JP2021196798A - Interactive system and control method for interactive system - Google Patents

Abstract

To provide an interactive system that accurately returns an appropriate response to various input expressions of a user and to provide a control method therefor.SOLUTION: In an interactive system 1, an interactive device 10 includes: in interactive processing unit 7 that outputs a response sentence to an input sentence; a storage unit that stores: a question/answer data that associates an assumed input sentence, assuming an input sentence, with a response sentence; a synonym dictionary; a list of discrimination terms, which are terms to be distinguished from each other according to a topic of an interaction; and a similarity degree calculation model that determines a degree of similarity between the input sentence and the assumed input sentence; and a similarity degree calculation model generating unit 6 that generates the similarity degree calculation model. The interactive system generates the similarity degree calculation model in which a degree of similarity is calculated such that synonyms recorded in the synonym dictionary have a high degree of similarity and discrimination terms in the list of discrimination terms have a low degree of similarity; obtains a degree of similarity between the input sentence and the assumed input sentence by using the similarity calculation model; selects the assumed input sentence, based on the obtained degree of similarity; and outputs the response sentence corresponding to the selected assumed input sentence.SELECTED DRAWING: Figure 2

Description

The present invention relates to a dialogue system and a method for controlling the dialogue system.

Patent Document 1 describes a dialogue system configured for the purpose of enabling efficient dialogue with a user. The dialogue system acquires the user utterance converted into a predetermined format by the natural language understanding unit, updates the current dialogue state based on the user utterance in the predetermined format, uses the first policy model, and updates the dialogue state. The current subdomain is determined based on, and the behavior is determined based on the dialogue state using the second policy model associated with the current subdomain. The dialogue system also reduces the complexity of the dialogue state in the dialogue sampled from the database.

Japanese Unexamined Patent Publication No. 2019-191517

In recent years, an information processing system that automatically responds to user input such as voice and text, such as automatic response to inquiries and consultations, interactive search, and interactive device operation (hereinafter referred to as "dialogue system"). Development is in progress. According to the dialogue system, it is possible to provide a user with services such as constant and immediate inquiry response, simple information retrieval, and device operation guidance.

The dialogue system is required to, for example, respond to various input expressions of the user by absorbing the synonyms of the expressions, distinguishing the expressions that require distinction, and returning an accurate and appropriate response. The dialogue system described in Patent Document 1 aims to realize an efficient dialogue between the dialogue system and the user in the composite dialogue domain. Regarding a mechanism for improving the quality of the dialogue system from the above viewpoints. Is not disclosed.

The present invention has been made in view of such a background, and provides a dialogue system and a control method of the dialogue system capable of returning an accurate and appropriate response to various input expressions of the user. I am aiming.

One of the present inventions for achieving the above object is a dialogue system configured by using an information processing apparatus, assuming a dialogue processing unit that outputs a response statement to an input statement and the input statement. Question-and-answer data in which the assumed input sentence, which is a sentence, and the response sentence are associated with each other, a synonym dictionary, a distinctive term list, which is a list of distinctive terms that should be distinguished from each other according to the topic of dialogue, and an input sentence. A storage unit that stores a similarity calculation model for obtaining the similarity between the sentence and the assumed input sentence, and a similarity calculation model generation unit that generates the similarity calculation model, and the similarity calculation model generation unit is described above. A similarity calculation model is generated in which the similarity is calculated so that the synonyms recorded in the synonym dictionary have a high degree of similarity and the distinctive words in the distinction term list have a low degree of similarity. Obtains the similarity between the input sentence and the assumed input sentence by using the similarity calculation model, selects the assumed input sentence based on the obtained similarity, and corresponds to the selected assumed input sentence. Output the response statement.

In addition, the problems disclosed in the present application and the solutions thereof will be clarified by the column of the form for carrying out the invention and the drawings.

According to the present invention, it is possible to return an accurate and appropriate response to various input expressions of the user.

It is a figure which shows the schematic structure of the dialogue system. It is a figure which shows the hardware configuration example of the information processing apparatus which constitutes a dialogue system. It is a figure which shows the outline of the basic operation of a dialogue system. It is a figure which shows the Web page of FAQ which is the origin of question answering data. It is a figure which shows the outline of the question answering data generated based on FAQ. It is a figure which shows the manual which is the origin of the question answering data. It is a figure which shows the outline of the question answering data generated based on a manual. It is a figure which shows the specification which is the origin of the question answering data. It is a figure which shows the outline of the question answering data of a scenario table format generated based on a specification. It is a figure explaining the dialogue procedure based on the question answering data of a scenario table format. This is an example of a synonym dictionary. This is an example of a list of distinguishing terms. This is an example of a list of main terms. This is an example of a screen displayed on a user terminal. This is an example of a screen displayed on a user terminal. This is an example of a screen displayed on a user terminal. This is an example of a screen displayed on a user terminal.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description and drawings are examples for explaining the present invention, and are appropriately omitted or simplified for the sake of clarification of the description. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

In the configuration of the invention described below, the same reference numerals may be used in common among different drawings for the same parts or parts having similar functions, and duplicate description may be omitted. Further, when there are a plurality of elements having the same or similar functions, the same code may be described with different subscripts. However, if it is not necessary to distinguish between multiple elements, the subscript may be omitted for explanation.

In the following description, various types of data may be described by the expression "data", but various types of data may be expressed by other data structures such as tables and lists. Further, when the identification information is described, expressions such as "identifier" and "ID" are used, but these can be replaced with each other. Further, in the following description, the character "s" added before the reference numeral means a processing step.

FIG. 1 shows a schematic configuration of a dialogue system 1 which is an information processing system exemplified as an embodiment. As shown in the figure, the dialogue system 1 includes a dialogue device 10 and a user terminal 40 that is communicably connected to the dialogue device 10 via a communication network 30.

The dialogue device 10 generates a response sentence to a text-formatted sentence (hereinafter, referred to as “input sentence”) received from the user such as an inquiry sentence from the user, and transmits the response sentence to the user terminal 40. The dialogue between the dialogue device 10 and the user is performed, for example, in a chat format via a Web page.

FIG. 2 is a system flow diagram illustrating the main functions of the dialogue device 10. As shown in the figure, the dialogue device 10 includes the functions of the dialogue content management unit 5, the similarity calculation model generation unit 6, and the dialogue processing unit 7. The dialogue content management unit 5 includes the functions of the question answering generation unit 51, the distinctive term list generation unit 52, and the main term list generation unit 53. The dialogue processing unit 7 includes the functions of the question answering generation unit 71 and the input assisting unit 72. The dialogue device 10 includes a storage unit (not shown). The storage unit stores the text data 21, the dialogue content 22 (question response data 221, synonym dictionary 222, distinctive term list 223), the similarity calculation model 23, and the main term list 24.

The question answering generation unit 51 of the dialogue content management unit 5 generates the question answering data 221 based on the text data 21. The text data 21 is, for example, text data obtained from a business manual, a business report, a specification, a Web page, or the like. The question answering data 221 may be manually created (input) by the user.

The question answering data 221 includes information on the correspondence between the question text and the response text, the flow of dialogue, and the like. Specifically, the question answering data 221 is a combination of a question from the user and a response to the question, and when the user returns some response sentence to the question from the dialogue device 10 side to the user, the response sentence and the response sentence. It is data that aggregates the combination with the question text to be sent to the user next.

The dialogue processing unit 7 performs dialogue processing with the user by transmitting and receiving text data via the user terminal 40. When the question response generation unit 71 of the dialogue processing unit 7 receives the input sentence from the user terminal 40, the question response generation unit 71 searches the question response data 221 for the response sentence corresponding to the received input sentence. In the search, the question answering generation unit 71 uses the synonym dictionary 222, the distinctive term list 223, and the similarity calculation model 23.

The synonym dictionary 222 includes information associated with synonyms. The synonym dictionary 222 may be an existing synonym dictionary, or is automatically generated, for example, by analyzing information (question sentence, response sentence, etc.) acquired by the dialogue device 10 during a dialogue with the user. But it may be.

The distinguishing term list 223 includes a list of combinations of terms (hereinafter referred to as "distinguishing terms") that require distinction for each topic (hereinafter referred to as "topic"). The distinctive term list 223 is generated by, for example, the distinctive term list generation unit 52 of the dialogue content management unit 5 using the question and answer data 221, the synonym dictionary 222, the text data 21, and the like. Further, the distinctive term list 223 may be manually created (input) by the user.

The reason why the distinctive term list 223 is used in the above search by the dialogue processing unit 7 is to return an accurate and appropriate response to various input expressions of the user. Even if words have similar notations, it may be necessary to distinguish expressions depending on the topic of dialogue, but the synonym dictionary 222 is for handling synonymous expressions in a unified manner, and is a synonym dictionary. If the question answering generation unit 71 performs a search based only on 222, it may not be possible to handle expressions that require distinction according to the topic of dialogue. By using the distinctive term list 223 together, the dialogue device 10 of the present embodiment absorbs the synonyms of expressions for various input expressions of the user, distinguishes expressions that require distinction, and provides an accurate and appropriate response. return.

The similarity calculation model 23 is a function or a machine learning model for obtaining the similarity between the input sentence and the assumed input sentence in the question response data 221. The similarity calculation model 23 is generated by the similarity calculation model generation unit 6. By using the similarity calculation model 23 when the dialogue processing unit 7 interacts with the user, an input sentence (hereinafter, referred to as “assumed input sentence”) matching the input sentence is defined in the question response data 221. Even if there is no input sentence, it is specified which input sentence in the question / answer data 221 corresponds to.

The main term list generation unit 53 of the dialogue content management unit 5 is a main term referred to by the input assist unit 72, which is a function of assisting the dialogue processing unit 7 of the dialogue device 10 to input an appropriate input sentence. Generate Listing 24. The main term list 24 may be manually created (input) by the user, for example.

FIG. 3 shows an example of the hardware configuration of the information processing device 100 constituting the dialogue device 10 and the user terminal 40. As shown in the figure, the information processing device 100 includes a processor 101, a main storage device 102, a communication device 103, an input device 104, an output device 105, and an auxiliary storage device 106.

The processor 101 is, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), an AI (Artificial Intelligence) chip, an FPGA (Field Programmable Gate Array), a SoC (System on Chip), and an ASIC. (Application Specific Integrated Circuit) etc. are used.

The main storage device 102 is a device for storing programs and data, and is, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), a non-volatile memory (NVRAM (Non Volatile RAM)), and the like.

The communication device 103 is a device that communicates with other information processing devices such as a user terminal via a communication network, a communication cable, or the like, and is a wireless or wired communication module (wireless communication module, communication network adapter, USB). Module etc.).

The input device 104 and the output device 105 form a user interface of the dialogue device 10. The input device 104 is a user interface that accepts user input and data input from the outside, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a voice input device (for example, a microphone), and the like. The output device 105 is a user interface that outputs various information to the user, and is a display device (liquid crystal display, organic EL panel, etc.) that displays various information, and a voice output device (for example, a speaker) that outputs various information by voice. ), Printers that print on paper media, etc.

The auxiliary storage device 106 is a device for storing programs and data, for example, an SSD (Solid State Drive), a hard disk drive, and an optical storage medium (CD (Compact Disc)).
), DVD (Digital Versatile Disc), etc.), IC card, SD card, etc. The auxiliary storage device 106 stores programs and data for realizing the functions of the dialogue device 10. The auxiliary storage device 106 can write / read programs and data via the reading device of the recording medium and the communication device 103. Programs and data stored (stored) in the auxiliary storage device 106 are read out to the main storage device 102 at any time. When the processor 101 reads out and executes the program stored in the main storage device 102, each function of the dialogue device 10 is realized.

All or part of the functions of the dialogue device 10 may be realized by hardware such as another arithmetic unit (for example, FPGA (Field Programable Gate Array) or ASIC (Application Specific Integrated Circuit)).

The information processing device 100 is, for example, a personal computer (desktop type or notebook type), a smartphone, a tablet, a general-purpose machine, or the like. All or part of the information processing apparatus 100 may be realized by using virtual information processing resources such as a cloud server provided by a cloud system.

Subsequently, the function of the question answering generation unit 51 of the dialogue content management unit 5 to generate the question answering data 221 will be specifically described.

4A and 4B illustrate the case where the question answering generation unit 51 generates the question answering data 221 based on the text data acquired from the Web page on which the FAQ (Frequently Asked Question) is posted.

FIG. 4A is an example of the text data 21 that is the generation source of the question answering data 221 and is an example of a Web page on which the FAQ is posted. As shown in the figure, in the illustrated Web page, the individual question sentence 301 and the response sentence 302 in the FAQ are classified and described by titles such as "how to apply for service A" and "how to change the address". Has been done.

FIG. 4B is an example of question answering data 221 generated based on the text data 21 shown in FIG. 4A. As shown in the figure, the illustrated question-answering data 221 is a text (hereinafter, referred to as "QA vs. data") composed of a pair of a question sentence and a response sentence. The dialogue content management unit 5 extracts a combination of the question sentence and the response sentence corresponding to the question sentence from the Web page, and generates QA pair data based on the extracted combination. In the case of this example, the FAQ question is the assumed input sentence.

5A and 5B illustrate the case where the question answering generation unit 51 generates the question answering data 221 from the text data 21 consisting of a manual such as a business manual described according to a certain structure. The dialogue content management unit 5 generates QA pair data and question / answer data 221 that aggregates the QA pair data based on the characteristics of the chapters and stylistic structure in the text data 21 composed of the manual.

FIG. 5A is an example of the text data 21 that is the generation source of the question answering data 221 and is the data extracted from a part of the manual. In the illustrated manual, "About Service A" is described as the title of Chapter 5, and "How to apply for Service A" is described in Section 1 (Section 5.1) of Chapter 5. In the case of this example, the dialogue content management unit 5 combines the title "Service A" in Chapter 5 and the title "Application method" in Section 5.1 with "no", "How to apply for Service A?" Is generated, and the content described in Section 5.1 is extracted as a response to the question. In addition, the illustrated manual has a chapter "6. Change of basic information", and there are items such as "6.1 address" for each type of basic information. The dialogue content management unit 5 asks "How to change the address?" By changing the description of "basic information" in "6. Change of basic information" to the title of the section (for example, "address"). Is generated, and the content described in Section 6.1 is extracted as a response to the question.

FIG. 5B is an example of question answering data 221 generated from the text data 21 shown in FIG. 5A. As shown in the figure, the dialogue content management unit 5 generates QA vs. data with the question text "How to apply for service A?" And the response text "You can complete the procedure from the store or website." do. In addition, the dialogue content management department generates QA vs. data that pairs the question text "How to change the address?" And the response text "You can complete the procedure from the website."

6A and 6B show text data 21 and question answering when the dialogue device 10 provides a service for presenting information (model number, product name, etc.) that identifies a product that meets the user's wishes through dialogue with the user. This is an example of data 221. For example, when a user wants to purchase a certain product, the dialogue device 10 presents the product according to the user's wishes through a dialogue with the user. 6A and 6B illustrate a case where a user intends to purchase a BTO (Build To Order) type personal computer (hereinafter referred to as "PC") as a certain product. The question answering generation unit 51 generates question answering data 221 based on the text data 21 of the specification table in which the specifications of the components of the PC as a product are described.

FIG. 6A is an example of text data 21, and is a specification table in which specifications of components of a personal computer, which is a commercial product, are described. As shown in the figure, the specification table includes information associated with the product name, price, storage device, port, etc. for each product.

FIG. 6B is an example of question answering data 221 generated by the question answering generation unit 51 based on the text data 21 of the specification table shown in FIG. 6A. As shown in the figure, the example question answering data 221 has an entry (record) identifier (ID), a question to be asked to the user, and a expected user response unit (assumed input sentence) to the question. ) And the information associated with the next question to be asked to the user.

The question answering data 221 shown in FIG. 6B includes information for determining the next question according to the response sentence (assumed input sentence) from the user to the question asked by the dialogue processing unit 7 of the dialogue device 10. That is, the question-answering data 221 is information (hereinafter, referred to as a "scenario table") that summarizes the scenarios for the dialogue processing unit 7 to proceed with the dialogue processing with the user in a table format.

FIG. 6C is a diagram illustrating a procedure of dialogue (chat) performed by the dialogue processing unit 7 with the user. The dialogue processing unit 7 first sends to the user terminal 40 the first question Q1 (for example, "Which storage device do you want?") In the scenario table shown in FIG. 6B. When the user returns a response sentence (for example, "HDD", "hard disk device", etc., here, as an example, it is assumed that the expected input sentence A1 or a response sentence close to the A1) is returned to the question Q1, the dialogue process is performed. Part 7 searches the scenario table and acquires the next question Q2 associated with the question Q1 and the assumed input sentence A1. If the question Q2 is, for example, a question "What is the standard of the USB port?" For the attribute of "USB", the dialogue processing unit 7 sends the question Q2 to the user terminal 40 and asks a question. The next question is acquired from the scenario table according to the user's response to Q2, and the acquired question is sent to the user terminal 40. As described above, the dialogue processing unit 7 finally identifies one product and sends the product name to the user terminal 40.

FIG. 7 shows an example of the synonym dictionary 222. As shown in the figure, the synonym dictionary 222 includes information associated with expressions having similar meanings such as synonyms and synonyms.

FIG. 8 shows an example of the distinctive term list 223. Illustrated Distinguished Term List 223 is an example of Distinguished Term List 223 for the topic "Insurance". The distinctive term list 223 is a list of terms that should be distinguished according to the topic of dialogue, even if the notations are similar, as distinctive terms that are independent of each other. For example, "insurance premium" and "insurance money" have similar notations and are related terms, but in "insurance industry", it is necessary to clearly distinguish between them. In the illustrated distinctive term list 223, "insurance premium" and "insurance claim" are registered as separate distinctive terms (as data of different entries (records)). In the illustrated distinguished term list 223, when a synonym exists in the distinguished term, one of the synonyms is used as the representative word, and the other synonyms are associated with the representative word (as data of the same entry (record)). be registered.

The distinctive term list 223 can take various aspects. For example, when the topic is "inquiries about products and services", for example, the expressions of product names and service names may be similar, such as "plan A" and "plan B", and in this case, distinction is made. "Plan A" and "Plan B" are registered as distinct terms in the term list 223. Further, for example, when the topic relates to the contents of the specification table shown in FIG. 5A, for example, "USB2.0" and "USB3.0" are registered as distinct terms.

The distinctive term list 223 is generated, for example, by the distinctive term list generation unit 52 of the dialogue content management unit 5 extracting the distinctive terms from the question answering data 221 and the text data 21 and aggregating the extracted distinctive terms.

A process of generating the distinctive term list 223 by the distinctive term list generation unit 52 will be described with reference to FIG.

For example, when the question / answer data 221 is in the form of a pair of a question sentence and a response sentence exemplified in FIG. 4B or FIG. Extract certain contained nouns and compound nouns. Terms that require distinction are often nouns or compound nouns, such as "plan A" and "plan B", "insurance premium" and "insurance money" for inquiries regarding insurance applications. It needs to be understood separately. The distinctive term is not limited to a noun or a compound noun, and may be another part of speech. The part of speech of the distinctive term can be set according to, for example, the topic of dialogue.

Further, the distinguishing term list generation unit 52 extracts the distinguishing term from the text data 21 by, for example, performing morphological analysis, syntactic analysis, or the like. For example, the distinctive term list generation unit 52 uses the extracted distinctive terms as representative words, and associates synonyms based on the synonym dictionary 222 with each representative word. For example, of the "insurance claims," "insurance premiums," and "passports" extracted as distinctive terms, "insurance premiums" are synonymous with "stakes," and "passports" are synonymous with the synonym dictionary 222. When registered in, the distinctive term list generation unit 52 registers "insurance money" and "insurance premium" as individual distinctive terms in another column of the distinctive term list 223, and "insurance premium" is "insurance premium". As a synonym for "insurance", and as a synonym for "passport" for "passport", they are registered in the same column of the distinctive term list 223, respectively.

When extracting the distinguishing terms from the text data 21 such as the specification table, the terms of the attribute values in the table of the specification table need to be distinguished from each other. Therefore, the distinguishing term list generation unit 52 uses these terms. Extract as a distinguishing term. Further, when the distinction term is extracted from the question response data 221 in the form of a scenario table, the distinction term list generation unit 52 uses a noun or a compound from the question sentence, the response sentence (assumed input sentence), and the following question. Extract nouns as distinctive terms.

The distinguishing term list generation unit 52 may accept editing of the contents of the distinguishing term list 223 from the user. Further, the distinctive term list 223 may be created by a person with reference to the question answering data 221 and the text data 21.

By the way, when the distinguishing term is extracted only from the question answering data 221, the number of terms may increase and the term (noise) that does not need to be distinguished may be extracted. Therefore, for example, among the distinguishing terms extracted from the question answering data 221, the terms not described in the predetermined place of the text data 21 are filtered (excluded), and only the terms described in the predetermined place are used as the distinguishing terms. It may be extracted. For example, a predetermined place in the text data 21 of FIG. 4A is a place where a category title is described, such as "method of applying for service A" and "method of changing an address". Further, for example, a predetermined place in the text data 21 of FIG. 5A is a place where chapter headings such as "service A", "application method", and "cancellation method" are described. Since the headings of categories and chapters are classified according to the products and inquiry contents that need to be distinguished, for example, the terms appearing in the headings of categories and chapters may be extracted as representative terms of the distinguishing terms. ..

The similarity calculation model 23 is generated by the similarity calculation model generation unit 6 based on the synonym dictionary 222 and the distinctive term list 223. In the similarity calculation model 23, for example, the input statement is X and the assumed input statement is Y, which can be expressed as a function f (X, Y). As the degree of similarity, for example, the editing distance between X and Y, the BLEU value of X and Y when X is regarded as a translated sentence and Y is regarded as a correct sentence (reference sentence), and the like can be adopted. Further, as the similarity, for example, a similarity other than the edit distance or the BLUE value may be used.

The similarity calculation model 23 treats words having different notations (surface expressions) as the same word based on the synonym dictionary 222. Therefore, it is preferable to unify synonyms into one term in advance using the synonym dictionary 222 and the distinctive term list 223 (all terms such as "price", "charge", and "price" are referred to as "price". Unify the terms, etc.).

With reference to FIG. 2, a process in which the similarity calculation model generation unit 6 generates the similarity calculation model 23 will be described. In the similarity calculation model generation unit 6, the similarity between the terms registered in the same column in the synonym dictionary 222 increases, and the similarity between the distinct terms registered in the distinct term list 223 decreases. Such a similarity calculation model 23 is generated. Hereinafter, an example of the similarity calculation model 23 (first model f1 (X, Y) and second model f2 (X, Y)) will be described.

In the calculation method using the first model f1 (X, Y), the input sentence X and the assumed input sentence Y are first decomposed into terms such as words and phrases by morphological analysis and syntactic analysis, and the set of decomposed terms is decomposed into each. S1 = {x1, x2, ..., Xn}, S2 = {y1, y2, ..., ym}, and the degree of similarity between S1 and S2 is calculated. It should be noted that a specific part of speech such as a particle or a specific term may be defined as a stop word, and they may be excluded from the sets S1 and S2.

The degree of similarity between S1 and S2 can be calculated as the degree of similarity between sets by a method such as a Jaccard index or a Dice coefficient. Further, the similarity s (xi, yj) between the words xi and yj is defined by the closeness of the notation of xi and yj (negative number of editing distance, etc.), and the set S1 is defined by WMD (Word Mover's Distance) or the like. It is also possible to calculate the similarity of the set S2. Further, a method based on the word duplication degree of the set S1 and the set S2 can also be used. There is also a method of using an editing distance such as the Levenshtein distance between X and Y. The degree of similarity between S1 and S2 may be calculated by another calculation method. In this way, in the first model f1 (X, Y), the similarity is calculated based on the notation of X and Y. The distance is converted to the degree of similarity by taking a negative number or the like.

In the calculation method based on the first model f1 (X, Y), the similarity is calculated based on the notation, so that the process of calculating the similarity is clear and highly explanatory. Therefore, in order to improve the response accuracy of the dialogue device 10, for example, when the administrator or the user of the dialogue device 10 wants to know why such a response is made, the similarity calculation process is performed. Can be tracked. Further, since the method of calculating the similarity based on the first model f1 (X, Y) is based on the notation, there is an advantage that the assumed input sentence can be correctly estimated when the notations are similar.

In the calculation method based on the first model f1 (X, Y), the closeness is calculated not by the meaning of the word but by the notation, so that even similar expressions may be treated as different words. For example, "use" and "use" have no common characters, so the degree of similarity is small.

Therefore, in the method of calculating the similarity by the second model f2 (X, Y), the interword similarity s (w1, w2) of the words w1 and w2 is used. The inter-word similarity s (w1, w2) is the similarity between individual words, and as a calculation method thereof, for example, there is a method using the cosine similarity of the distributed expression of w1 and w2. There is also a method of using the distance between word concepts of w1 and w2.

In the calculation method using the interword similarity s (w1, w2), for example, the length of the path between two words on WordNet can be used. Using these word-to-word similarities, the distance between the word sets S1 and S2 is calculated using, for example, WMD. Since the second model f2 (X, Y) uses the similarity of distributed expressions of words and the conceptual distance instead of the notation, the similarity between words having similar meanings is high even if the notations are different. However, since the interword similarity s (w1, w2) is not a calculation method based on the notation, the explanatory property may be low.

In the method of calculating the degree of similarity using the second model f2 (X, Y), the degree of similarity between words (w1, w2) was calculated using the degree of similarity and the conceptual distance of the distributed expression instead of the notation. However, the value of the degree of similarity may be corrected by using the synonym dictionary 222 and the distinctive term list 223. The following is an example of a method of correcting the similarity value when the similarity is normalized to take a real number between 0 and 1.

First, the terms w1 and w2 described as synonymous expressions in the synonym dictionary 222 are set to s (w1, w2) = c1 with a predetermined large value of 0 or more and 1 or less as c1. Note that w1 and w2 may be regarded as the same term and c1 = 1. Further, for the two distinct terms (w1, w2) described in the distinct term list 223, a predetermined small value of 0 or more and 1 or less is set as c2, and s (w1, w2) = c2. For example, c2 = 0 may be set. By making such a correction, the degree of similarity between the terms listed in the distinctive term list 223 becomes small.

In particular, when WordNet's interconceptual similarity is used, the similarity by the second model f2 (X, Y) may be the same value for a plurality of assumed question sentences Y. In such a case, the similarity calculation model f (X, Y) obtained by synthesizing f1 and f2 is used.
f (X, Y) = C × f1 (X, Y) + f2 (X, Y)
May be. The synthesis coefficient C is, for example, C = 0.1. Further, when f1 (X, Y) is sufficiently large, the notations are quite similar, and the value of f1 (X, Y) is reliable. Therefore, if the threshold value H is set in advance and f1 is larger than H as in the following equation, the value of f1 may be adopted as the degree of similarity.
When f (X, Y) = f1 (X, Y) f1 (X, Y)> H
... Equation 1
When f (X, Y) = C × f1 (X, Y) + f2 (X, Y) f1 (X, Y) ≦ H
・ ・ ・ Equation 2

By such a calculation method, f1 is adopted when the notations are similar, the explanation for the result is improved, and even when the value of f1 is low and the notations are not similar, the meaning between words is high. When are similar, a high degree of similarity can be obtained by mainly using f2 and using the synthetic similarity of f2 and f1.

The dialogue processing unit 7 calculates the similarity between the input sentence and the assumed input sentence using the similarity calculation model 23 generated as described above, and acquires the closest assumed input sentence. For example, in a dialogue with a user based on the scenario table shown in FIG. 6B, when the dialogue device 10 asks "Which storage device should I use?", The input sentence is "Hard Disk Drive". , Among the values (HDD, SSD) in the attribute of "storage device", which value is closest to is calculated.

From the above, it is possible to obtain a high degree of similarity when the notations are similar, and a high degree of similarity when the meanings are similar even if the notations are different. In addition, the similarity can be calculated low for terms that need to be distinguished. This makes it possible to improve the response accuracy. By using the similarity calculation model 23 as described above, the dialogue device 10 can identify the assumed input sentence having the highest similarity with the input sentence and return an appropriate response sentence.

By the way, since the expressions of the input sentence are various, the input sentence lacks information for the dialogue device 10 to return an appropriate response sentence, or a word to be input separately as a distinctive term and another word. May be mixed, and these are factors that reduce the response accuracy of the dialogue device 10. The input assisting unit 72 of the dialogue processing unit 7 shown in FIG. 2 assists the user so that an appropriate input sentence can be input, and aims to improve the response accuracy of the dialogue device 10.

The input assist unit 72 displays the terms used in the assumed input sentence in the process of inputting the input sentence by the user. The dialogue device 10 manages the terms included in the assumed input sentence as the main term list 24.

As shown in FIG. 2, the main term list generation unit 53 of the dialogue device 10 generates the main term list 24 based on the question answering data 221. In the dialogue processing unit 7, when the dialogue processing with the user is performed via the user terminal 40, the question response generation unit 71 estimates the assumed input sentence, and the input assisting unit 72 estimates from the main term list 24. The main word corresponding to the input sentence is presented to the user terminal 40.

FIG. 9 shows an example of the main term list 24. The main term list 24 is generated, for example, by the main term list generation unit 53 extracting specific parts of speech (nouns, verbs, etc.) included in each assumed input sentence of the question / answer data 221 by morphological analysis. .. The illustrated list of key terms 24 is generated based on the question answering data 221 shown in FIGS. 4A and 5A.

10A to 10D show an example of a screen displayed by the input assisting unit 72 in the process of inputting an input sentence by the user.

For example, as shown in FIG. 10A, when the user inputs "where can I apply?" In the user input field 41 displayed on the user terminal 40, the dialogue processing unit 7 responds to the input sentence X by each. The similarity f (X, Y) with respect to the assumed question sentence Y is calculated, and the upper p cases with high similarity (p is predetermined. For example, p = 10) are selected.

Here, when the value of the highest similarity is M (M = maxf (X, Y)), the assumed question sentence Y including the two assumed input sentences "service A" and "service B" is selected. And. At this time, in the main term list 24, the main term w (in this case, any of "service A", "service B", and "application") included in the selected assumed input sentence Y is used as the input sentence X. In addition, when X'is set, when the value of M'= maxf (X', Y) is larger than M by a predetermined constant value K or more, that is, when M'-M> K, w is set. It is selected as a candidate for supplementing insufficient information 42. That is, the main term w selected as the missing information complement candidate 42 is added to the input sentence X so that the similarity between the input sentence X and the assumed input sentence Y is high.

FIG. 10A shows a case where “service A” and “service B” are selected as the missing information complement candidate 42. The dialogue processing unit 7 displays the selected shortage information complement candidate 42 “service A” and “service B” so as to be selectable on the user terminal 40. In FIG. 10A, when the user selects, for example, "service A", the input assisting unit 72, as shown in FIG. 10B, "where is the application for service A" complemented by the selected word "service A". Is it possible? ”Is displayed.

In addition, for example, the user may input an erroneous term because he / she does not know the term or confuses the term. For example, as shown in FIG. 10C, when the input sentence X is the sentence "Where can I apply for service A?", The user confuses "service A" with "service B" or "service C". May be. In this case, the dialogue processing unit 7 displays "service B" and "service C" as different candidates 43 at the display position of "service A" in the input sentence X so as to be selectable on the user terminal 40. As a result, the user can be made aware that the terms may be confused, and erroneous input of the input sentence X can be reduced. In this mechanism, for example, the user is asking where the application for "service A" can be made, and since "service A" is in the main term list 24, "service A" is set to w1 and the main term list 24. It can be realized by calculating the inter-word similarity s (w1, w2) with the term in the above as w2 and selecting the term w2 similar to w1 as the term to be presented. The term w2 may be selected in a predetermined number in descending order of similarity with the term w1, or the term w2 having a similarity of a certain threshold value or more may be selected. The user can modify the input, if desired, by selecting one of the presented terms w2 and replacing the term w1 with the term w2.

When the user inputs an input sentence, if the user can know the terms that are effective for calculating the similarity of the dialogue device 10 and the terms that are not, the user can efficiently input the input sentence. For example, the word set of the input sentence is S1, and then the reference word set is S3. The reference word set S3 may be a set of words listed in all assumed input sentences, or may be a set of terms listed in the distinctive term list 223. The dialogue device 10 calculates g = max {s (w1, w2)} for the word w1 of the element of S1. In addition, "max" in the formula is taken for all words w2 in S3. At this time, as shown in FIG. 10D, in the input field of the input sentence X (hereinafter referred to as the user input field 41), the term w1 in which the value of g is equal to or higher than a predetermined threshold value can be distinguished from other terms. It may be displayed. In the example of the figure, the term w1 is highlighted (highlighted) (rectangular dotted frame 44 in the figure). Further, the term w1 that is equal to or higher than the threshold value may be indicated by a gradation display in which the density is changed according to the value of g. By making the term w1 distinguishable from other terms in this way, the user can understand the term that the dialogue device 10 emphasizes in responding, and the input for the dialogue device 10 to respond accurately. Can be prompted to the user.

The word-to-word similarity s (w1, w2) may be one or a combination of the various word-to-word similarities (cosine similarity of distributed expressions, distance between word concepts, etc.). good. Further, the interword similarity s (w1, w2) may be used for weighting corresponding to the synthetic count C of the similarity calculation model f (X, Y), for example. The similarity f (X, Y) may be used instead of the similarity s (w1, w2) between words.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and includes various modifications. Further, for example, the above-described embodiment is described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations. Further, it is possible to add, delete, or replace a part of the configuration of each embodiment with other configurations.

Further, each of the above configurations, functional units, processing units, processing means and the like may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be stored in memory, hard disks, and recording devices such as SSDs (Solid State Drives).
It can be placed on a recording medium such as an IC card, SD card, or DVD.

Further, the arrangement form of various functional units, various processing units, and various databases of each information processing apparatus described above is only an example. The arrangement form of various function units, various processing units, and various databases can be changed to the optimum arrangement form from the viewpoint of the performance, processing efficiency, communication efficiency, and the like of the hardware and software included in these devices.

Further, the configuration of the database (schema, etc.) for storing various data described above can be flexibly changed from the viewpoints of efficient use of resources, improvement of processing efficiency, improvement of access efficiency, improvement of search efficiency, and the like.

1 Dialogue system, 5 Dialogue content management unit, 51 Question response generation unit, 52 Distinguished term list generation unit, 53 Main term list generation unit, 6 Similarity calculation model generation unit, 7 Dialogue processing unit, 21 Text data, 22 Dialogue content 221 Question and answer data, 222
Synonym dictionary, 223 distinct term list, 23 similarity calculation model, 24 main term list, 30 communication network, 40 user terminal, 41 user input field

Claims (16)

It is a dialogue system configured using an information processing device.
A dialogue processing unit that outputs a response statement to the input statement,
A distinction that is a list of distinctive terms that should be distinguished from each other according to the question response data, the synonym dictionary, and the topic of dialogue in which the assumed input sentence and the response sentence are associated with each other. A storage unit that stores a list of terms, a similarity calculation model for finding the similarity between an input sentence and an assumed input sentence, and a storage unit.
The similarity calculation model generation unit that generates the similarity calculation model,
Equipped with
The similarity calculation model generation unit calculates the similarity so that the synonyms recorded in the synonym dictionary have high similarity and the distinct terms in the distinct term list have low similarity. Generate a calculated model and
The dialogue processing unit obtains the similarity between the input sentence and the assumed input sentence by using the similarity calculation model, selects the assumed input sentence based on the obtained similarity, and selects the assumed input. Output the response statement corresponding to the statement,
Dialogue system. The dialogue system according to claim 1.
A dialogue content management unit is further provided to generate the question-and-answer data based on text data consisting of a document having a chapter structure, and to extract the distinctive terms from the chapter headings in the text data or the notation indicating the chapter classification. Prepare, prepare
Dialogue system. The dialogue system according to claim 1.
The similarity calculation model generation unit includes a first similarity calculation model based on the similarity between a set of terms included in each of the input sentence and the assumed input sentence, and each of the input sentence and the assumed input sentence. A function obtained by generating a second similarity calculation model based on the similarity between words included in the above and synthesizing the first similarity calculation model and the second similarity calculation model is the similarity calculation model. Generate as,
Dialogue system. The dialogue system according to claim 3.
When the similarity between the input sentence calculated by the first similarity calculation model and the assumed input sentence is equal to or more than a predetermined threshold value, the dialogue processing unit uses the first similarity calculation model as the similarity. Adopted as a degree calculation model,
Dialogue system. The dialogue system according to claim 1.
The dialogue processing unit calculates the similarity after unifying the terms included in the input sentence into the terms included in the assumed input sentence using the synonym dictionary or the distinctive term list.
Dialogue system. The dialogue system according to claim 1.
The storage unit stores a list of main terms extracted by using the terms included in each of the assumed input sentences as the main terms.
The dialogue processing unit outputs the main term for the assumed input sentence similar to the input sentence in a state in which the user can select the input sentence, and outputs the input sentence with the main term selected by the user. Complement,
Dialogue system. The dialogue system according to claim 1.
The storage unit stores a list of main terms extracted by using the terms included in each of the assumed input sentences as the main terms.
The dialogue processing unit outputs the main terms of the assumed input sentence similar to the input sentence in a state where it can be replaced with some terms included in the input sentence, and outputs some of the terms. When the user's instruction to replace is received, a part of the input sentence is replaced with the instructed main term.
Dialogue system. The dialogue system according to claim 1.
The storage unit stores a list of main terms extracted by using the terms included in each of the assumed input sentences as the main terms.
The dialogue processing unit displays, among the words included in the input sentence, words whose similarity with the word included in the assumed input sentence is equal to or higher than a predetermined threshold value so as to be identifiable in the input sentence.
Dialogue system. It is a control method of a dialogue system configured by using an information processing device.
The dialogue system
Step to output response statement to input statement,
A distinction that is a list of distinct terms that should be distinguished from each other according to the question response data, the synonym dictionary, and the topic of dialogue in which the assumed input sentence and the response sentence are associated with each other. A step to store a list of terms and a similarity calculation model for finding the similarity between an input sentence and an assumed input sentence.
Steps to generate the similarity calculation model,
A step of generating a similarity calculation model for calculating the similarity so that the synonyms recorded in the synonym dictionary have a high degree of similarity and the distinctive terms in the distinction term list have a low degree of similarity, and
The similarity between the input sentence and the assumed input sentence is obtained by using the similarity calculation model, the assumed input sentence is selected based on the obtained similarity, and the response sentence corresponding to the selected assumed input sentence is selected. To output the step,
How to control the dialogue system. The method for controlling a dialogue system according to claim 9.
The dialogue system generates the question-and-answer data based on the text data consisting of a document having a chapter structure, and extracts the distinctive terms from the chapter headings in the text data or the notation indicating the classification of the chapters. Steps to further prepare the content management department,
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 9.
The dialogue system includes a first similarity calculation model based on the similarity between a set of terms included in each of the input sentence and the assumed input sentence, and words included in each of the input sentence and the assumed input sentence. A step of generating a second similarity calculation model based on the similarity between each other, and generating a function obtained by synthesizing the first similarity calculation model and the second similarity calculation model as the similarity calculation model. ,
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 11.
When the dialogue system has a similarity between the input sentence calculated by the first similarity calculation model and the assumed input sentence of a predetermined threshold value or more, the first similarity calculation model is used as the similarity degree. Steps to be adopted as a calculation model,
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 9.
A step in which the dialogue system unifies the terms included in the input sentence into the terms included in the assumed input sentence using the synonym dictionary or the distinctive term list, and then calculates the similarity.
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 9.
The dialogue system
A step of storing a list of main terms extracted from the terms included in each of the assumed input sentences as the main terms, and
A step of outputting the main term of the assumed input sentence similar to the input sentence in a state in which the user can select it, and complementing the input sentence with the main term selected by the user.
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 9.
The dialogue system
A step of storing a list of main terms extracted from the terms included in each of the assumed input sentences as the main terms, and
A user who outputs the main term of the assumed input sentence similar to the input sentence in a state where it can be replaced with a part of the terms included in the input sentence and replaces the part of the terms. When the instruction of the above is received, a part of the input sentence is replaced with the instructed main term.
How to control the dialogue system to further execute. The method for controlling a dialogue system according to claim 9.
A step in which the dialogue system stores a list of main terms extracted from the terms included in each of the assumed input sentences as the main terms, and
A step of identifiablely displaying a word included in the input sentence whose similarity with the word included in the assumed input sentence is equal to or higher than a predetermined threshold value in the input sentence.
How to control the dialogue system to further execute.

Images