is an individual utterance constituting a series of conversations, the number of occurrences of common keywords, which is the number of occurrences of words common before and after the utterance, the total keyword importance, which is the sum of the importance of peripheral keywords per utterance, and the number of filler words existing within 1 to 5 utterances after the utterance. A pattern recognition model is known in which a classifier is trained by supervised learning with a feature value including at least one of these features as input for each record and a candidate label indicating whether it is a Q candidate corresponding to a question by a questioner or an A candidate corresponding to an answer by an answerer as output (see Patent Document 1).

[0003] Furthermore, information processing devices are known that perform document clustering to assist in reaching the target document through full-text search results. Such information processing devices include: a document search means that performs a full-text search of documents to find search documents that match the search criteria; a document scoring means that orders the search documents according to a relevance score that represents the degree of relevance to the search criteria; a feature word file database in which a document identifier for identifying the document, a feature word extracted from the full-text data of the document, and a weight value representing the weight of the feature word are registered in association with each other; and a document clustering means that uses the document identifier of the search document as a key and performs clustering processing on the search documents based on the feature word obtained from the feature word file database and the weight value of the feature word. An information processing device is known that creates a list of search results and includes a document grouping means that groups the search documents based on the relevance score, and the document clustering means performs clustering processing on the search documents within each group grouped by the document grouping means (see Patent Document 2). [Prior art documents] [Patent Documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-71690 [Patent Document 2] Japanese Patent Publication No. 2010-9577 [Overview of the project] [Problems that the invention aims to solve]

[0005] In the case of an FAQ system using the pattern recognition model described in Patent Document 1, and when the information processing device described in Patent Document 2 is applied to the FAQ system, the usability of the FAQ system is insufficient, and there is room for improvement.

[0006] This disclosure provides a data management device, a data management system, a data management method, and a program that can improve the usability of an FAQ system. [Means for solving the problem]

[0007] One aspect of the present disclosure is a data management system for managing question texts and answers to said question texts, comprising a data management device, an administrator terminal, and an FAQ knowledge base, wherein the data management device comprises a control unit and a communication unit, the control unit uses a natural language processing model to detect personal information contained in a question text Q, performs an anonymization process to delete the detected personal information, and generates an anonymized question text Q which is the question text Q that does not contain the personal information, the communication unit transmits anonymization information indicating that the anonymization process has been performed, the question text Q containing the personal information, and the anonymized question text Q to the administrator terminal, the administrator terminal accepts input specifying which of the question text Q and the anonymized question text Q should be included in the representative QA data to be registered in the FAQ knowledge base, and the communication unit transmits registration recommendation information to the administrator terminal recommending that the representative QA data including the specified question text be registered in the FAQ knowledge base.

[0008] One aspect of this disclosure is a data management method for managing question texts and answer texts to said question texts, comprising: using a natural language processing model to detect personal information contained in a question text Q; performing an anonymization process to delete the detected personal information to generate an anonymized question text Q which is a question text Q that does not contain the personal information; transmitting anonymized information indicating that the anonymization process has been performed, the question text Q containing the personal information, and the anonymized question text Q from a data management device that manages said question texts and answer texts to an administrator terminal used by an administrator who manages the data management device; the administrator terminal receiving input specifying which of the question text Q and the anonymized question text Q should be included in representative QA data to be registered in an FAQ knowledge base; and transmitting registration recommendation information from the data management device to the administrator terminal recommending that representative QA data including the specified question text be registered in the FAQ knowledge base. [Effects of the Invention]

[0009] This disclosure can improve the usability of the FAQ system. [Brief explanation of the drawing]

[0010] [Figure 1] A schematic diagram showing an example configuration of the FAQ processing system according to the first embodiment of this disclosure. [Figure 2] Block diagram showing an example configuration of an FAQ server. [Figure 3] Diagram illustrating the overview of the FAQ chat screen. [Figure 4] Sequence diagram showing the operation overview of the FAQ processing system. [Figure 5] A diagram showing an example of an inquiry email. [Figure 6] A diagram showing an example of how inquiry emails can be classified. [Figure 7] This diagram shows an example of clustering of derived question sentences and recommended information for registering QA data corresponding to the central data of each cluster. [Figure 8]Flowchart showing an operation example related to QA registration support for a FAQ knowledge base by a FAQ server [Figure 9] Flowchart showing an example of QA text extraction processing [Figure 10] Flowchart showing another example of extracting representative answer sentences corresponding to representative question sentences [Figure 11A] Block diagram showing a configuration example of a FAQ server according to a second embodiment of the present disclosure [Figure 11B] Diagram showing an image of a FAQ chat screen where questions are narrowed down [Figure 12] Diagram showing an image of clustering in the second embodiment [Figure 13] Diagram showing an example of a search result (candidate) including a Q number and a question candidate sentence [Figure 14] Diagram showing an example of the result of vectorizing a question candidate sentence [Figure 15] Diagram showing an example of clustering vectors and each cluster and a question candidate sentence [Figure 16] Diagram showing an example of the result of vectorizing question candidate sentences for each cluster [Figure 17] Diagram showing an example of characteristic words extracted for each cluster [Figure 18] Diagram showing an example of a FAQ chat screen considering question narrowing [Figure 19] Network diagram showing an example of the relationship between each characteristic word [Figure 20] Block diagram showing a configuration example of a FAQ server according to a third embodiment [Figure 21] Diagram showing an example of information highlighting the correspondence of email destinations and the parts that need to be confirmed in the email body

Modes for Carrying Out the Invention

[0011] [[ID=B49]] The embodiments will be described in detail below, with reference to the drawings as appropriate. However, unnecessary details may be omitted. For example, detailed explanations of already well-known matters or redundant explanations of substantially identical configurations may be omitted. This is to avoid the following explanation becoming unnecessarily verbose and to facilitate understanding by those skilled in the art. The accompanying drawings and the following explanation are provided to enable those skilled in the art to fully understand this disclosure and are not intended to limit the subject matter described in the claims.

[0012] (The process leading to the acquisition of the first embodiment) In the FAQ system using the pattern recognition model described in Patent Document 1, after morphological analysis of the text, features designed by a human are extracted as candidates for question sentences Q and answer sentences A, resulting in insufficient extraction accuracy. Furthermore, unnecessary text contained in emails or chat logs that form the basis for creating the FAQ may be mixed in with the candidate Q or A. In addition, manually extracting representative question sets for Q and A is time-consuming. The first embodiment improves upon the above points.

[0013] (FAQ system configuration) Figure 1 is a block diagram of an FAQ system 5 according to the first embodiment of this disclosure. The FAQ system 5 includes an FAQ server 100, an FAQ knowledge base 150, a user terminal 200, and an administrator terminal 300. The FAQ server 100, the user terminal 200, and the administrator terminal 300 are connected to each other via, for example, wired or wireless communication. The FAQ system 5 is an example of a data management system.

[0014] The FAQ server 100 is a device that processes FAQs, and as described later, it performs tasks such as assisting with registration to the FAQ knowledge base 150 and narrowing down questions. The FAQ server 100 has a search function for searching question texts and answer texts. The FAQ server 100 can also operate as an FAQ chatbot CB. The FAQ chatbot CB receives questions from users, refers to the FAQ knowledge base 150, and provides answers to the questions.

[0015] The FAQ Knowledge Base 150 is a database for knowledge management related to FAQs and is constructed, for example, as a database server. The FAQ Knowledge Base 150 may have control functions, storage functions, and communication functions, similar to a general server. The FAQ Knowledge Base 150 may be included in the FAQ Server 100, or it may be configured as an integral part of the FAQ Server 100.

[0016] User terminal 200 is a terminal used by users. These users are those who use the FAQ service provided by FAQ system 5 (i.e., questioners). Administrator terminal 300 is a terminal used by administrators (operators). These administrators are those who manage the FAQ service provided by FAQ system 5. Administrators may include answerers who respond to questions from questioners. Alternatively, administrators may not include answerers, and answerer terminals used by answerers may be provided separately from the administrator terminal. Here, as an example, we assume that administrators include answerers. User terminal 200 and administrator terminal 300 can communicate directly with each other.

[0017] The FAQ server 100 receives a question from the user terminal 200 and attempts to answer it. If the FAQ server 100 can answer the question, it sends the answer to the user terminal 200. On the other hand, if the FAQ server 100 cannot answer the question, the user terminal 200 and the administrator terminal 300 communicate the question and the answer using email or chat functions. The administrator terminal 300 and the user terminal 200 may also communicate via the FAQ server 100.

[0018] Figure 2 is a block diagram showing an example configuration of the FAQ server 100. The FAQ server 100 comprises a control unit 110, a storage unit 130, and a communication unit 140.

[0019] The control unit 110 is configured, for example, by a processor and implements various functions by executing programs stored in the memory unit 130. The processor may include an MPU (Microprocessing Unit), a CPU (Central Processing Unit), a DSP (Digital Signal Processor), etc. The processor may also include various integrated circuits (for example, an LSI (Large Scale Integration) or an FPGA (Field Programmable Gate Array)). The control unit 110 coordinates the operation of each part of the FAQ server 100 and performs various processes. For example, the control unit 110 performs processing related to FAQs.

[0020] The storage unit 130 includes primary storage devices (e.g., RAM (Random Access Memory) or ROM (Read Only Memory)). The storage unit 130 may also include secondary storage devices (e.g., HDD (Hard Disk Drive) or SSD (Solid State Drive)) or tertiary storage devices (e.g., optical discs or SD cards). The storage unit 130 may also include other storage devices. The storage unit 130 may be removable from the FAQ server 100. The storage unit 130 stores various data, information, or programs.

[0021] The memory unit 130 includes a non-chat log database (DB) 131, a chat log database (DB) 132, and a QA database (DB) 133.

[0022] The non-chat log database 131 holds inquiry logs other than chat logs. Non-chat logs, which are logs of inquiries other than chat logs, include, for example, logs (historical information) related to inquiries made by email or telephone. Non-chat logs may also include data showing the content of emails or telephone calls.

[0023] The chat log database 132 stores logs (historical information) of inquiries made via chat. The chat log may also include the chat messages themselves.

[0024] Thus, the non-chat log database 131 and the chat log database 132 store various data and emails that were previously communicated between the user and the respondent.

[0025] The QA database 133 holds QA data obtained from chat logs or non-chat logs. The QA data includes a question and an answer to that question. The QA database 133 may also hold data associated with the QA data, such as the mapping results from clustering described later, feature vectors of the question obtained from the text data (question vector), feature vectors of the answer (answer vector), and information on links that allow access to the QA data. The mapping results from clustering, the question vector, and the answer vector serve as criteria for determining the similarity of multiple QA data sets.

[0026] The communication unit 140 transmits various data or information. The communication method used by the communication unit 140 may be wired communication or wireless communication, and may include communication methods such as LAN (Local Area Network), WAN (Wide Area Network), power line communication, or cellular communication for mobile phones (e.g., 5G, LTE (Long Term Evolution)). The communication unit 140 can communicate with, for example, the FAQ knowledge base 150, the user terminal 200, and the administrator terminal 300, and may also communicate with other communication devices.

[0027] The control unit 110 has a functional configuration that includes a text extraction unit 111, a word segmentation unit 112, a vectorization unit 113, a classification unit 114, a vector similarity calculation unit 115, a clustering unit 116, a cluster center calculation unit 117, a named entity recognition unit 119, and an anonymization processing unit 121.

[0028] The text extraction unit 111 extracts text data from chat logs or non-chat logs stored in the storage unit 130. The word segmentation unit 112, for example, uses a natural language processing model to segment the extracted text data into word units through morphological analysis. The control unit 110 may segment the data not only into word units but also into character units, subword units, etc.

[0029] The vectorization unit 113 uses a natural language processing model (for example, BERT (Bidirectional Encoder Representations from Transformers) or another model) to vectorize the extracted text data and generate feature vectors (text vectors).

[0030] The classification unit 114 uses a natural language processing model to classify the text data based on the generated text vectors. The classification unit 114 may, for example, pre-annotate the text data for classification, classify the extracted text data, and extract question sentences and answer sentences from the text data. The classification unit 114 performs a QA mapping process on the extracted question sentences and answer sentences to generate QA data that includes the question sentences and their answers. In the QA mapping process, if there is only one question sentence in the email body stored in the storage unit 160, the classification unit 114 may extract the answer sentence from the reply email from the administrator and map it to the question sentence. If there are multiple question sentences in the email body stored in the storage unit 160, the classification unit 114 may map each question sentence to the corresponding reply email for each question email.

[0031] Furthermore, among the text vectors, those that represent the question are also called question vectors, and those that represent the answer are also called answer vectors. In addition, the vectors of the candidate questions (candidate question sentences) described later are also called candidate question sentence vectors.

[0032] The vector similarity calculation unit 115 uses a natural language processing model to calculate the similarity (correlation) of each sentence vector generated by the vectorization unit 113.

[0033] The clustering unit 116 clusters the question sentences based on the similarity of each derived sentence vector (each question sentence vector) using a predetermined clustering method (e.g., k-Means method or other clustering method), and divides them into multiple clusters (question clusters). Through clustering, multiple question sentences are classified to belong to one of the multiple question clusters. Similarly, using a predetermined clustering method (e.g., k-Means method), the answer sentences are clustered based on the similarity of each derived sentence vector (each answer sentence vector), and divided into multiple clusters (answer clusters). The answer sentences may also be clustered in association with the question sentences. Through clustering, multiple answer sentences are classified to belong to one of the multiple answer clusters.

[0034] Clustering does not require training data. Therefore, the FAQ server 100 does not require prior training and can quickly classify, for example, question sentences and answer sentences.

[0035] The cluster-centered calculation unit 117 extracts cluster-centered data for each cluster based on a predetermined clustering method. In other words, the cluster-centered calculation unit 117 derives the question sentence located at the cluster center as the representative question sentence QR. Note that the cluster-centered calculation unit 117 may extract cluster-centered data using methods other than the k-Means method.

[0036] The named entity recognition unit 119 analyzes words contained in the text data using, for example, a natural language processing model, and extracts proper nouns (named entities) such as names of people, company names, addresses, or telephone numbers.

[0037] The anonymization processing unit 121 performs anonymization on the extracted named entities. Anonymization is data processing that removes or modifies personally identifiable information. Anonymization may include, for example, changing named entities to generic or common names, changing named entities to specific expressions (e.g., XXX), deleting named entities, or other anonymizing processes.

[0038] Next, we will describe the configuration of the user terminal 200 and the administrator terminal 300.

[0039] The user terminal 200 and the administrator terminal 300 include, for example, a PC (Personal Computer), a smartphone, or a tablet terminal. The user terminal 200 and the administrator terminal 300 include, for example, a control unit, a storage unit, a communication unit, an operation unit, and a display unit.

[0040] Next, we will explain the overall overview of FAQ System 5. Figure 3 is a diagram illustrating the overview of the FAQ chat screen.

[0041] The FAQ knowledge base 150 stores QA data. When a user terminal 200 sends a question Q to the FAQ chatbot CB, and does not receive an answer A for this question Q, the user terminal 200 sends the question Q to the administrator terminal 300, inquiring as a Q&A. In this case, the control unit 110 of the FAQ server 100 analyzes the question Q from the user terminal 200 and the answer A from the administrator terminal 300, and generates new QA data. The generated QA data can be registered in the FAQ knowledge base 150. As a result, the amount of QA data registered in the FAQ knowledge base 150 increases, thereby increasing the number of questions that the FAQ chatbot CB can answer. When the administrator wishes to register QA data to the FAQ knowledge base 150 using the administrator terminal 300, the FAQ server 100 performs registration support processing to the FAQ knowledge base 150 (first embodiment).

[0042] Furthermore, the FAQ server 100 receives an arbitrary question sentence Q (Q0) and searches for FAQs related to this question from the QA data stored in the FAQ knowledge base 150. Multiple QA data are obtained as candidate question sentences QC for the search results (candidate results). The candidate results vary depending on the question. Therefore, the FAQ server 100 selects a keyword KW by creating a keyword KW for narrowing down the candidate results each time according to the candidate results. In this way, the FAQ server 100 performs question narrowing processing (second embodiment).

[0043] Figure 4 is a sequence diagram showing an overview of the operation of FAQ System 5.

[0044] First, in the FAQ server 100, in order to generate answer text A for user questions Q using the FAQ chatbot CB, the control unit 110 registers QA data, including FAQ questions Q and answer text A, in the FAQ knowledge base 150 in advance (before the FAQ chatbot CB is put into operation) (S11). Note that FAQ QA data may also be registered in the FAQ knowledge base 150 during the operation of the FAQ chatbot CB. Note that there may be cases where only one of the FAQ questions Q or answer text A is registered in the FAQ knowledge base 150.

[0045] The user terminal 200, for example, uses the FAQ chat screen G1 to input an arbitrary question Q0 into the FAQ chat (S21). The user terminal 200 then sends this question Q0 to the FAQ server 100 (FAQ chatbot CB).

[0046] In the FAQ server 100, the communication unit 140 receives the question sentence Q0, and the control unit 110 extracts candidate question sentences QC related to the question sentence Q0 from the FAQ knowledge base 150 (S12). The control unit 110 generates a group of candidate questions (candidate question cluster) containing similar candidate questions from the extracted candidate question sentences QC (S13). The control unit 110 sets a word indicating the classification of the candidate question group (for example, a feature word KW indicating the features in the classification) (S14). The control unit 110 instructs the user terminal 200 to display the set feature word KW via the communication unit 140 (S15).

[0047] The user terminal 200 receives one or more feature words KW and instructions to display the feature words KW from the FAQ server 100, and displays the feature words KW on the display unit according to these instructions. The user terminal 200 accepts selection operations from the user, for example via the operation unit, and selects the feature word KW that is closest to the content the user wants to inquire about from among the received feature words KW (S22). The user terminal 200 sends the selected feature words KW to the FAQ server 100.

[0048] In the FAQ server 100, the communication unit 140 receives the selected characteristic word KW from the user terminal 200. The control unit 110 instructs the user terminal 200 to display the candidate question QC included in the group of candidate questions corresponding to the selected characteristic word KW (S16). The user terminal 200 receives the candidate question QC included in the group of candidate questions corresponding to the selected characteristic word KW and the instruction to display it from the FAQ server 100, and displays the received question.

[0049] Alternatively, if the FAQ server 100 has many questions in the group of candidate questions corresponding to the selected feature word KW, it recreates a group of candidate questions containing similar questions from this group, similar to step S13 (S16). Compared to the creation of the candidate question group in step S13, the recreation of the candidate question group in step S16 narrows the scope of related questions and refines the questions. Note that when recreating the candidate question group, the processes in steps S14 to S16 that follow step S13 are repeated.

[0050] Step S11 relates to the process of recommending registration to the FAQ knowledge base 150 and will be described in detail in the second embodiment. Steps S12 to S16 and steps S21 and S22 relate to the process of narrowing down questions and will be described in detail in the second embodiment.

[0051] Next, we will explain in detail the process for assisting with registration to the FAQ Knowledge Base 150.

[0052] Figure 5 shows an example of inquiry email M1. Figure 6 shows an example of the classification of inquiry email M1. Inquiry email M1 is an inquiry email containing question text Q. Figure 7 shows an example of the clustering of the derived question text Q and the registration recommendation information for QA data corresponding to the central data of the cluster.

[0053] An inquiry email M1 is, for example, obtained by the FAQ server 100 from a user terminal 200 of any user and stored in the non-chat log database 131 of the storage unit 130. The text extraction unit 111 extracts text data from this inquiry email M1. The text extraction unit 111 may perform preprocessing, such as deleting quoted text and signatures included in the inquiry email M1, or normalizing the text data included in the inquiry email M1. The vectorization unit 113 vectorizes the extracted text data and generates text vectors. The classification unit 114 classifies the text data based on the generated text vectors. The inquiry email M1 is classified, for example, into a greeting text including a greeting from the user, a question text Q including a question, a background text including the background of the question, a thank you message added to the end of the email, etc., as shown in Figure 6. In other words, the question text Q can be extracted from the inquiry email M1.

[0054] Similarly, the response email to the above inquiry email M1 is, for example, obtained by the FAQ server 100 from any administrator terminal 300 and stored in the non-chat log database 131 of the storage unit 130. The text extraction unit 111 extracts text data from this response email. The text extraction unit 111 may perform preprocessing, such as deleting quoted text and signatures included in the response email, or normalizing the text included in the response email. The vectorization unit 113 vectorizes the extracted text into feature vectors and generates text vectors. The classification unit 114 uses a natural language processing model to classify the text data based on the generated text vectors. Although not shown in the diagram, the response email is classified into a greeting including a greeting from the respondent, a response A including the answer to the question Q, a reason statement including the reason for the answer, a closing thank you message, etc. Note that the reason statement may be included in the response A. In other words, the response A can be extracted from the response email.

[0055] The control unit 110 generates QA data that includes a question sentence Q derived from an inquiry email M1 and an answer sentence A derived from an answer email to this inquiry email M1 as a pair, and stores it in the QA database 133. In this way, the control unit 110 may generate a large amount of QA data from a large amount of inquiry emails M1 and a large amount of answer emails stored in the non-chat log database 131 and store it in the QA database 133. Furthermore, although this example illustrates the generation of QA data based on emails that are not chat logs, the control unit 110 may similarly generate QA data based on chat logs of inquiries and chat logs of answers stored in the chat log database 132 and store it in the QA database 133. As a result, a large amount of historically proven QA data will be stored in the QA database 133.

[0056] It should be noted that the QA data stored in the QA database 133 is in a rough state. Specifically, the QA database 133 may contain a large amount of QA data that is similar in content.

[0057] The clustering unit 116 clusters the question sentences Q based on the question sentence vectors obtained by the vectorization unit 113. Figure 7 shows an example of the mapping information MP for question clusters C1, C2, and C3. Although Figure 7 shows the results of two-dimensional clustering, clustering of two dimensions or more (e.g., three dimensions) is also possible. The cluster-center calculation unit 117 extracts cluster-center data for each question cluster; that is, the cluster-center calculation unit 117 derives the cluster-center question sentence Q as a representative question sentence QR, which is a representative question.

[0058] Similarly, the clustering unit 116 clusters the answer sentences A for each question sentence Q based on the answer sentence vectors obtained by the vectorization unit 113. The cluster center calculation unit 117 derives the answer sentence A at the center of each answer cluster as the representative answer sentence AR, which is a representative answer. The representative question sentence QR and the representative answer sentence AR together constitute the representative QA data. If there are multiple answer sentences A for each question sentence Q, clustering of the answer sentences A is performed. However, if there is only one answer sentence A for each question sentence Q, clustering of the answer sentence A is unnecessary, and the single answer sentence A becomes the representative answer sentence AR. Similarly, the clustering unit 116 may also perform clustering on the question sentence Q, and the cluster center calculation unit 117 may calculate the representative question sentence QR.

[0059] The communication unit 140 instructs the administrator terminal 300 to display registration recommendation information I1, which recommends registering representative QA data in the FAQ knowledge base 150. The administrator terminal 300 receives the instruction to display registration recommendation information I1 and displays it accordingly. In Figure 7, the representative question text QR and its representative answer text AR for question cluster C3 are displayed.

[0060] For example, if an administrator who has confirmed the representative QA data via the control panel specifies that the representative QA data should be registered, the administrator terminal 300 registers (stores) the representative QA data in the FAQ knowledge base 150. This adds QA data that was not previously stored in the FAQ knowledge base 150. Therefore, the FAQ chatbot CB can answer questions from the user terminal 200 based on this QA data.

[0061] In this way, the FAQ server 100 extracts the central data for each cluster from among the similar question sentences Q and similar answer sentences A held in the QA database 133 by the control unit 110. The central data for each cluster is assumed to be the representative question sentence Q and answer sentence A in each cluster that contains similar question sentence Q and answer sentence A. In other words, the FAQ server 100 can extract the representative question sentence QR and representative answer sentence AR that are at the center of the cluster. Therefore, the FAQ knowledge base 150 will register the representative question sentence QR from among the similar question sentence Q and the representative answer sentence AR from among the similar answer sentence A. In other words, the FAQ server 100 can suppress the registration of unnecessary question sentences Q and answer sentence A in the FAQ knowledge base 150.

[0062] Figure 8 is a flowchart illustrating an example of the operation of the FAQ server 100 in supporting the registration of QA to the FAQ knowledge base 150.

[0063] First, the control unit 110 inputs emails from the non-chat log database 131 (S31). The control unit 110 inputs chat logs from the chat log database 132 (S31). The control unit 110 performs QA text extraction processing (S32). In the QA text extraction processing, the question text Q and the answer text A are extracted from the input emails, etc.

[0064] Figure 9 is a flowchart showing an example of the QA text extraction process in step S32. In the QA text extraction process, the vectorization unit 113 vectorizes the text data in each email or chat log and generates a text vector (S41). The classification unit 114 classifies the text data based on the generated text vectors (S42). By classifying the text data, for example, a question sentence Q, an answer sentence A, etc., are obtained. The control unit 110 registers the obtained question sentence Q, the answer sentence A corresponding to the question sentence Q (corresponding answer sentence), the feature vector of the question sentence Q (question sentence vector), and the feature vector of the answer sentence A (answer sentence vector) in the QA database 133 (S43).

[0065] Returning to Figure 8, the clustering unit 116 clusters the question vectors registered in the QA database 133 (S33). The cluster center calculation unit 117 extracts a central question Q (representative question QR) from each question Q of each question cluster obtained by clustering (S34). The cluster center calculation unit 117 extracts an answer A (representative answer AR) corresponding to the extracted representative question QR from, for example, the QA database 133 (S35). The communication unit 140 instructs the administrator terminal 300 used by the administrator to display the representative question QR and representative answer AR, and to display registration recommendation information recommending registration to the FAQ knowledge base (S36).

[0066] Note that the representative answer sentence AR may be extracted by a process other than the process in step S35 shown in Figure 8. Figure 10 is a flowchart of another example of extracting the representative answer sentence AR corresponding to the representative question sentence QR. For example, a question sentence cluster may contain not only one or more question sentences Q (question sentence vectors), but also one or more answer sentences A (answer sentence vectors). The clustering unit 116 may cluster the feature vectors of the answer sentences A within each question cluster and extract the answer sentence A that is at the center of the answer cluster obtained by clustering as the representative answer sentence AR (S35A).

[0067] Furthermore, if the inquiry email contains only one question (Q), the control unit 110 may extract an answer (A) from the response email to this inquiry email, associate this question (Q) with the answer (A), and generate QA data. Alternatively, if the inquiry email contains multiple question (Q) statements, the control unit 110 may refer to the multiple answer (A) statements stored in the non-chat log database 131, and generate multiple QA data sets by having a human (i.e., the administrator) associate each question (Q) with each answer (A) statement via the operation panel of the administrator terminal 300.

[0068] In other words, the following processes are performed when assisting with QA registration in the FAQ Knowledge Base 150. The control unit 110 extracts only the text of the question sentence Q and answer sentence A from email or chat logs, etc., creates rough FAQ information, and stores it in the QA database 133. The control unit 110 clusters the question sentence Q of the rough FAQ information to group similar question sentence Qs together (clustering). The control unit 110 instructs the administrator to present the representative question sentence QR and the representative answer sentence AR corresponding to this representative question sentence QR that are at the center of the group (cluster). Alternatively, the control unit 110 may present the representative answer sentence AR to the administrator by clustering the answer sentence A of the same question sentence group.

[0069] Before sending the registration recommendation information to the administrator terminal 300, the vector similarity calculation unit 115 of the FAQ server 100 may determine whether the representative question text QR does not match or is not similar (has low similarity) to a past question text (question text Q from the QA data held in the QA database 133). Not being similar means, for example, that the similarity between the two is below a predetermined threshold. If the representative question text QR does not match or is not similar to a past question text (question text Q from the QA data held in the QA database 133), the registration recommendation information is sent; if they match or are similar, the registration recommendation information does not need to be sent. This allows the FAQ server 100 to prevent the administrator from registering representative QA data similar to QA data containing past question text Q in the FAQ knowledge base 150.

[0070] Next, variations of the FAQ system 5 of this embodiment will be described.

[0071] In the FAQ server 100, the control unit 110 may perform the following actions. For example, the control unit 110 may use a natural language processing model to summarize one or more question sentences Q. This reduces redundant information in the question sentences Q, resulting in question sentences Q that can be read quickly.

[0072] Furthermore, the control unit 110 may use a natural language processing model to reconstruct one or more question sentences Q into question sentences Q that have a correct Japanese structure. This will result in question sentences Q that are easy to understand.

[0073] Furthermore, the control unit 110 may have the named entity recognition unit 119 detect predetermined expressions contained in the question sentence Q (for example, typographical errors, compliance violations, offensive expressions, or monetary information) based on the feature vectors of the words in the question sentence Q, for example, using a natural language processing model. The control unit 110 may also correct the detected expressions.

[0074] Furthermore, the control unit 110 may have the named entity recognition unit 119 detect personal information (e.g., name, address, or telephone number) contained in the question sentence Q based on the word feature vectors of the question sentence Q, for example, using a natural language processing model. The anonymization processing unit 121 may perform anonymization processing (e.g., deletion) on the detected personal information. This results in a question sentence Q that does not contain personal information. In this case, the FAQ server 100 may have the communication unit 140 send anonymization information to the administrator terminal 300 indicating that anonymization processing of personal information has been performed. When the administrator terminal 300 receives the anonymization information, it may display a selection screen on the display unit for selecting between a question sentence Q containing personal information and a question sentence Q that does not contain personal information. The administrator terminal 300 The user can perform this selection operation via the control unit from the selection screen, and depending on the selection operation, it may be recommended to register either representative QA data containing question text Q with personal information, or representative QA data containing question text Q without personal information, in the FAQ knowledge base 150. In addition, the communication unit 140 may send the registration recommendation information to the administrator terminal 300 after anonymizing the personal information.

[0075] When a user attempts to ask a question to the FAQ chatbot CB and obtain an answer, as a pre-processing step, representative QA data is registered in the FAQ knowledge base 150 using the registration support process for the FAQ knowledge base 150 described above. The FAQ chatbot CB refers to the FAQ knowledge base 150 and, if it receives a question Q from the user terminal 200 that it can answer, it sends an answer A to question Q to the user terminal 200. The FAQ chatbot CB refers to the FAQ knowledge base 150 and, if it receives a question Q from the user terminal 200 that it cannot answer question Q, it sends a message to the user terminal 200 indicating that it cannot answer question Q.

[0076] Then, if the user confirms that they cannot obtain an answer to question Q via the display unit of the user terminal 200, the user terminal 200 sends question Q to the administrator terminal 300 via email or chat, etc. (hereinafter also referred to as "direct question"). In addition, if the user does not wish to receive an answer from the FAQ chatbot CB, the user terminal 200 may also directly ask a question to the administrator terminal 300. Similarly, when the administrator terminal 300 sends an answer A to question Q received via email or chat, etc., to the administrator terminal 300, this is also referred to as "direct answer". When a direct question or direct answer is made, the following processing may be performed.

[0077] When a user terminal 200 directly asks a question to an administrator terminal 300, the administrator terminal 300, upon receiving the question Q from the user terminal 200, queries the FAQ server 100, including the question Q. The control unit 110 of the FAQ server 100 retrieves QA data from the QA database 133 or FAQ knowledge base 150 that contains a question Q similar to the question Q included in the inquiry from the administrator terminal 300, and the communication unit 140 transmits this data to the administrator terminal 300. Whether or not two question Qs are similar may be derived by the vector similarity calculation unit 115 based on the question vector. The administrator terminal 300 receives the QA data from the FAQ server 100 and sends the answer A included in the QA data to the user terminal 200 in response to the question Q from the user terminal 200. This allows the responder using the administrator terminal 300 to answer by referring to past QA similar to the questioner's question, reducing the effort required to create the answer A.

[0078] Furthermore, when a user terminal 200 directly asks a question to the administrator terminal 300, the administrator terminal 300, upon receiving the question Q from the user terminal 200, queries the FAQ server 100, including the question Q. The FAQ server 100's control unit 110 refers to the QA database 133 and obtains information about a link that allows access to QA data containing the question Q included in the query, or QA data containing a similar question Q (similar question) (similar QA data), and the communication unit 140 transmits this information to the administrator terminal 300. The administrator terminal 300 receives the above link information from the FAQ server 100 and transmits it to the user terminal 200. As a result, a respondent using the administrator terminal 300 can access the link to check the answer A included in the QA data or similar QA data, and can use this answer A as a reference to answer, thereby reducing the effort required to create the answer A.

[0079] Furthermore, when a user terminal 200 directly asks a question to the administrator terminal 300, the administrator terminal 300, upon receiving the question Q from the user terminal 200, queries the FAQ server 100, including the question Q. The FAQ server 100's control unit 110 identifies the cluster to which the QA data containing the question Q included in the query belongs, refers to the QA database 133, and extracts other QA data belonging to this cluster. The control unit 110 also generates mapping information that maps each QA data from the extracted cluster onto a two-dimensional plane or three-dimensional space, and generates tooltip information. The communication unit 140 transmits the generated mapping information and tooltip information to the user terminal 200. The administrator terminal 300 receives the mapping information and tooltip information from the FAQ server 100 and displays it on the display unit. As a result, respondents using the administrator terminal 300 can intuitively grasp the relationships between each QA data in the question cluster (group of questions) to which the received question belongs by checking the mapping information. Therefore, respondents using administrator terminal 300 can refer to past Q&A similar to the question from the questioner to provide an answer, thereby reducing the effort required to create answer A.

[0080] Furthermore, when a user terminal 200 directly asks a question to the administrator terminal 300, the administrator terminal 300, upon receiving the question Q from the user terminal 200, queries the FAQ server 100, including the question Q. The FAQ server 100's control unit 110 calculates the similarity between each question Q in each QA data for each question cluster held in the QA database 133 or FAQ knowledge base 150 and the question Q included in the query from the administrator terminal 300, for example, using a natural language processing model. In this case, the control unit 110 may calculate the similarity between each question Q in each QA data for each QA data held in the QA database 133 or FAQ knowledge base 150 and the question Q included in the query based on the feature vector of the question Q included in the query. If this similarity is below a predetermined threshold, the communication unit 140 may send early registration recommendation information to the administrator terminal 300, recommending that the QA data including the question Q included in the query be registered in the QA database 133 or FAQ knowledge base 150 as soon as possible. The administrator terminal 300 may receive early registration recommendation information from the FAQ server 100 and display it on its display unit.

[0081] When a service using FAQ System 5 is launched, new QA data will increase rapidly based on interactions between users and respondents. In response to this, FAQ System 5 can notify respondents if a newly submitted question Q differs significantly in similarity to a question Q in the QA database 133 or FAQ knowledge base 150, and encourage them to register the QA data containing the question Q from the questioner in the QA database 133 or FAQ knowledge base 150 as soon as possible. Therefore, FAQ System 5 can prioritize the registration of QA data with high deficiency, and reduce the workload associated with user questions and respondent answers.

[0082] After a direct question and answer exchange is completed between the user terminal 200 and the administrator terminal 300, the administrator terminal 300 sends registration application information to the FAQ server 100, which includes the completed question text Q and answer text A, and requests the registration of this QA data. The FAQ server 100 receives the registration application information including the QA data via its communication unit 140, and the control unit 110 stores the QA data in the QA database 133 or the FAQ knowledge base 150. The control unit 110 also clusters the question text Q into question clusters based on the feature vector of the question text Q, and tags the type of question text Q corresponding to this question cluster. This tagging information may be stored in the QA database 133 or the FAQ knowledge base 150.

[0083] Furthermore, the FAQ server 100's control unit 110 divides the question sentence Q into words and extracts proper nouns (proper expressions) such as names of people or companies. The communication unit 140 then transmits registration completion information, indicating that the QA data related to the registration application information has been registered in the QA database 133 or FAQ knowledge base 150, tagging completion information, indicating that it has been tagged, and proper expression extraction information, indicating that proper expressions have been extracted, to the administrator terminal 300. The administrator terminal 300 receives the registration completion information, tagging completion information, and proper expression extraction information and displays them on the display unit. This increases the motivation of users and respondents to apply for registration of QA data. In addition, QA data including the question sentence Q and answer sentence A that have been successfully answered can be registered in the QA database 133 and FAQ knowledge base 150 as past performance information and made available for use.

[0084] Furthermore, after a direct question and answer exchange is completed between the user terminal 200 and the administrator terminal 300, the administrator terminal 300 sends QA data, including the completed question Q and answer A, to the FAQ server 100. The FAQ server 100's communication unit 140 receives this completed QA data, and the control unit 110 refers to the QA database 133 or the FAQ knowledge base 150 to search for similar QA data to the completed QA data. Based on the presence or absence of similar QA data, the control unit 110 determines whether or not to retain the completed QA data in the QA database 133 or the FAQ knowledge base 150. In other words, similar QA data serves as a basis for deciding whether to register new entries in the FAQ knowledge base 150. If no similar QA data exists, the FAQ server 10 retains the completed QA data in the QA database 133 or the FAQ knowledge base 150; if similar QA data exists, it does not retain the completed QA data in the QA database 133 or the FAQ knowledge base 150. As a result, the FAQ server 100 can register completed QA data in the QA database 133 or the FAQ knowledge base 150, reducing the effort required for user questions and respondent answers.

[0085] Furthermore, after a direct question and answer exchange is completed between the user terminal 200 and the administrator terminal 300, the administrator terminal 300 sends QA data, including the completed question Q and answer A, to the FAQ server 100. The FAQ server 100's communication unit 140 receives this completed QA data, and the control unit 110 extracts the named entities from the question Q and answer A contained in the QA data, anonymizes the extracted named entities, and may store the anonymized QA data in the QA database 133 or the FAQ knowledge base 150. The communication unit 140 sends the QA data, including the anonymized question Q and answer A, to the administrator terminal 300. The named entities include, for example, personal information or other information that should be kept confidential. The administrator terminal 300 receives the anonymized QA data and displays it on the display unit. As a result, the administrator terminal 300 can receive registration instructions from the administrator via the control panel, allowing it to store anonymized QA data in the QA database 133 or FAQ knowledge base 150. Therefore, personal information and other information that needs to be kept confidential can be registered in the QA database 133 or FAQ knowledge base 150 with reduced risk of information leakage, thereby reducing the workload involved in questions from users and answers from respondents.

[0086] Furthermore, after a direct question and answer exchange is completed between the user terminal 200 and the administrator terminal 300, the administrator terminal 300 sends an inquiry email containing the completed question Q and an answer email containing the answer A to the FAQ server 100. The FAQ server 100's communication unit 140 receives the completed inquiry email and answer email, and the control unit 110 may obtain signature information from these emails and perform anonymization processing on the affiliation information and name information included in the signature information. The FAQ server 100 may then store the anonymized emails in the non-chat log database 131. This allows the FAQ server 100 to extract and store past QA data necessary for the FAQ chatbot CB and the respondent from the inquiry emails and answer emails in a state where the person who sent the past inquiry email or answer email cannot be identified. In addition, personal information can be reliably made undecipherable.

[0087] Furthermore, if QA data is registered in the FAQ knowledge base 150, the FAQ server 100 may have its communication unit 140 send knowledge base registration completion information to the administrator terminal 300, indicating that the QA data has been registered in the FAQ knowledge base 150. The administrator terminal 300 may receive the knowledge base registration completion information from the FAQ server 100 and display it on its display unit. The knowledge base registration completion information may include a message such as, "This interaction has been registered as an FAQ. You can check it at the URL XXX." This allows the administrator to confirm that the QA data to be registered has been successfully registered in the FAQ knowledge base 150, and to confirm that the FAQ chatbot CB, which provides answers by referring to the FAQ knowledge base 150, is growing.

[0088] Furthermore, while the above primarily describes encouraging users to register questions in the FAQ Knowledge Base 150 before automatically providing answers to user questions Q using the FAQ chatbot CB (i.e., before operation), it is not limited to this. For example, after a direct question and answer exchange is completed between the user terminal 200 and the administrator terminal 300, the administrator terminal 300 sends an inquiry email containing the completed question Q and an answer email containing the answer A to the FAQ server 100. The FAQ server 100's communication unit 140 receives the completed inquiry email and answer email, and the control unit 110 stores them in the non-chat log database 131. The control unit 110 clusters the question Q extracted from each email stored in the non-chat log database 131 or each chat log stored in the chat log database 132, and extracts the question Q at the center of each cluster as the representative question QR. The communication unit 140 sends registration recommendation information to the user terminal 200, recommending that representative QA data, including a representative question QR code and a representative answer AR for the representative question QR code, be registered in the FAQ knowledge base 150. The administrator terminal 300 receives the registration recommendation information and displays it on its display unit. This allows the FAQ system 5 to recommend registration in the FAQ knowledge base 150 even after the direct question and direct answer process is completed. Furthermore, the FAQ system 5 may repeatedly perform the registration support process to the FAQ knowledge base 150 at any time before, during, and after the operation of the FAQ chatbot CB. This makes it possible to sequentially register QA data (representative QA data) in the FAQ knowledge base 150.

[0089] Furthermore, the user terminal 200 transmits all or part of the question Q to the FAQ server 100 (FAQ chatbot CB). When the communication unit 140 of the FAQ server 100 receives all or part of this question Q, the control unit 110 identifies QA data similar to all or part of this question Q and containing the question Q stored in the FAQ knowledge base 150, and the communication unit 140 transmits information (e.g., a URL) of a link to access this QA data to the user terminal 200. The user terminal 200 receives the link information and displays it on the display unit. As a result, the user can easily check question Qs similar to all or part of the question Q they entered by accessing the link, thereby reducing the effort required to ask a question.

[0090] As described above, the FAQ system 5 of this embodiment displays registration recommendation information on the administrator terminal 300, allowing the administrator to easily determine whether or not to register QA data in the FAQ knowledge base 150. Furthermore, since the FAQ system 5 extracts features using question vectors, which are vectorized versions of the question text, and performs clustering, there is no need for features to be used as input data for the pattern recognition model, thus reducing the processing load on the administrator related to clustering. In addition, it is possible to suppress the extraction of features designed by humans as registration candidates for question text Q and answer text A, thereby improving the accuracy of registration in the FAQ knowledge base 150. Moreover, it is possible to suppress the inclusion of some unnecessary text data contained in emails and chat logs that are the source of FAQ creation in the question text Q and answer text A to be registered. Furthermore, manual extraction of representative QA data from multiple QA data is unnecessary. Therefore, the convenience of the FAQ system 5 can be improved.

[0091] (The process leading to the acquisition of the second embodiment) An FAQ system returns search results when any word or phrase is entered. However, when the technology of Reference 2 is applied to an FAQ system, it is difficult for the user to obtain the desired search results in a short time if there are a large number of search result candidates. Furthermore, while the conventional full-text search system FESS allows for the narrowing of search results by assigning words (labels) in advance to narrow down candidates, these words must be set manually. Therefore, when FESS is applied to an FAQ system to narrow down questions, since words are set manually by each person, there is a high possibility that candidates containing similar words will be selected as the result of the narrowing, making it difficult to narrow down questions appropriately. The second embodiment improves upon the above points.

[0092] (Second embodiment) In the second embodiment, the FAQ system 5 performs question filtering when it automatically answers a question using the FAQ chatbot CB.

[0093] Figure 11A is a block diagram showing an example configuration of an FAQ server according to a second embodiment of this disclosure. In Figure 11A, components similar to those in the FAQ server 100 shown in Figure 2 are denoted by the same reference numerals, and their descriptions are omitted or simplified.

[0094] The FAQ server 100A of this embodiment comprises a control unit 110A, a storage unit 130A, and a communication unit 140. The storage unit 130A includes a non-chat log database (DB) 131, a chat log database (DB) 132, a QA database (DB) 133, and an inspection result database (DB) 134. The control unit 110A comprises a text extraction unit 111, a word segmentation unit 112, a vectorization unit 113, a vector similarity calculation unit 115, a clustering unit 116, and a feature word extraction unit 118.

[0095] The search results database 134 holds search results obtained through a predetermined search process based on the user's question. The search results database 134 may store past search results as history, or it may be rewritten with each search. The search results here are question sentences that serve as additional candidate questions, obtained as a result of searching for the relevant FAQ using the question sentence. Multiple possible question sentences may be derived from the search results.

[0096] The text extraction unit 111 extracts text data from the QA database 133 or the search results database 134. In this case, the text extraction unit 111 may extract question sentences from the QA data held in the QA database 133 as candidate question sentences. Alternatively, the text extraction unit 111 may extract candidate question sentences held in the search results database 134.

[0097] The vectorization unit 113 uses a natural language processing model (for example, Word2Vec, TF-IDF (Term Frequency - Inverse Document Frequency), or other clustering method) to vectorize the candidate question sentences and obtain feature vectors (question complement vectors) for the candidate question sentences.

[0098] The vector similarity calculation unit 115 uses a natural language processing model to calculate the similarity (correlation) of each feature vector (each question complement vector) generated by the vectorization unit 113.

[0099] The clustering unit 116 uses a predetermined clustering method (e.g., k-means method or other clustering method) to classify the derived candidate question sentences so that they belong to one of several clusters (candidate question clusters) based on the similarity of each candidate question sentence vector.

[0100] The word segmentation unit 112 divides each candidate question sentence belonging to each candidate question cluster obtained by clustering into individual words.

[0101] The feature word extraction unit 118 uses a natural language processing model (e.g., TF-IDF or other feature word extraction method) to extract, for example, frequently occurring words as feature words KW from among the words contained in each candidate question sentence belonging to each candidate question cluster. For example, the feature word extraction unit 118 extracts one or more feature words from the question sentences of each candidate question cluster. Furthermore, if there are multiple candidate question clusters, the feature word extraction unit 118 may extract different feature words KW for each candidate question cluster that represent the characteristics of each cluster.

[0102] The communication unit 140 transmits each characteristic keyword KW of each extracted cluster to the user terminal 200. The user terminal 200 receives each characteristic keyword KW of each cluster from the FAQ server 100A and displays each characteristic keyword KW as a candidate for clarifying the question on the display unit. This allows the user to check the candidate for clarifying the question and narrow down the search results so that they can obtain the desired result (question text) as the search result.

[0103] In the second embodiment, for example, QA data is registered in the FAQ knowledge base 150 by the registration support process described in the first embodiment, but this is not limited to this. For example, QA data may be registered in the FAQ knowledge base 150 by a method other than that of the first embodiment.

[0104] Figure 11B shows an image of the FAQ chat screen G2 in this embodiment where question filtering is performed. As shown in Figure 11B, when a user sends an arbitrary question Q0 using the user terminal 200, the FAQ server 100A performs a search using the question Q0. As a result of the search, Figure 11B shows that approximately 100 registered FAQ question candidates are obtained. At this point, the FAQ server 100A sends and displays characteristic words KW to the user terminal 200 to narrow down the questions, prompting the user to specify the characteristic words KW.

[0105] In this embodiment, the FAQ server 100A's control unit 110A uses clustering to cluster candidate documents and divide them into multiple clusters (groups). The control unit 110A then extracts characteristic words KW from the documents contained in each cluster and uses these characteristic words KW as candidates to be used for follow-up questions. When the user terminal 200 specifies a characteristic word KW to be used as a follow-up question candidate via the operation unit, it transmits this specification operation information to the FAQ server 100A. The FAQ server 100A's communication unit 140 receives the specification operation information, and the control unit 110A identifies a candidate question sentence QC included in the candidate question cluster related to the specified characteristic word KW, and the communication unit 140 transmits the identified candidate question sentence QC to the user terminal 200. The user terminal 200 receives the candidate question sentence QC from the FAQ server 100A and displays it on the display unit. In this way, the user can obtain a related characteristic word KW from any question sentence Q0, and by specifying the characteristic word KW, can obtain a candidate question sentence QC related to the characteristic word KW.

[0106] Figure 12 shows an image of clustering. As shown in Figure 12, the FAQ server 100A's control unit 110A divides a large number of search results into a small number of candidate question clusters CL1 to CL3. In conventional clustering, users cannot easily grasp the characteristics of each candidate question cluster. On the other hand, as shown in Figure 12, the FAQ server 100A makes the characteristic words KW of each candidate question cluster into candidate search results, making them understandable to humans (users). The FAQ server 100A can clearly distinguish the characteristic differences of each word through dynamic word repetition, enabling highly accurate question narrowing. For example, when a characteristic word KW corresponding to candidate question cluster CL1 is selected, the FAQ server 100A can narrow down the candidates to each candidate question sentence QC belonging to candidate question cluster CL1 characterized by the characteristic word KW, which can be reduced to, for example, 5 choices.

[0107] Figure 13 shows an example of search results (candidates) including Q numbers and candidate question sentences (QC). In Figure 13, the Q number that identifies the candidate question sentence (QC) and the candidate question sentence (QC) are shown in association. In Figure 13, six candidate question sentences (QC) are obtained as search results (candidates).

[0108] Figure 14 shows an example of the results of vectorizing candidate question sentences (QC). In Figure 14, the Q number and the candidate question sentence vector, which is the vectorized version of the candidate question sentence (QC), are shown in relation to each other.

[0109] Figure 15 shows an example of clustering candidate question vectors and the corresponding candidate question QCs for each cluster. In Figure 15, the cluster number that identifies each cluster is associated with one or more candidate question QCs belonging to each cluster. In Figure 15, four clusters are generated; for example, cluster 0 contains two candidate question QCs.

[0110] Figure 16 shows an example of the vectorization results of the candidate question sentences QC for each cluster. In Figure 16, the cluster number is shown in association with the candidate question sentence vectors of one or more candidate question sentences QC belonging to each cluster.

[0111] Figure 17 shows an example of characteristic keywords extracted for each cluster. In Figure 17, the cluster number and the characteristic keywords for each cluster are shown in association. Multiple characteristic keywords may be derived for each cluster.

[0112] In this way, the control unit 110A can derive candidate follow-up questions to narrow down the search results from the search results obtained for any question sentence Q0. Furthermore, each characteristic word KW obtained as a follow-up question candidate can be transmitted to and displayed on the user terminal 200.

[0113] Figure 18 shows an example of the FAQ chat screen G3, which takes question filtering into consideration. In Figure 18, after an arbitrary question is asked, it is shown that 20 search results have been obtained, and the user is further prompted to filter using characteristic keywords (KW). The characteristic keywords here are the characteristic keywords for each cluster shown in Figure 17. In Figure 18, the characteristic keywords are displayed separately for each cluster. Furthermore, the user uses the user terminal 200 to specify the characteristic keywords for cluster number 1, namely contract period, expiration date, and automatic renewal, as the characteristic keywords for filtering. As a result, the question candidate sentences QC belonging to cluster number 0, which correspond to the specified characteristic keywords, are displayed.

[0114] Next, variations of this embodiment will be described.

[0115] The administrator terminal 300 may input stop words that are not to be extracted as feature words KW by the feature word extraction unit 118 of the FAQ server 100A via the operation unit, and send the stop words to the FAQ server 100A. The FAQ server 100A may receive the stop word information via the communication unit 140, and the control unit 110A may set the stop words. The set stop words may be stored in the storage unit 130.

[0116] The administrator terminal 300 may, via its operation unit, accept an operation to generate a similar word table in which multiple similar words are registered, and transmit this operation information to the FAQ server 100A. The FAQ server 100A's communication unit 140 receives this operation information, and the control unit 110A may generate a similar word table based on this operation information. The similar word table stores, for example, multiple similar words and a representative word that represents the multiple similar words in association. Similar words are, for example, words with variations in spelling. The configured similar word table may be stored in the storage unit 130. When the feature word extraction unit 118 obtains a feature word KW, if the similar word table contains words similar to this feature word KW, it extracts a representative feature word corresponding to this feature word KW.

[0117] This allows the FAQ server 100A to improve the quality of extracted feature words (KW), and thus improve the quality of suggested clarification questions. Consequently, the FAQ server 100A can improve the speed or accuracy of narrowing down search result candidates.

[0118] Furthermore, the FAQ server 100A may have its control unit 110A automatically or manually specify the number of clusters obtained by clustering. For example, the control unit 110A may determine the number of clusters according to the number of candidate question sentences QC as search results (candidates). The administrator terminal 300 may also accept an operation to specify the number of clusters obtained by clustering via the operation unit and transmit this operation information to the FAQ server 100A. The FAQ server 100A may have its control unit 110A receive the above operation information from the administrator terminal 300 via the communication unit 140 and determine the number of clusters based on the operation information. For example, if the number of clusters is set to a large number, the classification of candidate question sentences QC becomes more detailed.

[0119] FAQ Server 100A can narrow down the search results to a high-impact candidate by setting the number of clusters to an appropriate number. For example, the number of clusters can be set to any number of four or fewer. In this case, users can quickly grasp the trend of the refinement.

[0120] Furthermore, the FAQ server 100A may have its control unit 110A generate a network diagram NE by using a co-occurrence network to connect each feature word KW as a node ND and with an edge ED. Therefore, this network diagram NE will show the relationships between each feature word KW. Figure 19 is a network diagram NE showing an example of the relationships between each feature word KW.

[0121] In this network diagram NE, the closer the relationship between feature words KW, i.e., the higher the similarity, the shorter the distance between the nodes ND representing that feature word KW. The similarity of each feature word KW corresponds to the similarity of each cluster characterized by that feature word KW. The similarity of each cluster can be calculated by the vector similarity calculation unit 115. The communication unit 140 may transmit the generated network diagram NE to the user terminal 200. The user terminal 200 may receive and display the network diagram NE from the FAQ server 100A. This allows the user to visually understand the relationships between each feature word KW by checking the display of the network diagram NE.

[0122] Furthermore, each node ND included in the network diagram NE may be selectable by the user via the operation unit of the user terminal 200. For example, the user terminal 200 may select any node ND via the operation unit. The user terminal 200 transmits the information of the selected node ND to the FAQ server 100A. The communication unit 140 of the FAQ server 100A receives the information of the selected node ND from the user terminal 200, and the control unit 110A recognizes that a feature word KW corresponding to the selected node ND has been specified. In other words, instead of specifying a feature word KW for filtering on the FAQ chat screen, the FAQ server 100A can specify a single feature word KW by selecting a node using the network diagram NE, similar to selecting a button. Therefore, the FAQ server 100A can more intuitively narrow down the search result candidates using feature words KW.

[0123] Furthermore, each edge ED included in the network diagram NE may be selectable by the user via the operation unit of the user terminal 200. For example, the user terminal 200 may select any edge ED via the operation unit. The user terminal 200 may send information about the selected edge ED to the FAQ server 100A. The FAQ server 100A's communication unit 140 receives information about the selected edge ED from the user terminal 200 and recognizes that two feature words KW corresponding to the nodes ND connected to both ends of the selected edge ED have been specified. In other words, instead of specifying the feature words KW for filtering on the FAQ chat screen, the FAQ server 100A can specify two feature words KW simultaneously by selecting an edge using the network diagram NE, similar to selecting a button. Therefore, the FAQ server 100A can more intuitively narrow down the search result candidates using feature words KW.

[0124] In this way, the FAQ server 100A can easily grasp the relationships between each feature word KW by using the network diagram NE, and it is possible to visually narrow down the candidate search results. Narrowing using edge ED allows for a new way of displaying candidate search results. For example, if an edge ED connecting the two nodes "English" and "Template" is selected, the FAQ server 100A can use "English" and "Template" as feature words KW for narrowing down the results, and narrow down the search results (candidate question sentence QC) to "English template".

[0125] Furthermore, the control unit 110A may determine the display priority of each feature word KW based on the similarity of each feature word KW. For example, if a user specifies one feature word KW via the operation unit of the user terminal 200, other feature words KW that have a close relationship with the specified feature word KW (similarity is above a predetermined threshold) may be preferentially displayed on the display unit of the user terminal 200 and made selectable. In this case, in the network diagram NE using the co-occurrence network, the feature words KW displayed as nodes ND are limited to feature words KW with high priority, making it easier for the user to specify feature words KW that are similar to the already specified feature word KW.

[0126] Thus, using a network diagram (NE) based on a co-occurrence network makes it easier to visually understand the relationships between characteristic words (KW). Furthermore, it allows for a visual narrowing down of search result candidates.

[0127] Furthermore, the output of search result candidates may include not only text information but also image information. For example, the control unit 110A may train an image and descriptive text data before the question narrowing process to generate a trained model. The control unit 110A may use this trained model to output image information as a search result candidate, either in place of text information (question candidate sentence QC) or together with text information. The control unit 110A may also search for image information corresponding to one or more feature words KW and transmit this image information via the communication unit 140. For example, if "dog" and "white" are extracted as two feature words KW, an image of a white dog may be determined as image information and transmitted to the user terminal 200. The user terminal 200 may receive the image information from the FAQ server 100A and display the image of the white dog on the display unit.

[0128] Thus, according to the FAQ system 5 of this embodiment, the FAQ server 100A can not only cluster each candidate question sentence QC as a candidate for search results based on the similarity of the candidate question sentences QC based on the question sentence Q0 to generate a candidate question cluster, but can also display words (characteristic words) that indicate the characteristics of the candidate question cluster. As a result, the FAQ server 100A can accurately support users in narrowing down their questions.

[0129] (Third embodiment) The third embodiment is a modified version of the first embodiment. In the third embodiment, processing related to emails such as inquiry emails and reply emails is performed.

[0130] Figure 20 is a block diagram showing an example configuration of an FAQ server according to the third embodiment. In Figure 20, components similar to those in the FAQ server 100 shown in Figure 2 are denoted by the same reference numerals, and their descriptions are omitted or simplified.

[0131] The FAQ server 100B of this embodiment comprises a control unit 110B, a storage unit 130, and a communication unit 140. Compared to the control unit 110 of the first embodiment, the control unit 110B has a configuration that includes a deviation processing unit 122 and an emphasis processing unit 123.

[0132] The deviation processing unit 122 determines the correspondence between the recipient of an email (sent and received email) and the recipient in the email body, and determines whether the two recipients match or not (i.e., whether there is a deviation). If the two recipients do not match, the deviation processing unit 122 performs actions such as outputting a warning message indicating that the two recipients do not match, delaying the sending and receiving of the email, or forcibly separating the email with mismatched recipients. Outputting the above warning message can be done, for example, by sending a warning message to the administrator terminal 300. In this case, the administrator terminal 300 may receive and display the warning message (for example, by displaying it on a display unit). Therefore, if there is a mismatch in the above recipients, the FAQ server 100B suppresses the normal sending of emails, thus preventing the erroneous sending of emails.

[0133] The highlighting processing unit 123 identifies parts of the email body that require particular attention (also called parts requiring attention) and instructs the user terminal 200 or administrator terminal 300 to highlight these parts. These parts requiring attention are, for example, representative sentences in the email or sentences with high security risks. When the user terminal 200 or administrator terminal 300 receives instructions from the FAQ server 100B, it highlights the identified parts of the email body that require attention. This allows the user or respondent to easily grasp the important parts of the email.

[0134] Furthermore, the identification of parts of the email body that require verification may be performed using a natural language processing model, as described above. For example, the control unit 110B may vectorize the text data of the email body and classify the text data using the text vectors. The control unit 110B can classify the text data into multiple types (clusters) that include parts that require verification, such as representative text or highly secure text, and can verify the parts that require verification.

[0135] Figure 21 is an example of information that highlights the correspondence between email recipients and the parts of the email body that require review, BS1 and BS2. Part BS1 is, for example, a representative sentence, and part BS2 is, for example, a sentence with a high security risk.

[0136] In Figure 21, the recipient corresponding to the recipient (Cc:) is "Ito" in both the email recipient information indicating the recipient of the email and in the email body. On the other hand, in Figure 21, the recipient corresponding to the recipient (To:) is "Tanaka" in the email recipient information indicating the recipient of the email, and "Suzuki" in the email body. Therefore, the recipients (To:) do not match. The emphasis processing unit 123 may change the display format of both mismatched recipients (To:) to improve the visibility of both recipients (To:). This allows the user or respondent to easily confirm that the recipients (To:) are different and easily correct them to the correct recipient. In addition, since the recipients (To:) do not match, the deviation processing unit 122 may perform processes such as outputting warning information indicating that both recipients do not match, delaying the sending and receiving of the email, or forcibly separating the emails with mismatched recipients. This allows the FAQ server 100B to suppress email transmissions regardless of the user's or respondent's intention to send the email.

[0137] Furthermore, the emphasis processing unit 123 may improve the visibility of the parts BS1 and BS2 that need to be checked by changing the display manner of the parts BS1 and BS2 obtained by classification, for example by making them bold or using a prominent font color.

[0138] Thus, according to the FAQ system 5 of this embodiment, for example, when a respondent receives a Q&A inquiry from a user, points to note can be highlighted in the email. Therefore, the respondent can easily check the content of the question Q and detect risks in the answer A.

[0139] Next, variations of this embodiment will be described.

[0140] The functions of this embodiment may be implemented by email sending and receiving software or add-in software for email sending and receiving software.

[0141] The control unit 110B of the FAQ server 100B may derive a correspondence between the recipient of an outgoing email (e.g., To:) and the recipient extracted from the email body. The control unit 110B may derive a correspondence between the recipient of an incoming email (e.g., From:) and the sender extracted from the email body. When sending an email, the control unit 110B may determine if there is a deviation from the correspondence (e.g., a mismatch in recipients), and if there is a deviation, it may determine that there is a possibility of email misdelivery and take measures to mitigate misdelivery, such as issuing a warning, delaying transmission, or forcibly separating the email. Furthermore, when receiving an email, if there is a deviation from the correspondence of a predetermined trained model that has learned the correspondence (e.g., matching recipients), the control unit 110B may determine that there is a possibility of impersonation and take measures to mitigate misreception, such as issuing a warning, delaying reception, or forcibly separating the email.

[0142] Furthermore, the control unit 110B may have a function to determine deviations from the correspondence relationship using an AI model (trained model) that has been trained on the correspondence relationship between the email recipient (e.g., To:) in outgoing emails and the recipient extracted from the email body, and the correspondence relationship between the email recipient (e.g., From:) in received emails and the sender extracted from the email body.

[0143] Furthermore, the control unit 110B may train the above AI model each time via the email add-in, or it may train it based on email data stored in the non-chat log database 131. Also, the control unit 110B may perform inference for determining whether an email is a misdelivery or impersonation each time via the email add-in.

[0144] Furthermore, since users and respondents check the display on the user terminal 200 and administrator terminal 300 while working with emails, they can determine whether the judgment result of the AI ​​model's detection of email misdelivery and impersonation is correct or not. If the judgment result of the AI ​​model's detection of email misdelivery and impersonation is incorrect (i.e., a misjudgment), the user terminal 200 or administrator terminal 300 inputs information about the correct judgment via the terminal's operation unit and sends it to the FAQ server 100B as feedback information. The control unit 110B may input the feedback information via the communication unit 140 and input it as training data for the AI ​​model to train the AI ​​model. As a result, the FAQ system 5 can train the AI ​​model more correctly and improve the accuracy of the judgment of email misdelivery and impersonation using the AI ​​model.

[0145] As described above, the FAQ server 100 (an example of a data management device) of the above embodiment manages question sentences Q and answer sentences A to question sentences Q. The FAQ server 100 comprises a control unit 110 and a communication unit 140. The control unit 110 acquires multiple sentence data, vectorizes the multiple sentence data to create multiple sentence vectors, and extracts multiple question sentences Q from the multiple sentence data based on the multiple sentence vectors. Based on the multiple question sentence vectors, which are multiple sentence vectors corresponding to the multiple question sentences Q, the control unit 110 clusters the extracted multiple question sentences Q into question clusters that contain mutually similar question sentences Q. For each question cluster, the control unit 110 determines a representative question sentence QR that represents the question sentence Q belonging to the question cluster corresponding to the center of the question cluster. The communication unit 140 transmits registration recommendation information I1, which recommends registering the determined representative question sentence QR in the FAQ knowledge base 150 (an example of a storage device), to an administrator terminal 300 used by the administrator who manages the FAQ server 100.

[0146] As a result, the FAQ server 100 can extract question sentences from text data, streamlining the generation of QA data. Furthermore, since feature quantities that serve as input data for the pattern recognition model are not required, and feature quantities are extracted and clustered using the vectorized question sentence vectors, the processing load related to clustering can be reduced. In addition, by recommending representative question sentences Q within a cluster for registration, the FAQ server 100 can reduce the amount of modification made by the administrator to question sentences Q within a cluster, even when, for example, an administrator uses the administrator terminal 300 to fine-tune each question sentence Q before registering them in the FAQ knowledge base 150. In this way, the usability of the FAQ system 5 can be improved.

[0147] Furthermore, registration recommendation information I1 may include information recommending that the representative question text QR and the representative answer text AR, which is the answer text A corresponding to the representative question text QR, be registered in the FAQ knowledge base 150.

[0148] This allows the FAQ server 100 to manage and associate representative questions and answers from multiple similar question sentences Q. Therefore, users can obtain a representative answer A for question sentence Q by asking a question to the FAQ chatbot CB using the user terminal 200.

[0149] Furthermore, the control unit 110 may extract multiple answer sentences A from multiple text data based on multiple text vectors, and cluster the extracted multiple answer sentences A into answer clusters that contain mutually similar answer sentences A, based on multiple answer sentence vectors which are multiple text vectors corresponding to the multiple answer sentences A. For each answer cluster, the control unit 110 may determine a representative answer sentence AR that represents the answer sentence A belonging to the answer cluster corresponding to the center of the answer cluster.

[0150] As a result, the FAQ server 100 can extract answer sentence A from text data, thereby reducing the effort required to generate QA data. Furthermore, since feature quantities that serve as input data for the pattern recognition model are not required, and feature quantities are extracted and clustered using the vectorized answer sentence vectors, the processing load related to clustering can be reduced. In addition, by recommending representative answer sentence A within the answer cluster for registration, the FAQ server 100 can reduce the amount of modification made by the administrator to answer sentence A within the answer cluster, even when, for example, an administrator uses the administrator terminal 300 to make fine adjustments to each answer sentence before registering them in the FAQ knowledge base 150. In this way, the usability of the FAQ system 5 can be improved.

[0151] Furthermore, if the control unit 110 detects that a predetermined address or telephone number is included in the question or answer, it may delete the predetermined address or telephone number. After deleting the predetermined address or telephone number, the communication unit 140 may transmit the registration recommendation information I1. The predetermined address or telephone number may be, for example, an address or telephone number that has not been previously registered in the storage unit 130.

[0152] As a result, the administrator terminal 300 receives and displays the registration recommendation information, allowing the administrator to review the registration recommendation information and register the QA data with the specified address or phone number removed into the FAQ knowledge base 150.

[0153] Furthermore, the communication unit 140 may obtain an arbitrary question sentence Q0 from the user terminal 200 used by the user. The control unit 110A may perform a predetermined search process based on the arbitrary question sentence Q0, extract multiple candidate question sentences QC as search results, which are candidates for additional questions that the user may ask, and vectorize the candidate question sentences QC to create multiple candidate question sentence vectors, which are sentence vectors of the multiple candidate question sentences QC. Based on the multiple candidate question sentence vectors, the control unit 110A may cluster the extracted candidate question sentences QC into candidate question clusters that contain mutually similar candidate question sentences QC, and extract a characteristic word KW that indicates the characteristics of each candidate question cluster. The communication unit 140 may transmit the characteristic word KW of each candidate question cluster to the user terminal 200.

[0154] As a result, the FAQ server 100A not only classifies candidate question sentences (QC) into candidate question clusters, but also derives characteristic keywords (KW) that represent the characteristics of each candidate question cluster. In other words, it derives frequently occurring words in candidate question sentences (QC) within the same cluster as characteristic keywords (KW), and these characteristic keywords (KW) can be used as characteristic keywords (KW) for additional questions following any given question sentence (Q0). Therefore, users can obtain hints to narrow down their questions by asking broad questions, thereby improving questioning efficiency.

[0155] Furthermore, if there are multiple candidate question clusters, the control unit 110 may extract a different characteristic word KW for each candidate question cluster, indicating the characteristics of each cluster.

[0156] As a result, the FAQ server 100A can uniquely determine a cluster of candidate questions from the characteristic keywords KW. Therefore, the FAQ server 100 can improve its accuracy in narrowing down candidate question sentences QC that the user intends, by allowing the user to select the characteristic keywords KW.

[0157] Furthermore, the control unit 110 may determine whether the characteristic words KW of each candidate question cluster are different words based on a similar word table that defines similar words.

[0158] This allows FAQ server 100 to identify similar words that differ in spelling due to variations in notation, etc.

[0159] Furthermore, the control unit 110 may determine the number of candidate question clusters obtained by clustering according to the number of candidate question sentences QC extracted.

[0160] This allows the FAQ server 100 to appropriately maintain the number of question candidate sentences (QC) within each question candidate cluster. For example, the FAQ server 100 can prevent the number of question types, which corresponds to the number of question candidate clusters, from becoming excessive, making it difficult for users to select a question type. It can also prevent the number of question candidate sentences (QC) belonging to a question type from becoming excessive, thus preventing users from being presented with a large number of detailed questions.

[0161] Furthermore, the control unit 110 may derive co-occurrence relationship information (e.g., a network diagram NE) that shows the co-occurrence relationships of the extracted feature words KW. The communication unit 140 may transmit the co-occurrence relationship information to the user terminal 200.

[0162] As a result, the user terminal 200 can acquire co-occurrence relationship information and display a network diagram NE showing the co-occurrence relationships based on that information, allowing it to grasp the relationships between each feature word KW at a glance. Furthermore, by making node ND of the network diagram NE selectable, for example, the user terminal 200 can intuitively select feature words KE using the network diagram NE and easily select question candidates to narrow down the question.

[0163] Although various embodiments have been described above with reference to the drawings, it goes without saying that this disclosure is not limited to such examples. It is clear to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of this disclosure. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the disclosure.

[0164] Furthermore, in the above embodiment, the processor may be physically configured in any way. Also, using a programmable processor allows for changes in processing content through program modifications, thus increasing the design flexibility of the processor. The processor may consist of a single semiconductor chip, or it may be physically composed of multiple semiconductor chips. When composed of multiple semiconductor chips, each control in the above embodiment may be implemented by a separate semiconductor chip. In this case, these multiple semiconductor chips can be considered as forming a single processor. Furthermore, the processor may consist of a semiconductor chip and components having other functions (such as capacitors). Also, a single semiconductor chip may be configured to implement both the functions of the processor and other functions. Furthermore, multiple processors may be configured as a single processor. [Industrial applicability]

[0165] This disclosure is useful for data management devices, FAQ systems, data management methods, programs, etc., that can improve the usability of FAQ systems. [Explanation of symbols]

[0166] 5. FAQ System 100, 100A, 100B FAQ Server 110, 110A, 110B Control Unit 111 Sentence extraction part 112 word segmentation 113 Vectorization section 114 Classification Department 115 Vector Similarity Calculation Unit 116 Clustering section 117 Cluster-Central Computing Unit 118 Feature word extraction unit 119 Named entity extraction unit 121 Anonymization Processing Unit 122 Deviation Processing Unit 123 Emphasis Processing Unit 130,130A storage unit 131 Non-chat log databases 132 Chat Log Database 133 QA Database 134 Search Results Database 140 Communications Department 150 FAQ Knowledge Base 200 user terminals 300 Administrator terminals

Claims

1. A data management system for managing question texts and answers to said question texts, It comprises a data management device, an administrator terminal, and an FAQ knowledge base. The aforementioned data management device is It comprises a control unit and a communication unit, The control unit, Using a natural language processing model, we detect personal information contained in question Q. An anonymization process is performed to delete the detected personal information, and an anonymized question Q is generated, which is the question Q that does not contain the personal information. The aforementioned communications unit is The anonymized information indicating that the anonymization process has been performed, the question text Q containing the personal information, and the anonymized question text Q are transmitted to the administrator terminal. The aforementioned administrator terminal is The system accepts input specifying which of the aforementioned question sentence Q and the aforementioned anonymized question sentence Q should be included in the representative QA data to be registered in the FAQ knowledge base. The aforementioned communications unit is The system sends registration recommendation information to the administrator terminal, recommending that representative QA data containing the specified question be registered in the FAQ knowledge base. Data management system.

2. A data management system according to claim 1, The aforementioned administrator terminal is Upon receiving the aforementioned anonymized information, A selection screen is displayed on the display unit for selecting either the question Q containing the aforementioned personal information or the anonymized question Q. The input specified above accepts selection operations from the selection screen. Data management system.

3. A data management system according to claim 1 or 2, The aforementioned administrator terminal is The aforementioned question Q and the anonymized question Q are displayed. The system accepts permission to register the anonymized question Q in the FAQ knowledge base. The communication unit of the data management device is In response to the registration permission input, registration recommendation information is sent recommending that representative QA data, including the anonymized question text Q, be registered in the FAQ knowledge base, and, Only information containing the anonymized question Q will be transmitted to the aforementioned FAQ knowledge base. Data management system.

4. A data management system according to any one of claims 1 to 3, The control unit of the data management device is Using the aforementioned natural language processing model, the personal information contained in the question sentence Q is detected based on the word feature vectors of the question sentence Q. Data management system.

5. A data management system according to any one of claims 1 to 4, The aforementioned personal information includes at least one of the following: name, address, and telephone number. Data management system.

6. A data management system according to any one of claims 1 to 5, The control unit of the data management device is The aforementioned question Q or the anonymized question Q is summarized using the natural language processing model. Data management system.

7. A data management system according to any one of claims 1 to 6, The control unit of the data management device is The aforementioned question sentence Q or the anonymized question sentence Q is reconstructed into a question sentence with a correct Japanese structure using the natural language processing model. Data management system.

8. A data management system according to any one of claims 1 to 7, The control unit of the data management device is Using the natural language processing model, a predetermined expression contained in the question sentence Q or the anonymized question sentence Q is detected, and the detected predetermined expression is modified. Data management system.

9. A data management system according to claim 8, The aforementioned specified expression includes at least one of the following: an expression containing typographical errors, an expression containing non-compliance, an offensive expression, and an expression containing information relating to a monetary amount. Data management system.

10. A data management method for managing question texts and answer texts to said question texts, Using a natural language processing model, detect personal information contained in the question Q, The process involves anonymizing the detected personal information to generate an anonymized question Q which does not contain the personal information, The anonymized information indicating that the anonymization process has been performed, the question text Q containing the personal information, and the anonymized question text Q are transmitted from the data management device managing the question text and the answer text to the administrator terminal used by the administrator managing the data management device. The administrator terminal accepts input specifying which of the aforementioned question sentence Q and the aforementioned anonymized question sentence Q should be included in the representative QA data to be registered in the FAQ knowledge base. The data management device transmits registration recommendation information to the administrator terminal, recommending that representative QA data including the specified question text be registered in the FAQ knowledge base. A data management method having

Images