JP7334293B1 - Information processing system, information processing method and program - Google Patents

Abstract

The present invention provides a technology for automatically responding more appropriately to inquiries. An information processing system (1) includes a connector DB (174), a voice recognition processing section (153), a narrowing down processing section (155), and an integration processing section (156). The connector DB 174 stores first attribute information regarding the assumed inquiry. Connector DB 174 stores second attribute information regarding answers to assumed inquiries. The speech recognition processing unit 153 acquires the contents of the inquiry. The narrowing-down processing unit 155 and the integration processing unit 156 specify a response including an answer to the inquiry based on the attribute information included in the voice recognition result obtained by the voice recognition process, the first attribute information, and the second attribute information. [Selection diagram] Figure 4

Description

The present invention relates to an information processing system, an information processing method, and a program.

2. Description of the Related Art Conventionally, a technique of recognizing a user's speech and automatically responding to an inquiry has been known.
For example, in a voice automatic question answering device described in Patent Document 1, a voice recognition engine is used to perform voice recognition of a free speech type question, etc., and the voice recognition result is corrected by discourse analysis or the like, and the content of the question is corrected. We are aiming for appropriate recognition.

Japanese Patent Application Laid-Open No. 2003-263190

However, in the conventional technology, even if an attempt is made to identify an answer based on an inquiry input by voice, there are cases where an appropriate answer cannot be derived due to insufficient information included in the inquiry. .
In addition, when a freely spoken inquiry is speech-recognized using a general-purpose speech recognition engine, there is a possibility that the content of the inquiry cannot be appropriately speech-recognized.
In addition, the voice recognition process and the process of identifying the answer may take a long time, and there is a possibility that the inquiry cannot be answered promptly.
Furthermore, since the expressions used by the speakers making inquiries are diverse, the system may not be sufficiently capable of handling such diversity of expressions.
When one of these reasons or a plurality of reasons occur in combination, it becomes difficult to provide an appropriate automatic response to an inquiry by voice.
It should be noted that such a problem may occur not only in the case of voice recognition of voice inquiries, but also in inquiries by text input such as chat.

SUMMARY OF THE INVENTION An object of the present invention is to provide a technique for making a more appropriate automatic response to an inquiry.

In order to achieve the above object, an information processing system according to one aspect of the present invention includes:
An information processing system including one or more information processing devices,
a first attribute information storage means for storing first attribute information related to an envisioned inquiry;
a second attribute information storage means for storing second attribute information relating to answers to the assumed inquiries;
inquiry content acquisition means for acquiring the content of an inquiry;
response specifying means for specifying a response including an answer to the inquiry based on the attribute information of the content of the inquiry and the first attribute information and the second attribute information;
characterized by comprising

In order to achieve the above object, an information processing system according to another aspect of the present invention includes:
An information processing system including one or more information processing devices,
voice recognition means for acquiring voice recognition results obtained by performing voice recognition processing on an utterance for an inquiry;
language processing means for specifying, by language processing, candidate answers that match the inquiry from the semantic content expressed by the speech recognition results of the speech recognition means;
extracting words included in the inquiry from the speech recognition result of the speech recognition means; attribute processing means for identifying suitable answer candidates;
response specifying means for specifying a response including an answer to the inquiry based on the answer candidate specified by the language processing means and the answer candidate specified by the attribute processing means;
characterized by comprising

In order to achieve the above object, an information processing system according to another aspect of the present invention includes:
An information processing system including one or more information processing devices,
Consists of one or more attributes associated with answers to the assumed inquiry based on first attribute information on the assumed inquiry and second attribute information on the answer to the assumed inquiry. a search data generation means for generating search data for
inquiry content acquisition means for acquiring the content of an inquiry;
response identification means for identifying a response including an answer to the content of the inquiry based on the attribute information of the content of the inquiry and the search data;
characterized by comprising

In order to achieve the above object, an information processing system according to another aspect of the present invention includes:
An information processing system including one or more information processing devices,
voice recognition means for acquiring voice recognition results obtained by performing voice recognition processing on an utterance for an inquiry;
First attribute information related to an assumed inquiry, second attribute information related to an answer to the assumed inquiry, and expanded attribute information related to the first attribute information and the second attribute information Filter processing means for filtering attribute information included in a speech recognition result by the speech recognition means using a filter included as an element;
response identification means for identifying a response including an answer to the inquiry utterance based on the result of filtering by the filtering means;
characterized by comprising

ADVANTAGE OF THE INVENTION According to this invention, the technique which performs a more suitable automatic response with respect to an inquiry can be provided.

It is a mimetic diagram showing composition of information processing system 1 concerning this embodiment. 3 is a schematic diagram showing a hardware configuration of an information processing device 800 that constitutes the terminal device 10 or the server 20. FIG. 2 is a block diagram showing a functional configuration of the terminal device 10; FIG. 2 is a block diagram showing a functional configuration of server 20. FIG. It is a schematic diagram which shows an example of a response content table. It is a schematic diagram which shows an example of a connector table. FIG. 10 is a schematic diagram showing an example of extraction filter generation; FIG. 10 is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an extraction filter; 1 is a schematic diagram showing a specific functional configuration example of an information processing system 1 according to a first embodiment; FIG. 4 is a flowchart showing the flow of automatic response processing executed by the information processing system 1; 4 is a flow chart showing the flow of extraction filter generation processing executed by the information processing system 1. FIG. It is a schematic diagram which shows the concept of the process in embodiment. It is a schematic diagram which shows an example of the connector table in 2nd Embodiment. FIG. 3 is a schematic diagram showing an example of a refinement table 1; FIG. FIG. 10 is a schematic diagram showing an example of a refinement table 2; FIG. FIG. 3 is a schematic diagram showing an example of a refinement table 3; FIG. FIG. 10 is a schematic diagram showing an example of extraction filter generation; FIG. 10 is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an extraction filter; FIG. 10 is a schematic diagram showing an example of changes in narrow-down tables 1 to 3 when utterances for inquiry are input a plurality of times; 4 is a flowchart showing the flow of automatic search processing executed by the information processing system 1; FIG. 11 is a block diagram showing a functional configuration of a server 20 according to a third embodiment; FIG. FIG. 11 is a schematic diagram showing a specific functional configuration example of an information processing system 1 according to a third embodiment; FIG. 12 is a schematic diagram showing an example of a response content table in the third embodiment; FIG. It is a schematic diagram which shows an example of a 1st connector table. FIG. 11 is a schematic diagram showing an example of a second connector table; FIG. 10 is a schematic diagram showing an example of generating an inquiry extraction filter; FIG. 11 is a schematic diagram showing an example of generating an answer extraction filter; FIG. 10 is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an inquiry extraction filter; FIG. 10 is a schematic diagram showing an example of a process of specifying an answer candidate based on attribute relevance using an answer extraction filter; 4 is a flowchart showing the flow of automatic response processing executed by the information processing system 1;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First embodiment]
The information processing system according to the present embodiment performs speech recognition processing on an utterance for an inquiry, and language processing on the content of the utterance represented by the speech recognition result, thereby specifying a specific process (language Specific processing using processing), attributes expressed by words included in speech recognition results of utterances for inquiries, and attributes based on assumed inquiries and answers prepared for assumed inquiries specific processing (specific processing based on attribute relevance) for identifying answer candidates that match the uttered query based on attribute relevance. Then, the information processing system according to the present embodiment responds to the inquiry based on the answer candidate specified by the specifying process using language processing and the answer candidate specified by the specifying process based on the attribute relevance. Identify the final answer and output the response (dialogue containing the answer) to the query.
Therefore, according to the information processing system according to the present embodiment, it is possible to specify a more appropriate response content while shortening the processing time when performing an automatic response to an utterance.

In addition, in the present invention, the contents of inquiries and answers are implemented specifically for a specific purpose (that is, a specific application target), and as a result, an appropriate combination of inquiries and answers according to this purpose can be prepared. Then, for the prepared inquiries and answers, as will be described later, processing such as attribute classification, attribute expansion, extraction filter generation, etc. can be performed prior to the utterance of the inquiry. . Therefore, when an inquiry is uttered, it is possible to identify the answer at high speed.

It should be noted that, in the present embodiment, the matter specified as the answer to the uttered inquiry (that is, the result of the automatic response) is the person in charge with the ability to answer or the content representing the answer itself (for example, text, video, The person in charge specified as the answer to the spoken inquiry answers the inquiry based on their own knowledge, and the content specified as the answer is the user (the inquiry). provided to the speaker who made the call).
The information processing system according to this embodiment will be specifically described below.

FIG. 1 is a schematic diagram showing the configuration of an information processing system 1 according to this embodiment.
As shown in FIG. 1, an information processing system 1 includes a plurality of terminal devices 10 and a server 20. The plurality of terminal devices 10 and the server 20 are configured to communicate with each other via a network 30. It is
An information processing system 1 shown in FIG. 1 shows an example in which the information processing system according to the present invention is applied to a business system such as a call center that receives inquiries by voice via a network. is performed.

The terminal device 10 is configured by an information processing device such as a smart phone or a PC (Personal Computer). Also, the terminal device 10 is used by a user who makes an inquiry, receives an input of voice representing the contents of an inquiry by the user, and transmits data of the input voice to the server 20 . In addition, the terminal device 10 sends an answer to an inquiry transmitted from the server 20 (a person in charge capable of answering or content representing the answer itself (for example, text, video, audio content, web page, etc.)) to the user. Output for

The server 20 is configured by an information processing device such as a PC or a server computer. The server 20 receives voice data representing the content of the inquiry transmitted from the terminal device 10, executes voice recognition processing, and obtains text data (speech recognition result) representing the content of the inquiry. In this embodiment, the server 20 obtains a plurality of speech recognition results by executing different types of speech recognition processing in parallel. Specifically, the server 20 performs speech recognition processing for the purpose of dictation, which converts the entire input speech data into text data, and word extraction, which extracts specific words contained in the speech data and converts them into text data. , and the speech recognition processing are executed in parallel. However, the server 20 executes one speech recognition process (for example, dictation speech recognition process) and processes one speech recognition result (for example, word extraction process) to obtain a plurality of speech recognition results. You can do it.

The server 20 also identifies answer candidates using language processing based on the obtained speech recognition results. That is, the server 20 extracts the semantic content from the text data obtained by speech recognition of the user's utterance by natural language processing, and extracts the meaning content from the prepared answer (here, the content representing the person in charge of answering or the answer itself). , to identify candidate answers that are highly likely to match the content of the uttered query. Note that, hereinafter, answer candidates specified using language processing are appropriately referred to as “answer candidates by language processing”.

In addition, the server 20 extracts predetermined words from the utterances for inquiries by speech recognition processing, and extracts the attributes represented by the words included in the utterances, the assumed inquiries, and the answers prepared for the assumed inquiries. Identify candidate answers based on their relevance to the attributes of That is, the server 20 acquires synonyms of words representing attributes of the assumed inquiry and answers prepared for the assumed inquiry, thereby By expanding the attributes of the prepared answers, using the expanded attributes (that is, words representing the original attributes and their similar words) as a filter, any of the expanded attributes of the words contained in the utterance for the inquiry It is determined whether or not the Then, the server 20 identifies answer candidates corresponding to the words included in the utterance for the inquiry in the expanded attributes. Note that, hereinafter, answer candidates specified based on attribute relevance are appropriately referred to as “answer candidates based on attribute relevance”.

As described above, in this embodiment, based on the relationship between the attributes represented by the words included in the utterance of the inquiry and the attributes of the assumed inquiry and the answers prepared for the assumed inquiry, Identifies potential answers. Therefore, based only on the relationship between the attributes expressed by the words included in the utterance of the inquiry and the attributes of the assumed inquiry (that is, the match between the content of the uttered inquiry and the content of the assumed inquiry) , there is a high possibility that an answer that matches the uttered inquiry can be identified, compared to the case of identifying a prepared answer.
In addition, in this embodiment, the attributes of the assumed inquiry and the answers prepared for the assumed inquiry are expanded, and the expanded attributes (that is, the words representing the original attributes and their similar words) is used as a filter to determine matches with words included in the utterance for the inquiry. Therefore, it is possible to expand the scope of identification of relevance to the range of content close to the utterance of the inquiry, and the range of content close to the assumed inquiry and the answer prepared for the assumed inquiry. A broad range of intent can be used to identify answers that match the uttered query.

Furthermore, the server 20 executes processing for identifying answers by integrating the identified answer candidates based on the language processing and the answer candidates based on attribute relevance, and identifies the final answer to the uttered query. . In the present embodiment, when an answer candidate by language processing has a certain degree of certainty (for example, a score of a result of identification by language processing is equal to or greater than a threshold value) is specified, an answer candidate by language processing is given priority, and if the accuracy of the answer candidate by language processing is low, the answer candidate based on the relevance of attributes shall be selected. The server 20 then transmits the specified final answer to the terminal device 10 as voice or text data.
As a result, when an inquiry is uttered, it is possible to obtain a more accurate speech recognition result at high speed and specify an appropriate answer.

In addition, in the present embodiment, when attributes are set for the assumed inquiry and the answer prepared for the assumed inquiry, the data of the assumed inquiry and the answer prepared for the assumed inquiry It is possible to set tags or hashtags to
In addition, in order to enable the person in charge with the answering ability to be specified as an answer to the inquiry, the server 20 is provided with a telephone network for making calls with the person in charge with the answering ability. can be connected as appropriate.

[Hardware configuration]
FIG. 2 is a schematic diagram showing a hardware configuration of an information processing device 800 that constitutes the terminal device 10 or the server 20. As shown in FIG.
As shown in FIG. 2, the information processing apparatus 800 includes a CPU (Central Processing Unit) 811, a ROM (Read Only Memory) 812, a RAM (Random Access Memory) 813, a bus 814, an input section 815, an output A unit 816 , a storage unit 817 , a communication unit 818 , a drive 819 and an imaging unit 820 are provided.

The CPU 811 executes various processes according to programs recorded in the ROM 812 or programs loaded from the storage unit 817 to the RAM 813 .
The RAM 813 also stores data necessary for the CPU 811 to execute various types of processing.

The CPU 811 , ROM 812 and RAM 813 are interconnected via a bus 814 . An input unit 815 , an output unit 816 , a storage unit 817 , a communication unit 818 and a drive 819 are connected to the bus 814 .

The input unit 815 receives input of various information to the information processing apparatus 800 . In this embodiment, the input unit 815 includes an operation input unit 815a that receives operation input via a display screen using a mouse, keyboard, or the like, and a voice input unit 815b that receives voice input using a microphone or the like.
The output unit 816 includes a display, a speaker, and the like, and outputs images and sounds.
The storage unit 817 is composed of a hard disk, a DRAM (Dynamic Random Access Memory), or the like, and stores various data managed by each server.
A communication unit 818 controls communication with another device via a network.

A removable medium 831 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is mounted in the drive 819 as appropriate. A program read from the removable medium 831 by the drive 819 is installed in the storage unit 817 as required.
The imaging unit 820 is configured by an imaging device including a lens, an imaging device, and the like, and captures a digital image of a subject.
Note that the hardware configuration described above is the basic configuration of the information processing apparatus 800, and may be configured without a part of the hardware, with additional hardware, or with a different hardware implementation. be able to. For example, the information processing apparatus 800 may be configured without the imaging unit 820, or may include a DSP (Digital Signal Processor) for executing speech recognition processing at high speed. Further, the information processing apparatus 800 can also be configured to include a touch panel by configuring the input unit 815 with a touch sensor and arranging it over the display of the output unit 816 .

[Functional configuration]
Next, a functional configuration of the terminal device 10 will be described.
FIG. 3 is a block diagram showing the functional configuration of the terminal device 10. As shown in FIG.
As shown in FIG. 3, in the CPU 811 of the terminal device 10, a speech reception unit 51, a speech data transmission unit 52, a response reception unit 53, and a response output unit 54 function.

The utterance accepting unit 51 accepts an utterance made by a user who uses the information processing system 1 for an inquiry. The speech received by the speech reception unit 51 is held as voice data.
The utterance data transmission unit 52 transmits to the server 20 the utterance (speech data) for inquiry received by the utterance reception unit 51 .
The response receiving unit 53 receives a response (Web page data, voice data, video data or text data) transmitted from the server 20 in response to the inquiry utterance transmitted by the utterance data transmitting unit 52 .
The response output unit 54 outputs a response to the inquiry received by the response reception unit 53 by voice or screen display.

Next, a functional configuration of the server 20 will be described.
FIG. 4 is a block diagram showing the functional configuration of the server 20. As shown in FIG.
As shown in FIG. 4, in the CPU 811 of the information processing device 800, an utterance data receiving unit 151, a preprocessing unit 152, a speech recognition processing unit 153, a language processing unit 154, a narrowing processing unit 155, an integration processing 156 and , function. Further, the storage unit 817 of the information processing device 800 includes a speech recognition dictionary database (speech recognition dictionary DB) 171, a word dictionary database (word dictionary DB) 172, a response content database (response content DB) 173, and a connector database. (connector DB) 174 is formed.
In this embodiment, the speech recognition engine for executing speech recognition processing is for dictation, which converts the entire input speech data into text data, and for extracting specific words contained in the speech data. It is used for the purpose of voice recognition processing for extracting words and converting them into text data.

The speech recognition dictionary DB 171 stores data of various dictionaries and models (acoustic models, language models, pronunciation dictionaries, etc.) used by each of the plurality of speech recognition engines. For example, the speech recognition dictionary DB 171 includes an acoustic model in which speech feature values and phoneme models are stored in association with each other, a language model in which results of statistical processing of patterns in which character strings or word strings are used as languages are stored, Also stored is pronunciation dictionary data in which words in the language model and phonemes in the acoustic model are stored in association with each other.

The word dictionary DB 172 stores a dictionary for word extraction in which words to be extracted are registered in advance. Each word in the word dictionary DB 172 is given a reading (pronunciation), and when speech recognition processing for word extraction is executed, pre-registered words are extracted based on the reading (pronunciation) information. . As a result, the accuracy and speed of speech recognition can be improved.
The response content DB 173 stores text data representing assumed inquiry content, words included in the assumed inquiry content (words representing attributes of the inquiry content), and answers prepared for the assumed inquiry. Data of a response content table in which identification information (here, the name of an operator who deals with an inquiry or the name of content representing an answer) is associated is stored.

FIG. 5 is a schematic diagram showing an example of a response content table.
In addition, in FIG. 5, the column indicated by "..." means that some data exists or does not exist (hereinafter, the same applies to other drawings). ).
As shown in FIG. 5, the response content table is prepared with sentences (text data) representing assumed inquiry content, attributes (words) of the assumed inquiry content, and assumed inquiries. Information identifying an answer (operator name or content name representing an answer) is stored in association with a number assigned to an answer prepared for an assumed inquiry. For example, in the second row of the response content table in FIG. 5, the sentence "I would like to complete the registration procedure for automatic bank transfer." The words "transfer" and "registration" are stored. Also, "operator A" is stored as information for identifying an answer prepared for this inquiry. "Operator A" is given the number "#1". Further, for example, in the 10th row of the response content table in FIG. 5, the text "I would like to open a new account" is stored as the assumed inquiry content, and the attributes of this inquiry content are "new" and " The words "account" and "open" are stored. In addition, "opening guidance" (movie file name) is stored as information for identifying answers prepared for this inquiry. The number "#101" is assigned to the "opening guidance".

The connector DB 174 stores words representing the attributes of assumed inquiries and the attributes of answers prepared for the assumed inquiries (here, the operator's field of expertise, response record, or contents representing the answers). The data of the connector table associated with the word representing the category to which the product belongs, etc., is stored. In the following description, the attribute of the content of the assumed inquiry will be referred to as "type A attribute", and the attribute of the answer prepared for the assumed inquiry will be referred to as "type B" attribute. In this embodiment, the connector DB 174 stores words (explicit attributes) explicitly included in the assumed inquiry content as attributes of the assumed inquiry content (attributes of type A). Unexplicit words (implicit attributes) such as the meaning of the content itself and the name of the answer are stored. These words are sorted by name and similar word unification, and each is set as a unique word (original word or original attribute). Note that when creating other tables, the same rules as those used when generating the connector table are used when performing name identification and similar word unification. Various data (“response specific attribute table”, “response group specific attribute table”, “expansion attribute table” and “extraction filter” to be described later) generated based on the connector table are stored in one area of the RAM 813 . It is stored in the connector DB 174 as appropriate.

FIG. 6 is a schematic diagram showing an example of a connector table.
As shown in FIG. 6, the connector table contains words (type A attribute ) and a word (attribute of type B) representing the attribute of the answer prepared for the envisioned inquiry are stored in association with each other. For example, in the second row of the connector table shown in FIG. Stores the words "credit card", "debit card", "cash card", "automatic withdrawal", "transfer", etc. as attributes of answers (attributes of type B) prepared for inquiries. It is Note that in the connector table, implicit attributes of type A are stored on the leftmost side of each entry. One row of data in the connector table indicates an original attribute set in advance as an attribute of a set of assumed inquiries and answers prepared for the assumed inquiries.

In this embodiment, "attribute expansion" and "narrowing down" of answers based on the expanded attributes are performed in order to specify an appropriate answer from the content of the inquiry. In order to realize "attribute expansion", the word (response Groups of answers prepared for each inquiry are classified into two groups: one is a word that represents an attribute that can be specified (response group specific attribute), and each word (attribute) is expanded to a similar word, and the necessary A method is used in which all words (attributes) including expanded words (attributes), if any, are expanded to similar words acquired via the network (to generate an "expanded attribute table" to be described later). Appropriate answers are then identified by referencing the attributes augmented by these techniques and filtering the words contained in the query utterances.
Connector table, response specific attribute and response group specific attribute (specifically, "response specific attribute table" and "response group specific attribute table" described later), expanded attribute table, filter for word extraction (specifically, By preparing an "extraction filter" (to be described later) in advance of making an inquiry, it is possible to identify the answer at high speed when the inquiry is uttered. Data used to identify "answers prepared for expected inquiries" in the response content table, such as the response specific attribute table, response group specific attribute table, expanded attribute table, and filter for word extraction. is appropriately referred to as "search data".

The speech data receiving unit 151 receives a speech (speech data) for inquiry transmitted from the terminal device 10 .
The preprocessing unit 152 performs preprocessing such as noise removal on the speech data (utterance for inquiry) received by the speech data receiving unit 151 .

The speech recognition processing unit 153 performs speech recognition processing in parallel on the speech data after the preprocessing has been performed, using a plurality of speech recognition engines. Each speech recognition result obtained by speech recognition processing is output to the language processing unit 154 and the narrowing down processing unit 155 . As shown in FIG. 4, the speech recognition processing unit 153 can include a first speech recognition unit 153-1 to an n-th speech recognition unit 153-n (n is an integer of 2 or more). The 153-1 to n-th speech recognition units 153-n can execute speech recognition processing using different speech recognition engines. In the first speech recognition unit 153-1 to the n-th speech recognition unit 153-n, which speech recognition engine is used to execute speech recognition processing, and the first speech recognition unit 153-1 to the n-th speech recognition The number of units 153-n can be determined according to specific implementation conditions such as the processing purpose and information processing capability of the server 20. FIG. However, in this embodiment, the server 20 performs speech recognition processing for the purpose of dictation, which converts the entire input speech data into text data, and extracts specific words contained in the speech data and converts them into text data. Speech recognition processing for word extraction is executed in parallel.
Note that the speech recognition processing unit 153 may appropriately execute auxiliary processing such as filler removal.

The language processing unit 154 extracts the meaning of the text data of the dictation result acquired by the speech recognition processing unit 153 by language processing, and extracts a prepared answer (here, the person in charge of answering or Among the content representing the answer itself), candidates for the answer (the person in charge of answering or the content representing the answer itself) that is highly likely to match the answer of the uttered content are specified. At this time, the language processing unit 154 scores the suitability (accuracy) of the answer candidates specified from the semantic content as an answer to the inquiry, and selects the one with the highest score as an answer candidate by language processing. can be

The narrowing processing unit 155 determines whether the speech recognition result of word extraction acquired by the speech recognition processing unit 153 (that is, the word extracted from the utterance) matches the word included in the extraction filter, If there is a match, based on the matching words included in the extraction filter, the answers prepared in advance (the person in charge of answering or the content representing the answer itself) are narrowed down. This identifies answer candidates based on attribute relevance. Note that the narrow-down processing unit 155 generates the connector table, the response specific attribute table, the response group specific attribute table, the expansion attribute table, and the extraction filter used in this embodiment prior to input of the user's inquiry utterance. do.

FIG. 7 is a schematic diagram showing an example of extraction filter generation.
For the sake of convenience of explanation, only the columns showing specific data in FIG. the same shall apply).
As shown in FIG. 7, when generating an extraction filter, first, the connector table is referred to, words of response specific attributes are extracted, and table format data (hereinafter referred to as , called the “response specific attribute table”). In addition, the connector table is referred to, words of the response group specific attribute are extracted, and table format data (hereinafter referred to as "response group specific attribute table") in which the number group assigned to the answer is associated with each attribute. ). Furthermore, the words contained in the response specific attribute table and the words contained in the response group specific attribute table are expanded with similar words. For example, similar word group data prepared in advance for the word, word data that can be used as similar words obtained externally via the Internet (for example, similar concepts obtained by referring to an ontology) , etc.). Further, it generates table-format data (hereinafter referred to as an "expanded attribute table") in which the original words contained in the response specific attribute table and the response group specific attribute table are associated with similar words of the original words. Then, an extraction filter composed of word strings included in the extended attribute table is generated.

In addition, when adding a new answer candidate, the word representing the attribute of the answer candidate and the word representing the existing original attribute are grouped and similar words are unified, and the word representing the original attribute is updated. . Then, if the word representing the original attribute is not included in the response specific attribute table, including the word representing the existing original attribute, adding the word representing the attribute and the answer candidate to the response specific attribute table, If it is already included in the response specific attribute table, the word and answer candidates representing that attribute are deleted from the response specific attribute table and added to the "response group specific attribute table".

FIG. 8 is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an extraction filter.
For example, assume that the following utterance examples (1) to (3) are uttered.
Sample utterance (1) “Can I manage my assets online?”
Sample utterance (2) “Can I use my time deposit for a card loan?”
Example utterance (3) "How can I check the balance of my mortgage?"

At this time, it is assumed that words representing attributes are extracted from the utterance of each utterance example as follows.
Utterance example (1) "net" (similar word), "asset management" (original word)
Utterance example (2) “Time deposit” (original word), “Card loan” (original word)
Example utterance (3) “Housing loan” (original word), “Balance” (original word)

In the case of utterance example (1), "net" and "asset management" match the words included in the extraction filter, and the original word "internet" corresponding to the similar word "net" is found in the response specific attribute table. , it is possible to uniquely identify the prepared answer to the envisioned inquiry. That is, in the case of utterance example (1), the operator B (#2) is identified from the response identification attribute table and made an answer candidate based on the attribute relevance.

In the case of utterance example (2), "fixed deposit" and "card loan" match the words included in the extraction filter, and the original words "fixed deposit" and "card loan" are both included in the response group specific attribute table. include. At this time, the answers of #3, #4, and #5 (that is, operators B, C, and D) are specified by "time deposit." In addition, the answers of #2 and #4 (that is, operators B and D) are specified by "card loan". Operator D (#4) is identified by logical AND of the group of identified answers, and is made an answer candidate based on attribute relevance. In this example, the answer is narrowed down to one, but if only words included in the response group specific attribute table are extracted, multiple answers may be candidates. In such a case, it is possible to present a plurality of answer candidates to the user. , the answer may be uniquely selected. Also, when multiple answers are candidates, a response requesting further utterances to uniquely identify the answer may be output.

In the case of example utterance (3), "mortgage" and "balance" match the words included in the extraction filter, and the original words "mortgage" and "balance" are both included in the response group specific attribute table. be At this time, the answers #2 and #4 (that is, operators B and D) are specified by "housing loan". Also, the "balance" specifies the answers of #4 and #5 (that is, operators D and E). Operator D (#4) is identified by logical AND of the group of identified answers, and is made an answer candidate based on attribute relevance.

Returning to FIG. 4 , the integration processing unit 156 executes processing for integrating and identifying answer candidates based on the specified language processing and answer candidates based on attribute relevance, and final answers to the uttered queries. identify. At this time, as described above, the integration processing unit 156 identifies candidates with a certain degree of certainty as answer candidates by language processing (for example, those whose score of the identification result by language processing is equal to or greater than a threshold value). In this case, priority is given to answer candidates based on language processing, and if the accuracy of answer candidates based on language processing is low, answer candidates based on attribute relevance are selected. The integrated processing unit 156 then transmits the specified final answer to the terminal device 10 as voice or text data. If the answer candidates obtained by the language processing do not have a certain level of certainty and the answer candidates based on the relationship of attributes are not specified, the integrated processing unit 156 determines to output a response indicating that the answer cannot be specified. do.
Note that the response content table, connector table, and extraction filter used in the server 20 are prepared in advance and installed in the server 20 before accepting the user's speech (that is, before the automatic response process is executed).

[Specific functional configuration example]
FIG. 9 is a schematic diagram showing a specific functional configuration example of the information processing system 1 in this embodiment.
As shown in FIG. 9, when the speech recognition processing unit 153 performs speech recognition processing such as dictation and word extraction, the speech recognition processing unit 153 includes a first speech recognition unit 153-1 and a second speech recognition unit 153-2. is formed.

The first speech recognition unit 153 - 1 includes a dictation speech recognition unit 501 , a first recognition result identification unit 502 and a first recognition result output unit 503 .
A dictation speech recognition unit 501 processes speech data by a speech recognition engine for dictation, and converts the entire input speech data into text data.
A first recognition result specifying unit 502 uniquely specifies a speech recognition result from the processing result of the dictation speech recognition unit 501 . For example, the first recognition result specifying unit 502 selects the speech recognition result with the highest accuracy from among the speech recognition result candidates acquired by the dictation speech recognition unit 501, based on the score given during the speech recognition process. Specify as a recognition result.
First recognition result output section 503 outputs the speech recognition result specified by first recognition result specifying section 502 to language processing section 154 .

The second speech recognition unit 153 - 2 includes a word extraction speech recognition unit 511 , a second recognition result identification unit 512 and a second recognition result output unit 513 .
A word extraction speech recognition unit 511 performs speech recognition processing for word extraction (here, rule grammar speech recognition processing or speech recognition processing using a DNN (Deep Neural Network)). It processes the data, extracts specific words contained in the audio data, and converts them to text data.
A second recognition result specifying unit 512 uniquely specifies a speech recognition result from the processing result of the word extraction speech recognition unit 511 . For example, the second recognition result specifying unit 512 specifies words registered in the word dictionary DB 172 as voice recognition results from among the same-sounding words with different characters among the voice recognition result candidates acquired by the word extraction voice recognition unit 511. do.
The second recognition result output unit 513 outputs the speech recognition result specified by the second recognition result specifying unit 512 to the narrowing down processing unit 155 .
Note that the speech recognition processing in the first speech recognition unit 153-1 and the second speech recognition unit 153-2 is executed by one dictation speech recognition processing (for example, speech recognition processing by the dictation speech recognition unit 501), and the subsequent processing is performed. may be executed in parallel as described above. In this case, the word dictionary DB 172 is used to extract words from the text data resulting from the dictation speech recognition process.

[motion]
Next, the operation of the information processing system 1 will be described.
[Automatic response processing]
FIG. 10A is a flowchart showing the flow of automatic response processing executed by the information processing system 1. FIG.
The automatic response process is started in response to an operation instructing execution of the automatic response process being performed in the server 20 .
Before the automatic response process is executed (that is, the user's query utterance is input), the connector table, response specific attribute table, response group specific attribute table, expanded attribute table, extraction filter (search data ) is generated in advance (see FIG. 10B).

When the automatic response process is started, in step S1, the speech receiving unit 51 of the terminal device 10 determines whether or not the user's speech has been input.
If the user's utterance is not input, NO is determined in step S1, and the process of step S1 is repeated.
On the other hand, if the user's speech is input, YES is determined in step S1, and the process proceeds to step S2.

In step S<b>2 , the speech data transmission unit 52 of the terminal device 10 transmits the input voice data of the user's speech to the server 20 . The transmitted voice data is received by the speech data receiving section 151 of the server 20 .
In step S3, the preprocessing unit 152 of the server 20 performs preprocessing such as noise removal on the received audio data.

In step S4, the speech recognition processing unit 153 of the server 20 performs parallel speech recognition processing (S41-1 to S43-1 and S41 -2 to S43-2) are executed.
Specifically, in step S41-1, the dictation speech recognition unit 501 of the server 20 processes the speech data using a speech recognition engine for dictation, and converts the entire input speech data into text data.

In step S42-1, the first recognition result specifying unit 502 uniquely specifies a speech recognition result from the processing result of the dictation speech recognition unit 501. FIG.
In step S 43 - 1 , first recognition result output section 503 outputs the speech recognition result specified by first recognition result specifying section 502 to language processing section 154 .
After step S43-1, the process proceeds to step S5.

In step S41-2, the word extraction speech recognition unit 511 performs speech recognition processing for word extraction (here, rule grammar speech recognition processing or speech recognition processing using DNN). The engine processes speech data, extracts specific words from the speech data, and converts them to text data.
In step S42-2, the second recognition result identification unit 512 uniquely identifies the speech recognition result from the processing result of the word extraction speech recognition unit 511. FIG.
In step S 43 - 2 , the second recognition result output unit 513 outputs the speech recognition result specified by the second recognition result specifying unit 512 to the narrowing down processing unit 155 .
After step S43-2, the process proceeds to step S6.

In step S5, the language processing unit 154 extracts the meaning and content of the text data of the dictation result acquired by the speech recognition processing unit 153 by language processing, or content representing the answer itself), candidates for answers (person in charge of answering or content representing the answer itself) that are highly likely to match the uttered content are identified. The result of identifying the answer candidate by the language processing in step S5 is output to the integration processing unit 156. FIG.

In step S6, the narrow-down processing unit 155 determines whether the speech recognition result of word extraction acquired by the speech recognition processing unit 153 (that is, the word extracted from the utterance) matches the word included in the extraction filter. is determined, and if there is a match, prepared answers are narrowed down based on the matched word included in the extraction filter. That is, the narrow-down processing unit 155 identifies words representing attributes of the extended attribute table from the words included in the extraction filter, and further narrows down the words included in the response specific attribute table and the response group specific attribute table from the words representing the specified attributes. Identifies the number of answers received (ie, the number of answers contained in the connector table). Then, based on the identified answer number, the narrow-down processing unit 155 refers to the response content table and narrows down the specific answers. This identifies answer candidates based on attribute relevance. The answer candidate identification result based on the attribute relevance in step S<b>6 is output to the integration processing unit 156 .
After steps S5 and S6, the process proceeds to step S7.

In step S<b>7 , the integration processing unit 156 determines whether or not answer candidates have been identified by language processing (for example, whether or not the score of the identification result by language processing is equal to or greater than a threshold).
If answer candidates have been identified by language processing, YES is determined in step S7, and the process proceeds to step S9.
On the other hand, if the answer candidate cannot be identified by language processing, it is determined as NO in step S7, and the process proceeds to step S8.
In step S8, the integration processing unit 156 executes a process of specifying an answer for the specified answer candidates based on the relevance of attributes, or a response that the answer cannot be specified).

In step S9, the integration processing unit 156 outputs (transmits to the terminal device 10) the response to the inquiry as voice or text data. At this time, if an answer candidate by language processing has been specified as a response to the inquiry, integration processing unit 156 outputs the answer candidate by the specified language processing, and outputs the answer candidate by language processing. If not, output a candidate answer based on attribute relevance or a response that the answer cannot be specified. The response data transmitted in step S9 is received by the response receiving section 53 of the terminal device 10, and is output by the response output section 54 as voice or image display. It should be noted that the response to the inquiry output in step S9 may be a response that a uniquely specified answer or a response that an answer cannot be specified, a response that prompts an utterance for uniquely specifying a plurality of answer candidates, It may be a response that outputs a plurality of answer candidates.

In step S10, the integration processing unit 156 determines whether or not conditions for ending the automatic response process are met. As the conditions for ending the automatic response process, for example, it is necessary to define that a unique response is output to the spoken inquiry, or that the user performs an operation to end the inquiry. can be done.
If the end condition of the inquiry process is not satisfied, NO is determined in step S10, and the process proceeds to step S1.
On the other hand, if the conditions for terminating the automatic response process are met, a determination of YES is made in step S10, and the automatic response process ends.

[Extraction filter generation process]
FIG. 10B is a flow chart showing the flow of extraction filter generation processing executed by the information processing system 1 .
The extraction filter generation process is started in the server 20 in response to an operation instructing execution of the extraction filter generation process. A series of search data is generated by the extraction filter generation process.
When the extraction filter generation process is started, in step S21, the narrowing-down processing unit 155 acquires assumed inquiry contents and attributes of answers prepared for the assumed inquiry. These attributes can be acquired from words included in the content of the assumed inquiry and the answers prepared for the assumed inquiry, or preset attributes.

In step S<b>22 , the narrowing down processing unit 155 collects the names of words representing attributes and unifies similar words.
In step S23, the narrowing-down processing unit 155 selects a word (attribute of type A) representing the attribute of the assumed inquiry content and a word (attribute of type B) representing the attribute of the answer prepared for the assumed inquiry. ) to generate a connector table.

In step S24, the narrowing-down processing unit 155 replaces the original words (original attributes) in the connector table with words (response specific attributes) representing attributes that can uniquely specify the answers prepared for the envisioned inquiry. and a word (response group specific attribute) representing an attribute that can identify a group of answers prepared for an assumed inquiry.

In step S25, the narrowing processing unit 155 generates a response specific attribute table and a response group specific attribute table based on the classified response specific attribute and response group specific attribute.
In step S26, the narrow-down processing unit 155 expands the words included in the response specific attribute table and the words included in the response group specific attribute table to similar terms.

In step S27, the narrow-down processing unit 155 generates an expanded attribute table in which original words included in the response specific attribute table and the response group specific attribute table are associated with similar words of the original words.
In step S28, the narrow-down processing unit 155 generates an extraction filter composed of word strings included in the expanded attribute table.
After step S28, the extraction filter generation process ends.

As described above, in the information processing system 1 according to the present embodiment, the attribute represented by the words included in the utterance of the inquiry and the attribute of the assumed inquiry and the answer prepared for the assumed inquiry. Identify potential answers based on relevance. Also, a connector table, a response specific attribute table, a response group specific attribute table, an expansion attribute table, and an extraction filter (search data) are prepared prior to the inquiry. That is, as attributes of the assumed inquiry content, words (explicit attributes) that are explicitly included in the assumed inquiry content, the meaning of the assumed inquiry content itself, the name of the answer, etc. are not specified. Words (implicit attributes) are used to generate a connector table along with the attributes that the prepared answers to the envisioned queries have. In addition, attributes included in the assumed inquiry and answers prepared for the assumed inquiry are expanded to similar words, and an extraction filter composed of words representing the original attributes and their similar words is generated. Also, a predetermined word is extracted from the utterance for inquiry by voice recognition processing, and it is determined whether or not the word extracted from the utterance matches the word included in the extraction filter. Then, when the word extracted from the utterance matches the word included in the extraction filter, the information processing system 1 provides a prepared answer based on the matched word included in the extraction filter. content representing the person in charge or the answer itself). This identifies answer candidates based on attribute relevance. In addition, the information processing system 1 extracts the meaning content by language processing for the text data of the dictation result obtained by speech recognition processing of the utterance, and prepares a prepared answer (a person in charge of answering or an answer content), candidates for answers (content representing the person in charge of answering or the answer itself) that are highly likely to match the answer of the uttered content are identified. Furthermore, the information processing system 1 executes a process of integrating and identifying the answer candidates based on the identified language processing and the answer candidates based on the attribute relevance, and identifies the final answer to the uttered query. .

Through such processing, in the information processing system 1, in response to an uttered inquiry, not only a specification method based on language processing, but also a specification method based on the matching of words representing attributes is used to obtain an answer to the inquiry. can be specified.
Therefore, when performing an automatic response to an utterance, it is possible to specify a more appropriate response content while shortening the processing time.
In addition, based only on the relationship between the attributes expressed by the words included in the utterance of the inquiry and the attributes of the assumed inquiry (that is, the consistency between the content of the uttered inquiry and the content of the assumed inquiry) , there is a high possibility that an answer that matches the uttered inquiry can be identified, compared to the case of identifying a prepared answer.
In addition, a connector table, a response specific attribute table, a response group specific attribute table, an extended attribute table, and an extraction filter (data for search) used in the process of identifying answers to inquiries are prepared prior to the inquiry. Therefore, when an inquiry is uttered, it is possible to identify the answer at high speed.

In addition, when judging the matching of words representing attributes, expand the words representing attributes based on presumed inquiries and answers prepared for the presumed inquiries to generate an extraction filter, It determines the match with the attribute represented by the word included in the speech recognition result of the utterance for the purpose.
Therefore, even if unexpected expressions are used, such as when expressions fluctuate or words with similar meanings are used, it is possible to easily identify answers that match the uttered inquiry. Become. That is, since it is possible to expand the target of identification of the relevance to the range of contents close to the utterance of the inquiry, and contents close to the assumed inquiry and the answers prepared for the assumed inquiry, A broad range of intent can be used to identify answers that match the uttered query.

[Second embodiment]
Next, a second embodiment of the invention will be described.
In the first embodiment, the item specified as the answer is the person in charge who has the ability to answer or the content representing the answer itself (e.g., text, video, audio content, web page, etc.). . In addition, the attributes possessed by the assumed inquiries and answers prepared for the assumed inquiries are classified into response specific attributes that uniquely identify answers and response group specific attributes that identify answers as a group. , these attributes are associated and converted into data, and by further expanding the attributes, the words being spoken are filtered to specify the answer to the inquiry.

On the other hand, in the present embodiment, in the first embodiment, a specific part of the entire content representing the answer or the answer of the person in charge (for example, the specific part of the text, video, audio content or web page) A part (for example, a paragraph, a section, a time period of a moving image, or a specific part of a web page) is also an object to be specified as an answer. In addition, in the present embodiment, a plurality of processes having different functions (hereinafter referred to as "narrowing down process" as appropriate) are executed in parallel as the specific process based on the relevance of attributes, and these parallel processes The identification result and the identification result obtained by the identification processing using the language processing are integrated to identify the content of the response (the content of the dialogue that induces the utterance of the attribute effective for identifying the answer or the answer). For example, as one narrowing process, a narrowing process 1 is executed for narrowing down answers based on words included in the latest utterances for all words representing attributes included in the original connector table. In narrowing down process 1, the result of narrowing down all the words representing attributes contained in the original connector table by the latest single utterance is output. In addition, as another narrowing process, the result of narrowing down the words representing attributes included in the original connector table by a predetermined number of past utterances is targeted, and narrowing is performed by the words contained in the latest utterances. Narrowing down process 2 is executed. In the narrowing process 2, the result of narrowing down the words representing the attribute contained in the original connector table by the utterances of a predetermined number of times in the past is output. In addition, as still another narrowing process, the attribute contained in the original connector table is changed by a predetermined number of past utterances (for example, all past utterances after the inquiry was made) different from the narrowing process 2. Targeting the result of narrowing down the words to be expressed, narrowing down processing 3 is executed for narrowing down by words included in the latest utterance. In the narrowing process 3, the result of narrowing down the words representing the attributes contained in the original connector table by a predetermined number of past utterances different from those in the narrowing process 2 (for example, all past utterances after the inquiry was made). is output. Then, the identification results of the narrowing down processes 1 to 3 and the identification results of the identification process using language processing are integrated by selecting based on preset selection conditions, and the content of the response to the inquiry (answer or answer) The content of dialogue that induces the utterance of a particularly effective attribute) is specified.

In this embodiment, narrowing down processes 1 to 3 are used together with specific processing using language processing, and narrowing down processes 1 to 3 are configured to autonomously interact (automatically interact) with the user. Therefore, the possibility of identifying the appropriate answer is higher.
Main parts such as the system configuration of the information processing system 1, the hardware configuration of the information processing device, and the functional configurations of the terminal device 10 and the server 20 in this embodiment are the same as those in the first embodiment. Mainly explained.

FIG. 11 is a schematic diagram showing the concept of processing in this embodiment.
As shown in FIG. 11, in the information processing system 1 according to the present embodiment, the identification results obtained by the narrowing down processes 1 to 3 executed in parallel and the identification results obtained by the identification processing using language processing are integrated to obtain a response. Identify content. At this time, in narrowing down processes 1 to 3, the original connector table, the response specific attribute table generated from the original connector table, and the response group specific attribute table (hereinafter referred to as "connector tables") are used as starting points for narrowing down. Narrowing tables 1 that are the result of narrowing down connector tables by process 1, narrowing down tables 2 that are results of narrowing down tables 1 by narrowing process 2, and narrowing down tables 2 by narrowing process 3. Answer candidates based on attribute relevance are specified by generating a narrowing down table class 3 as a result of the inclusion. However, with regard to the narrowing down table type 3, when a series of inquiries by the user reaches four times or more, the output result of the narrowing down table type 3 becomes the input of the narrowing down processing 3 for the next inquiry, and the narrowing down by the narrowing down processing 3 is performed recursively. is done. The narrow-down tables 1-3 include narrow-down tables 1-3, which are the results of the narrow-down processes 1-3, respectively. Narrowing tables 1 to 3 are tabular data indicating which answers (specifically, answer numbers) the words narrowed down by the narrowing processes 1 to 3 are related to. In narrow-down tables 1-3, for example, for each of the words narrowed down by narrow-down processes 1-3, "1" having relevance to the answer number and "0" having no relevance are stored. In addition, if the answer candidate is not uniquely specified, the attributes included in the narrowing down tables 1 to 3 are updated to attributes that can narrow down the answers more than the current ones, and then accepted as the input for the next narrowing down process. Passed (input replaced).

In FIG. 11, the response specific attribute table included in the connector tables is "response specific attribute table ATS0", the response group specific attribute table is "response group specific attribute table ATG0", and the response specific attribute The table is "response specific attribute table ATS1", the response group specific attribute table is "response group specific attribute table ATG1", the response specific attribute table included in narrowing table class 2 is "response specific attribute table ATS2", and the response group specific attribute table is called "response group specific attribute table ATG2", the response group specific attribute table included in narrowing table class 3 is called "response specific attribute table ATS3", and the response group specific attribute table is called "response group specific attribute table ATG3". In the following description, the same notation will be used as appropriate.

[Functional configuration]
In this embodiment, the functional configuration of the terminal device 10 is the same as the functional configuration of the first embodiment shown in FIG.
Also, the functional configuration of the server 20 differs from the functional configuration of the first embodiment shown in FIG. .
The connector DB 174 stores words representing the attributes of assumed inquiries and the attributes of answers prepared for the assumed inquiries (here, the answers of persons in charge who have the ability to answer, or the answers The data of the connector table in which the entire content representing the answer and the word representing the attribute of the part of the content representing the answer are associated with each other are stored.

FIG. 12A is a schematic diagram showing an example of a connector table in this embodiment.
As shown in FIG. 12A, in the connector table, as in the first embodiment, for each set of assumed inquiries and answers prepared for the assumed inquiries, attributes of assumed inquiry contents (attribute of type A) and a word (attribute of type B) representing the attribute of an answer prepared for an assumed inquiry are associated and stored.

In this embodiment, the connector DB 174 stores words (explicit attributes) explicitly included in the assumed inquiry content as attributes of the assumed inquiry content (attributes of type A). Unexplicit words (implicit attributes) such as the meaning of the content itself and the name of the answer are stored. These words are sorted by name and similar word unification, and each of them is treated as a unique word (original word, original attribute).

Further, in the present embodiment, as attributes of the content representing the answer (attribute of type B), a class to which the entire content representing the answer belongs and a class to which a part constituting the content representing the answer belongs are set. For wholes and parts, words representing attributes are stored in the connector table. The part that constitutes the content representing the answer can be arbitrarily selected from the entire content representing the answer, for which attributes can be set separately. It can be a segmented portion.
In FIG. 12A, "response address" indicates the location of data in the network, such as the URL (Uniform Resource Locator) of the content representing the response or the address in the server.

In this embodiment, as in the first embodiment, the connector table is referred to, and in order to specify an appropriate answer from the contents of the inquiry, "attribute expansion" and an answer " Narrowing down is performed. In order to realize "attribute expansion", (1) in the connector table, in addition to the words (explicit attributes) explicitly included in the assumed inquiry content, the assumed inquiry A method that stores unspecified words (implicit attributes) such as the meaning of the content itself, the name of the answer, etc., and when generating the response specific attribute table from the connector table, (2) express attributes (3) Expanding a word representing an attribute to a similar word acquired via a network is used. Then, refer to the attributes expanded by these methods, filter the words included in the utterance for the inquiry, and if necessary, the information processing system 1 identifies the original attributes effective for narrowing down. Thus, an appropriate answer can be specified while performing a dialogue that induces utterance.

The narrowing-down processing unit 155 acquires synonyms of attributes included in an assumed inquiry and answers prepared for the assumed inquiry. For example, the narrowing-down processing unit 155 uses similar word group data prepared in advance for words representing attributes, or word data that can be used as similar words externally acquired via the Internet (for example, an ontology). data of words representing similar concepts acquired by referring to them, etc.).

Then, the narrow-down processing unit 155 uses words (original attributes stored in the connector table) that represent the attributes (attributes of type A and type B) of the assumed inquiry and the answers prepared for the assumed inquiry. word) and words acquired as similar words (similar words of the original attribute), and extract the relevance between the assumed inquiry and the answers prepared for the assumed inquiry Generate an extraction filter for
At this time, the narrow-down processing unit 155 generates the response specific attribute table ATS0 and the response group specific attribute table ATG0 from the attributes in which the words included in the connector table are expanded to generate the extraction filter. The response specific attribute table ATS0 and the response group specific attribute table ATG0 constitute connector tables together with the connector table.

In addition, in narrowing down processing 1, the narrowing down processing unit 155 narrows down the words representing attributes contained in the connector tables based on the words contained in the latest utterances. A refinement table 1 is generated whose elements are information indicating relevance (for example, "1" indicates relevance and "0" indicates no relevance). Specifically, the narrow-down processing unit 155 refers to the connector table, and generates the narrow-down table 1 by words related to any of the answer candidates related to the word included in the latest utterance. Note that the connector table has significance as a narrowing table before being narrowed down by utterance.

FIG. 12B is a schematic diagram showing an example of the refinement table 1. As shown in FIG.
In FIG. 12B , narrowing down table 1 includes, among the words representing attributes contained in the connector table, words related to any of the answer candidates related to the word contained in the latest utterance, and Information indicating whether or not each number is related (“1” indicating related or “0” indicating not related) is stored in association with each other.
For example, in FIG. 12B, "1" is stored for the word "meeting minutes product" representing the attribute in the refinement table 1, indicating that it is related to the answers #1, #2, and #X. As for the word "XX system" representing another attribute, "0" is stored to indicate that it has no relevance to the answer #1.

Since the numbers of the answer candidates stored in the refinement table 1 are narrowed down to those related to the words included in the latest utterance, the refinement table 1 contains the answer candidates narrowed down by the latest utterance. It becomes information indicating Further, since the narrowing-down table 1 also stores words representing attributes associated with answer candidates, it contains information necessary for narrowing down the answer candidates by the next utterance. there is
At this time, the narrow-down processing unit 155 generates a response specific attribute table ATS1 and a response group specific attribute table ATG1 from the attributes included in the narrow-down table 1. FIG. These response specific attribute table ATS1 and response group specific attribute table ATG1 together with the narrow-down table 1 constitute narrow-down table class 1. FIG.

Based on the narrowing down table 1 generated in the narrowing down process 1, the narrowing down processing unit 155 determines whether or not an answer candidate can be uniquely identified by the words included in the latest utterance. If a candidate answer can be uniquely identified, the candidate answer is used to determine candidate answers based on attribute relevance.

Further, based on the narrowing down table 1 generated in the narrowing down process 1, the narrowing down processing unit 155 can further narrow down the answers from the user when the answer candidates cannot be uniquely identified by the words included in the latest utterance. Attributes (that is, attributes effective for specifying answers) are uttered in a dialogue. Specifically, the narrow-down processing unit 155 selects an attribute (that is, an attribute that can uniquely identify the (or an attribute that reduces the number of answer candidates compared to the current one) as an attribute to be presented to the user as information for narrowing down. In addition, the narrow-down processing unit 155 updates the attributes included in the narrow-down table 1 to those that can narrow down the answers more than the current ones. That is, the narrow-down processing unit 155 excludes words (that is, attributes that are not effective for specifying an answer) that are related to all answer candidates to which the word included in the latest utterance is related in the narrow-down table 1, The refinement table 1 is updated by leaving the words (that is, the attribute effective for specifying the answer) related to some of the answer candidates related to the words contained in the utterance of . Specifically, the narrow-down processing unit 155 identifies a group of answers that are related to the word included in the latest utterance in the narrow-down table 1, and all elements of the table corresponding to the identified answer group are “ The narrowing down table 1 is updated by deleting the attribute of the column that is 1” and leaving the attribute of the column that has at least one “1” in the elements of the other attributes. In addition, along with the update of the narrow-down table 1, the response specific attribute table ATS1 and the response group specific attribute table ATG1 of the narrow-down table class 1 are also updated, and the updated narrow-down table class 1 is passed as an input of the narrow-down process 2. .
Through such processing, the user can be urged to utter an attribute that is effective for narrowing down the answers from the system.

Further, in the narrow-down processing 2, the narrow-down processing unit 155 receives the narrow-down table 1 as an input, narrows down the words representing the attributes contained in the updated narrow-down table 1 based on the words contained in the latest utterance, and narrows down the words. A refinement table 2 is generated whose elements are information indicating the relevance between the words representing the attributes obtained and the answer numbers (for example, "1" indicates relevance, and "0" indicates no relevance). Specifically, the narrow-down processing unit 155 refers to the narrow-down table 1, and generates a narrow-down table 2 based on words related to any of the answer candidates related to the word included in the latest utterance.

FIG. 12C is a schematic diagram showing an example of the refinement table 2. As shown in FIG.
In FIG. 12C , refinement table 2 includes, among the words representing attributes contained in refinement table 1, words that are related to any of the answer candidates related to the word included in the latest utterance, and , and information indicating whether or not there is a relationship ("1" indicating that there is a relationship or "0" indicating that there is no relationship) are stored in association with each other.
For example, in FIG. 12C, "1" is stored for the word "XX system" representing the attribute in the refinement table 2, indicating that it is related to the answers #2 and #X. For the word "remote conference" representing another attribute, "0" is stored to indicate that it has no relevance to the answer #2.

Since the numbers of the answer candidates stored in the refinement table 2 are narrowed down to those related to the words included in the latest utterance and the previous utterance, the refinement table 2 includes the latest utterance and the previous utterance. This is information indicating answer candidates narrowed down by In addition, since the narrow-down table 2 also stores words representing attributes associated with answer candidates, it contains information necessary for narrowing down the answer candidates by the next utterance. there is
At this time, the narrow-down processing unit 155 generates a response specific attribute table ATS2 and a response group specific attribute table ATG2 from the attributes included in the narrow-down table 2. FIG. These response specific attribute table ATS2 and response group specific attribute table ATG2 together with the narrow-down table 2 constitute narrow-down tables 2 .

In addition, based on the narrowing down table 2 generated in the narrowing down process 2, the narrowing down processing unit 155 determines whether or not the answer candidate can be uniquely identified by the words included in the latest utterance. If a candidate answer can be uniquely identified, the candidate answer is used to determine candidate answers based on attribute relevance.

Further, based on the narrowing-down processing unit 155, based on the narrowing-down table 2 generated in the narrowing-down processing 2, if the answer candidates cannot be uniquely specified by the words included in the latest utterance, the user can further narrow down the answers. Attributes (that is, attributes effective for specifying answers) are uttered in a dialogue. Specifically, the narrow-down processing unit 155 selects an attribute (that is, an attribute that can uniquely identify the (or an attribute that reduces the number of answer candidates compared to the current one) as an attribute to be presented to the user as information for narrowing down. In addition, the narrow-down processing unit 155 updates the attributes included in the narrow-down table 2 to those that can narrow down the answers more than the current ones. That is, the narrow-down processing unit 155 excludes words (that is, attributes that are not effective for specifying an answer) that are related to all of the answer candidates to which the word included in the latest utterance is related in the narrow-down table 2, The refinement table 2 is updated by leaving words (that is, attributes effective for specifying answers) related to some of the answer candidates related to the words included in the utterance of . Specifically, the narrow-down processing unit 155 identifies a group of answers that are related to the word included in the latest utterance in the narrow-down table 2, and all elements of the table corresponding to the identified answer group are “ The narrowing down table 2 is updated by deleting the attribute of the column that is 1” and leaving the attribute of the column that has at least one “1” in the elements of the other attributes. At this time, the refinement table 2 is updated by excluding words included in the latest utterance. Further, in accordance with the update of the narrow-down table 2, the response specific attribute table ATS2 and the response group specific attribute table ATG2 of the narrow-down table 2 are also updated, and the updated narrow-down table 2 is passed as an input of the narrow-down process 3. .
Through such processing, the user can be urged to utter an attribute that is effective for narrowing down the answers from the system.

Further, in the narrow-down processing 3, the narrow-down processing unit 155 uses the narrow-down table class 2 as an input from the start of the inquiry to the third utterance, and uses the word representing the attribute contained in the updated narrow-down table 2 as the latest utterance. Based on the words contained in the element A refinement table 3 is generated. Specifically, the narrow-down processing unit 155 refers to the narrow-down table 2, and generates a narrow-down table 3 based on words related to any of the answer candidates related to the word included in the latest utterance.

Further, in the narrow-down processing 3, the narrow-down processing unit 155 receives the narrow-down table 3 updated in the previous narrow-down processing 3 after the fourth utterance from the start of the inquiry, and is included in the updated narrow-down table 3. The word representing the attribute is narrowed down based on the words included in the latest utterance, and the information indicating the relationship between the narrowed word representing the attribute and the answer number (for example, "1" is related, " 0” is not relevant) is generated. Specifically, the narrow-down processing unit 155 refers to the narrow-down table 2, and generates a narrow-down table 3 based on words related to any of the answer candidates related to the word included in the latest utterance.

FIG. 12D is a schematic diagram showing an example of the refinement table 3. As shown in FIG.
In FIG. 12D, in narrow-down table 3, words representing attributes included in narrow-down table 2 or narrow-down table 3 are associated with some of the answer candidates related to words included in the latest utterance. A word representing an attribute that is the result of narrowing down to words, and information indicating whether each word has a relationship with each answer number ("1" indicating that there is a relationship, or "0" indicating ) are stored in association with each other.
For example, in FIG. 12D, the word “remote meeting” representing the attribute in the refinement table 3 stores “1” indicating that it is related to the answer #X.

Since the numbers of candidate answers stored in the refinement table 3 are narrowed down to those related to words included from the start of the inquiry to the latest utterance, the refinement table 3 contains the most recent utterances from the start of the inquiry. It becomes information indicating answer candidates narrowed down by the utterance. Further, since the narrowing-down table 3 also stores words representing attributes associated with answer candidates, it contains information necessary for narrowing down the answer candidates by the next utterance. there is
At this time, the narrow-down processing unit 155 generates a response specific attribute table ATS3 and a response group specific attribute table ATG3 from the attributes included in the narrow-down table 3. FIG. These response specific attribute table ATS3 and response group specific attribute table ATG3 together with the narrow-down table 3 constitute narrow-down tables 3 .

In addition, based on the narrowing down table 3 generated in the narrowing down process 3, the narrowing down processing unit 155 determines whether or not an answer candidate can be uniquely identified by the words included in the latest utterance. If a candidate answer can be uniquely identified, the candidate answer is used to determine candidate answers based on attribute relevance.

Further, based on the narrowing down table 3 generated in the narrowing down process 3, the narrowing down processing unit 155 can further narrow down the answers from the user when the answer candidates cannot be uniquely specified by the words included in the latest utterance. Attributes (that is, attributes effective for specifying answers) are uttered in a dialogue. Specifically, the narrow-down processing unit 155 selects an attribute (that is, an attribute that can uniquely identify the (or an attribute that reduces the number of answer candidates compared to the current one) as an attribute to be presented to the user as information for narrowing down. In addition, the narrow-down processing unit 155 updates the attributes included in the narrow-down table 3 so that the answers can be narrowed down more than the current ones. That is, the narrow-down processing unit 155 excludes words (that is, attributes that are not effective for specifying the The refinement table 3 is updated by leaving the words (that is, attributes effective for specifying the answers) related to some of the answer candidates related to the words contained in the utterance of . Specifically, the narrow-down processing unit 155 identifies a group of answers that are related to the word included in the latest utterance in the narrow-down table 3, and all elements of the table corresponding to the identified answer group are “ The refinement table 3 is updated by deleting the attribute of the column that is "1" and leaving the attribute of the column that has at least one "1" in the elements of the other attributes. At this time, the narrow-down table 3 is updated by excluding words included in the latest utterance. Along with the update of the narrow-down table 3, the response specific attribute table ATS3 and the response group specific attribute table ATG3 of the narrow-down table 3 are also updated.
Through such processing, the user can be urged to utter an attribute that is effective for narrowing down the answers from the system.

The narrow-down processing unit 155 executes narrow-down processing 1, narrow-down processing 2, and narrow-down processing 3 in parallel, and sequentially generates and updates narrow-down tables 1-3 by the narrow-down processing 1-3. Then, the narrow-down processing unit 155 outputs to the integration processing unit 156 the identification results of the answer candidates by the narrow-down processes 1 to 3. FIG. Note that the narrowing process 1 is executed from the first utterance in the inquiry, and the narrowing process 2 is executed from the second utterance in the inquiry (utterance for the first dialogue performed to narrow down the answers from the system), and the narrowing process 3 is executed from the third utterance in the inquiry (the utterance for the second dialogue performed to narrow down the answers from the system).

Here, in the narrowing processes 1 to 3, regarding the generation (or updating) of the narrowing down tables 1 to 3 and the replacement (passing) of the narrowing down tables that are the inputs of the narrowing processes 2 and 3, it is possible to narrow down the answers for each utterance. Although it is possible to take various possible forms, as an example, the following procedure is defined in this embodiment.
(1) Narrowing table 1 is generated sequentially by narrowing down the words included in the connector table by narrowing process 1 when words are extracted by the extraction filter in the latest utterance. If the answer is not uniquely identified and Narrowing Table 1 is updated, the input of Narrowing Process 2 is replaced with the updated Narrowing Table 1 .
(2) The refinement table 1 is not updated if no word is extracted by the extraction filter in the latest utterance.

(3) The narrow-down table 2 is not generated (the narrow-down process 2 is not executed) for the utterance immediately after initialization (immediately after the narrow-down tables 1 to 3 are reset).
(4) For utterances other than those immediately after being initialized, the refinement table 2 has attributes that match the words extracted from the latest utterance in the refinement table class 1, which is the output of the refinement processing 1 for the previous utterance. If there is, it is generated by narrowing down the words contained in the narrowing down table 1 by the narrowing down process 2 . If the answer is not uniquely identified and Narrowing Table 2 is updated, the input of Narrowing Process 3 is replaced with the updated Narrowing Table 2 .
(5) For utterances other than those immediately after being initialized, the refinement table 2 has attributes that match the words extracted from the latest utterance in the refinement table class 1, which is the output of the refinement processing 1 for the previous utterance. If not, it will not be generated.

(6) The narrow-down table 3 is not generated (the narrow-down process 3 is not executed) for the utterance immediately after being initialized (after the narrow-down tables 1 to 3 are reset).
(7) The narrow-down table 3 is not generated (the narrow-down process 3 is not executed) for the second utterance after being initialized (after the narrow-down tables 1 to 3 are reset).
(8) The narrow-down table 3 is the latest utterance in the narrow-down table 2, which is the output of the narrow-down processing 2 for the previous utterance, for utterances other than the second utterance immediately after initialization and after initialization. If there is an attribute that matches the extracted word, it is generated by narrowing down the words contained in the narrowing down table 2 by narrowing down processing 2 . When the answer is not uniquely specified and the narrowing-down table 3 is updated, the input of the narrowing-down processing 3 is replaced with the updated narrowing-down table 3 when the utterance for the inquiry is continued.
(9) The narrow-down table 3 is the output of the narrow-down processing 2 for the previous utterance or the narrow-down table 2 for the previous utterance for utterances other than the second utterance immediately after being initialized or after being initialized. If there is no attribute that matches the word extracted from the latest utterance in the refinement table class 3, which is the output of process 3, it is not generated.

FIG. 13 is a schematic diagram showing an example of extraction filter generation.
For convenience of explanation, only the columns showing specific data in FIG. the same shall apply).
In addition, in the example shown in FIG. 13, for convenience of explanation, the attributes of the entire content representing the answer are shown, but the attributes of the part of the content representing the answer can be handled in the same way.

As shown in FIG. 13, when an extraction filter is generated, first, the narrowing down processing unit 155 refers to the connector DB 174, and selects an attribute that can uniquely identify the prepared answer to the assumed inquiry. The word representing the answer is extracted, and table format data (response specific attribute table ATS0) associated with the number assigned to each answer is generated. In addition, the narrow-down processing unit 155 refers to the connector DB 174, extracts a word representing an attribute that can specify a group of answers (whole or partial) prepared for an assumed inquiry, table format data (response group specific attribute table ATG0) in which the group of numbers assigned to are associated with each other. Further, the narrow-down processing unit 155 acquires similar terms to the words included in the response specific attribute table ATS0 and the words included in the response group specific attribute table ATG0, and obtains the words included in the response specific attribute table ATS0 and the response group specific attribute table ATG0. Generate table format data (extended attribute table) in which the original word is associated with similar words of the word. Then, the narrow-down processing unit 155 generates an extraction filter composed of word strings included in the expanded attribute table.

FIG. 14A is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an extraction filter.
Also, FIG. 14B is a schematic diagram showing an example of changes in the refinement tables 1 to 3 when an inquiry utterance is input a plurality of times.
FIG. 14A shows an example in which answer candidates are identified from words extracted by the extraction filter via the response specific attribute table ATS0 and the response group specific attribute table ATG0 included in the connector tables. However, even in narrow-down tables 1-3, answer candidates are similarly specified via response specific attribute tables ATS1-3 and response group specific attribute tables ATG1-3.
For example, assume that the following utterance example (1) is uttered as an utterance for starting an inquiry.
Example utterance (1) “I want to know about the minutes product.”

At this time, it is assumed that the following words representing attributes are extracted from the utterance of the utterance example (1).
Extracted word: "minutes product" (original word)

In the case of utterance example (1), "minutes product" matches the words included in the extraction filter, and the original word "minutes product" is included in the response group specific attribute table ATG0. At this time, the responses #1, #2, and #X are specified by the “meeting minutes product”, and the narrowing-down table 1 is generated by the narrowing-down process 1 (see “result of utterance example (1)” in FIG. 14B). Specifically, in the specified answer groups #1, #2, and #X, attributes associated with each answer are obtained. Then, a refinement table 1, a response specific attribute table ATS1, and a response group specific attribute table ATG1 are generated, each element of which is information representing the relationship between a word representing the acquired attribute and the number of each answer.

In this case, since the answer cannot be uniquely identified, the narrowing down table 1 is referred to, and among the attributes containing "1" in the column elements of the narrowing down table 1, the attributes that can narrow down the answers more than the current one ( That is, an attribute that can uniquely identify an answer or an attribute that reduces the number of answer candidates compared to the current one) is adopted as an attribute to be presented to the user as information for narrowing down. In addition, in the narrowing down table 1, words that are related to all answer candidates related to the word contained in the latest utterance (that is, attributes that are not effective for specifying the answer) are excluded, and the word contained in the latest utterance Refinement table 1 is updated, leaving words associated with some of the candidate answers (ie, attributes effective in identifying answers). As a result, a dialogue is executed that induces the user to utter an attribute that is effective for specifying the answer.
Specifically, referring to the narrowing down table 1, the answers #1, #2, and #X will narrow down the answers more than now if either "XX system" or "remote conference" is uttered. Therefore, if you tell the user "'XX system' and 'remote conference', you can answer. ” is output.

When the user speaks in response to the information processing system 1 conducting a dialogue that guides an utterance of an attribute effective for specifying an answer, narrowing down processes 1 and 2 are performed based on the words included in this utterance. Executed, narrowing down tables 1 and 2 are generated. As a result, the answer candidates are further narrowed down.
For example, assume that after the information processing system 1 conducts a dialogue that induces an utterance of an attribute effective for specifying an answer, the user utters an utterance such as the following additional utterance example as the second utterance.
Additional utterance example (1) "I'm interested in the XX system"

In the additional utterance example (1), "XX system" matches the words included in the extraction filter.
For the additional utterance example (1), in narrowing down process 1, the same process as in the case of the above-described "meeting minutes product" is executed.
On the other hand, in narrowing process 2, the original word "XX system" is included in the response group specific attribute table ATG1. At this time, the answers #2 and #X are specified by the “minutes product”, and the narrowing down table 2 is generated by the narrowing process 2 (see “result of additional utterance example (1)” in FIG. 14B). Specifically, in the identified answer groups #2 and #X, attributes associated with each answer are obtained. Then, a refinement table 2, a response specific attribute table ATS2, and a response group specific attribute table ATG2 are generated, each element of which is information representing the relationship between the words representing the acquired attributes and the number of each answer.

In this case, since the answer cannot be uniquely specified, the narrowing down table 2 is referred to, and among the attributes containing "1" in the column elements of the narrowing down table 2, the attributes that can narrow down the answers more than the current one ( That is, an attribute that can uniquely identify an answer or an attribute that reduces the number of answer candidates compared to the current one) is adopted as an attribute to be presented to the user as information for narrowing down. In addition, in the narrowing down table 2, words that are related to all of the answer candidates related to the word included in the latest utterance (that is, attributes that are not effective for specifying the answer) are excluded, and the word included in the latest utterance Refinement table 2 is updated, leaving words associated with some of the candidate answers (ie, attributes effective in identifying answers).
As a result, a dialogue is executed to guide the user to utter an attribute that is effective in specifying the answer.
Specifically, referring to the narrowing down table 2, the answers #2 and #X can be uniquely narrowed down if “remote meeting” is uttered. If you talk about "meeting", I can answer. ” is output.

When the user speaks in response to the information processing system 1 conducting a dialogue that induces an utterance of an attribute effective for specifying an answer, narrowing down processes 1 to 3 are performed based on the words included in this utterance. Executed to generate refinement tables 1-3. As a result, the answer candidates are uniquely narrowed down in the narrowing down table 3 .
For example, assume that after the information processing system 1 conducts a dialogue that induces an utterance of an attribute effective for specifying an answer, the user utters an utterance such as the following additional utterance example as the third utterance.
Additional utterance example (2) “What is a remote conference?”

In the additional utterance example (2), "remote meeting" matches the words included in the extraction filter.
For the additional utterance example (2), in narrowing down processes 1 and 2, the same processes as in the above-described "minutes product" and "XX system" are executed.
On the other hand, in narrowing process 3, the original word "remote conference" is included in the response group specific attribute table ATG2. At this time, the answer #X is specified by "remote conference", and narrowing down table 3 is generated by narrowing down process 3 (see "result of additional utterance example (2)" in FIG. 14B). As a result, in the example shown here, the answer candidates are uniquely narrowed down in the narrowing-down table 3 .

The integration processing unit 156 executes a process of integrating and specifying the answer candidates by the specified language processing and the results of the narrowing processes 1 to 3, and the contents of the response to the uttered inquiry (the final answer or the specification of the answer The content of the dialogue that induces the utterance of the attribute effective for) Identify the final answer. In this embodiment, the integration processing unit 156 identifies the final answer based on the following selection criteria.

That is, the integration processing unit 156
(i) When candidates with a certain level of accuracy are specified as answer candidates by language processing (for example, those whose score of the identification result by language processing is above a threshold), the answer candidates by language processing are finally selected. selected as a valid answer.
(ii) If an answer candidate with a certain degree of accuracy is not identified as an answer candidate by language processing, narrow-down table 1, narrow-down table 2, or narrow-down table 3 is selected as an answer candidate based on attribute relevance. If the answer is narrowed down to one answer, select the narrowed down answer as the final answer.
(iii) If the answer candidate based on the attribute relevance is other than the result of (ii), the answer candidate by language processing is included in the answer candidates in the narrowing table 1, narrowing table 2, or narrowing table 3. If so, the answer candidate by language processing is selected as the final answer.
(iv) If the answer candidate based on the relevance of attributes is other than the result of (iii), narrowing down tables that had a narrowing effect on the latest utterance (that is, narrowing down processing that narrowed down the answer 1 to 3) to generate a message that prompts the next user to speak, including an attribute that has the effect of narrowing down (an attribute that reduces the number of answer candidates compared to the current one), and uses it as a response. However, if there are multiple narrowing-down tables that have had a narrowing-down effect on the latest utterance, the response group specific attribute table that is the output of the narrowing-down processing with a larger number is referred to.

[motion]
Next, the operation of the information processing system 1 will be described.
[Automatic search process]
FIG. 15 is a flow chart showing the flow of automatic search processing executed by the information processing system 1. As shown in FIG.
The automatic search process is started in the server 20 in response to an operation instructing execution of the automatic search process.
In the flowchart shown in FIG. 15, the processes of steps S1 to S5 are the same as the automatic response process of the first embodiment shown in FIG. 10A.
In this embodiment, the connector table, response specific attribute table ATS0, and response group specific attribute table ATG0 (connector tables) used in the narrowing process 1 are set prior to execution of the automatic search process.

In step S6A, the narrow-down processing unit 155 executes narrow-down processing 1 to 3 in parallel, and outputs each identification result to the integration processing unit 156. FIG.
In step S7A, the integration processing unit 156 determines whether or not answer candidates have been identified by language processing (whether or not answer candidates with a certain degree of certainty have been identified).
If an answer candidate can be identified by language processing, a determination of YES is made in step S7A, and the process proceeds to step S13A.
On the other hand, if the answer candidate cannot be identified by the language processing, it is determined as NO in step S7A, and the process proceeds to step S8A.

In step S8A, the integration processing unit 156 identifies answer candidates according to preset selection criteria based on the results of narrowing down processes 1-3.
In step S9A, the integration processing unit 156 determines whether or not the answer candidates have been uniquely identified by the narrowing down processes 1-3.
If an answer candidate can be uniquely specified by narrowing down processes 1 to 3, YES is determined in step S9A, and the process proceeds to step S13A.
On the other hand, if the answer candidate cannot be uniquely specified by the narrowing down processes 1 to 3, NO is determined in step S9A, and the process proceeds to step S10A.

In step S10A, the narrow-down processing unit 155 updates the narrow-down tables 1-3. As a result, the attributes included in the narrowing down tables 1 to 3 are updated so that the answers can be further narrowed down. If the narrow-down tables 1 to 3 have not been narrowed down by the words included in the latest utterance (answer candidates have not decreased from the current number), the narrow-down tables to which the narrow-down table belongs are not updated (step The processing of S10A is skipped).
In step S11A, the narrowing-down processing unit 155 determines the content of dialogue that induces the utterance of the attribute.
In step S12A, the narrow-down processing unit 155 replaces (passes) the updated narrow-down tables as inputs for other narrow-down processing. If the narrowing down tables 1 to 3 have not been narrowed down by the words included in the latest utterance (answer candidates have not decreased from the current number), the narrowing down tables to which the narrowing down table belongs are processed by other narrowing down processes. is not replaced (passed) as the input of (step S12A is skipped).

In step S13A, the integrated processing unit 156 outputs (transmits to the terminal device 10) the response to the inquiry as voice or text data. At this time, if an answer candidate by language processing has been specified as a response to the inquiry, integration processing unit 156 outputs the answer candidate by the specified language processing, and outputs the answer candidate by language processing. If not, it outputs candidate answers based on attribute relevance or dialogue content that induces attribute utterances. The response data transmitted in step S13A is received by the response reception unit 53 of the terminal device 10, and is output by the response output unit 54 as voice or image display. At this time, if multiple answer candidates have been narrowed down, these multiple answer candidates may be output as the final answer.

In step S14A, the integration processing unit 156 determines whether or not the condition for ending the automatic response process is met. As conditions for ending the automatic response process, for example, it is necessary to define that an answer has been uniquely specified for an uttered inquiry and that the response has been output, or that the user has performed an operation to end the inquiry. can be done.
If the end condition of the inquiry process is not satisfied, NO is determined in step S14A, and the process proceeds to step S1.
On the other hand, if the conditions for terminating the automatic response process are met, a determination of YES is made in step S14A, and the automatic search process ends.

As described above, in the information processing system 1 according to the present embodiment, a plurality of narrowing down processes 1 to 3 having different functions are executed in parallel as processes for selecting answer candidates based on attribute relevance. Then, the specific result of the parallel processing and the specific result of the specific processing using language processing are integrated to specify the content of the response. Then, when an answer cannot be specified by an utterance for a single inquiry, the information processing system 1 automatically narrows down attributes effective for specifying an answer, and conducts a dialogue that induces a meaningful utterance by the user. Do it autonomously.
As a result, it is possible to increase the possibility that an appropriate answer is specified as an answer candidate based on attribute relevance.

In addition, in the information processing system 1 according to the present embodiment, in addition to specifying the entire content representing an answer as an answer, a portion of the content (for example, a paragraph, a section, or a time period of a video) is specified as an answer. It is intended to be
As a result, even if a suitable answer is included as part of the content of the answer prepared for the envisioned inquiry, it is possible to identify a part of it as an answer. is more likely to be identified.
Therefore, according to the information processing system 1, when performing an automatic response to an utterance, it is possible to specify a more appropriate response content while shortening the processing time.

[Third embodiment]
Next, a third embodiment of the invention will be described.
In the information processing system 1 of the first embodiment, the information processing system 1 according to the present embodiment does not specify answer candidates by language processing, and an inquiry extraction filter generated from attributes of an assumed inquiry. using a first identification process for identifying answer candidates based on attribute relevance, and an answer extraction filter generated from the attributes of the answer prepared for the assumed inquiry, A second identification process for identifying answer candidates based on attribute relevance is executed in parallel, and the answer identified by the first identification process or the second identification process is output as a final answer. It is.

Main parts such as the system configuration of the information processing system 1, the hardware configuration of the information processing device, the functional configuration of the terminal device 10, and the like in the present embodiment are the same as those in the first embodiment, so the different portions will be mainly described. .
The functional configuration of the server 20 and the flowchart of the automatic response process, which are different from the first embodiment, will be mainly described below.

[Functional configuration]
FIG. 16 is a block diagram showing the functional configuration of the server 20 in this embodiment.
FIG. 17 is a schematic diagram showing a specific functional configuration example of the information processing system 1 in this embodiment.
The response content DB 173 stores text data (here, the name of the person in charge) representing the content of the assumed inquiry and answers (here, the telephone number of the person in charge) prepared for the assumed inquiry. Data of the associated response content table is stored.

FIG. 18 is a schematic diagram showing an example of a response content table in this embodiment.
As shown in FIG. 18, the response content table of the present embodiment contains text data (name of the person in charge) representing the content of the assumed inquiry, and answers (phone number of the person in charge) prepared for the assumed inquiry. number), a number assigned to an assumed inquiry, and a number assigned to an answer prepared for the assumed inquiry are stored in association with each other.

The connector DB 174 stores data of a first connector table corresponding to assumed inquiries and data of a second connector table corresponding to answers prepared for the assumed inquiries.
FIG. 19 is a schematic diagram showing an example of the first connector table, and FIG. 20 is a schematic diagram showing an example of the second connector table.

As shown in FIG. 19, the first connector table contains a number assigned to an assumed inquiry, text data representing the assumed inquiry (that is, the name of the person in charge), and attributes of the assumed inquiry. (call name of the person in charge, etc.) and the number given to the answer (telephone number of the person in charge) prepared for the envisioned inquiry are stored in association with each other. For example, in the second row of the first connector table shown in FIG. 19, for "Saburo Kono" (name) representing an assumed inquiry, the word "Kono" (last name) is included as an attribute of the assumed inquiry. stored.

In addition, as shown in FIG. 20, the second connector table contains words and words representing the attributes of response capability and responsibility (department, position, product in charge, etc.) for each envisioned inquiry (that is, the person in charge). , and the number assigned to the answer prepared for the envisioned inquiry are stored in association with each other. For example, the second row of the second connector table shown in FIG. /...) and a number assigned to an answer prepared for an envisioned inquiry.

Returning to FIGS. 16 and 17, the server 20 in this embodiment does not specify answer candidates by language processing, and therefore does not include the language processing unit 154 and the integration processing unit 156 . In addition, since the server 20 in this embodiment executes speech recognition processing for extracting two words in parallel without executing speech recognition processing for dictation, the dictation speech recognition unit 501 of the first embodiment shown in FIG. , a word extraction speech recognition unit 501A is provided.

Here, one of the two word extraction speech recognition processes is a speech recognition process for extracting words contained in an assumed inquiry, and the other is a speech recognition process for extracting an answer prepared for the assumed inquiry. It is a speech recognition process for extracting contained words.
Therefore, in FIG. 17, words included in an assumed inquiry are registered in the dictionary data of the word dictionary DB 172 referred to by the word extraction speech recognition unit 501A. Words included in answers prepared for envisioned inquiries are registered in the dictionary data.

16 and 17, the narrowing down processing unit 155 is a first specifying process of specifying answer candidates based on attribute relevance using an inquiry extraction filter generated from attributes of an assumed inquiry. , and using an answer extraction filter generated from the attributes of the answers prepared for the envisioned inquiry, a second identification process is performed to identify candidate answers based on the relevance of the attributes. do.
Then, the answer to the question is uniquely specified from the results of the first specifying process and the second specifying process executed by the narrowing down processing unit 155, and is used as the final answer.

FIG. 21 is a schematic diagram showing an example of query extraction filter generation.
For the sake of convenience of explanation, only the columns showing specific data in FIG. The same applies to FIG. 25).
As shown in FIG. 21, when generating an inquiry extraction filter, first, the narrow-down processing unit 155 refers to the first connector table of the connector DB 174, extracts words of the response specific attribute, and adds Generate table format data (response specific attribute table) associated with the assigned number. In addition, the narrowing down processing unit 155 refers to the first connector table of the connector DB 174, extracts the words of the response group specific attribute, and associates the number group assigned to the answer for each attribute in table format data ( response group specific attribute table). Further, the narrow-down processing unit 155 expands the words included in the response specific attribute table and the words included in the response group specific attribute table with similar words. For example, similar word group data prepared in advance for the word, word data that can be used as similar words obtained externally via the Internet (for example, similar concepts obtained by referring to an ontology) , etc.). Further, the narrow-down processing unit 155 generates table format data (expanded attribute table) in which original words included in the response specific attribute table and the response group specific attribute table are associated with similar words of the original words. Then, the narrow-down processing unit 155 generates an inquiry extraction filter composed of the word strings included in the expanded attribute table.

FIG. 22 is a schematic diagram showing an example of generating an answer extraction filter.
In this embodiment, the answer extraction filter is generated from the attributes of the answer prepared for the assumed inquiry, in addition to the inquiry extraction filter generated from the attributes of the assumed inquiry.
As shown in FIG. 22, when an answer extraction filter is generated, first, the narrow-down processing unit 155 refers to the second connector table of the connector DB 174, extracts words of the response specific attribute, and selects each answer. Generates table format data (response specific attribute table) associated with the numbers assigned to the . In addition, the narrow-down processing unit 155 refers to the second connector table of the connector DB 174, extracts the words of the response group specific attribute, and associates the number group assigned to the answer for each attribute in table format data ( response group specific attribute table). Further, the narrow-down processing unit 155 expands the words included in the response specific attribute table and the words included in the response group specific attribute table with similar words. For example, similar word group data prepared in advance for the word, word data that can be used as similar words obtained externally via the Internet (for example, similar concepts obtained by referring to an ontology) , etc.). Further, the narrow-down processing unit 155 generates table format data (expanded attribute table) in which original words included in the response specific attribute table and the response group specific attribute table are associated with similar words of the original words. Then, the narrow-down processing unit 155 generates an answer extraction filter composed of the word strings included in the expanded attribute table.

FIG. 23 is a schematic diagram showing an example of a process of identifying answer candidates based on attribute relevance using an inquiry extraction filter. Also, FIG. 24 is a schematic diagram showing an example of a process of specifying an answer candidate based on attribute relevance using an answer extraction filter.
For example, assume that the following utterance examples (1) to (3) are uttered.
Sentence example (1) "Mr. Yamada in sales, please."
Sample utterance (2) “Mr. Kawano from development, please.”
Sample utterance (3) “I would like someone who is skilled in understanding XX products.”

At this time, it is assumed that words representing attributes are extracted from the utterance of each utterance example as follows.
Utterance example (1) "Yamada" (from the inquiry extraction filter), "Sales" (from the response extraction filter)
Utterance example (2) "Kawano" (from the extraction filter for inquiries), "Development" (from the extraction filter for answers)
Example utterance (3) "XX product" (from the answer extraction filter), "Technology" (from the answer extraction filter)

In the case of utterance example (1), "Yamada" matches the words included in the inquiry extraction filter, "Sales" matches the words included in the answer extraction filter, and #2 and #3 from "Yamada" The answers are specified, and the answers #3 and #4 are specified from "sales". That is, in the case of utterance example (1), the answer #2 (Taro Yamada) is specified from the logical product of the results of the first specifying process and the second specifying process, and is taken as the final answer.

In the case of utterance example (2), "Kawano" matches the words included in the query extraction filter, and "Development" matches the words included in the answer extraction filter. #1 is also identified. That is, in the case of the utterance example (2), the answer #1 (Saburo Kono) is specified from the result of the first specifying process or the second specifying process, and is taken as the final answer.

In the case of utterance example (3), the attributes do not match the words included in the query extraction filter, and "XX product" and "technique" match the words included in the answer extraction filter. The answers #1 and #3 are specified, and the answer #1 is specified from "technology". That is, in the case of utterance example (3), the answer #1 (Saburo Kono) is specified from the result of the second specifying process, and is taken as the final answer.

[motion]
Next, the operation of the information processing system 1 will be described.
FIG. 25 is a flow chart showing the flow of automatic response processing executed by the information processing system 1 .
The automatic response process is started in the server 20 in response to an operation instructing execution of the automatic response process.
In contrast to the automatic response processing of the first embodiment shown in FIG. 10A, the automatic response processing in this embodiment is performed by changing the processing of step S41-1 to step S41-1A and the processing of step S41-2 to step S41-2A. , and steps S5 to S9 are changed to steps S105 to S107.

That is, in step S4, the speech recognition processing unit 153 of the server 20 performs parallel speech recognition processing (S41-1A to S43-1 and S41-2A to S43-2).
Specifically, in step S41-1A, the word extraction speech recognition unit 501A performs speech recognition processing for word extraction (here, rule grammar speech recognition processing or speech recognition processing using DNN). Speech recognition engine processes speech data, extracts specific words from the speech data, and converts them to text data. Note that the speech recognition processing in step S41-1A is speech recognition processing for extracting words included in an assumed inquiry. A speech recognition process is executed with reference to the file.

In step S41-2A, the word extraction speech recognition unit 511 performs speech recognition processing for word extraction (here, rule grammar speech recognition processing or speech recognition processing using DNN). The engine processes the voice data, extracts specific words from the voice data, and converts them into text data. It should be noted that the speech recognition processing in step S41-2A is speech recognition processing for extracting words included in answers prepared for an assumed inquiry, and is prepared for an assumed inquiry. A speech recognition process is executed by referring to a dictionary for word extraction in which words included in answers are registered in advance.
After steps S43-1 and S43-2, the process proceeds to step S105.

In step S105, the narrow-down processing unit 155 executes the first identification process and the second identification process to identify the final answer to the uttered inquiry. That is, the narrowing-down processing unit 155 determines that the speech recognition result of word extraction acquired by the speech recognition processing unit 153-1 (that is, the word included in the assumed query) matches the word included in the query extraction filter. If there is a match, the answers prepared in advance are narrowed down based on the matched word included in the inquiry extraction filter. In addition, the narrowing down processing unit 155 extracts the voice recognition result of word extraction acquired by the voice recognition processing unit 153-2 (that is, the word included in the answer prepared for the assumed inquiry), and the extraction for answer. It is determined whether or not the words included in the filter match, and if they match, the prepared answers are narrowed down based on the matching words included in the answer extraction filter. Then, the narrowing-down processing unit 155 specifies the final answer to the uttered inquiry based on the narrowing-down results of the first specifying process and the second specifying process.

In step S106, the integration processing unit 156 transmits the specified final answer to the terminal device 10 as voice or text data. The transmitted response data is received by the response receiving section 53 of the terminal device 10 and output by the response output section 54 in the form of voice or image display. Note that the response output in step S106 may be a uniquely specified answer, a response prompting an utterance for uniquely specifying a plurality of answer candidates, or a plurality of answer candidates. It may be used as a response.

In step S107, the integration processing unit 156 determines whether or not the end condition of the automatic response process is met. As conditions for ending the automatic response process, for example, it is necessary to define that an answer has been uniquely specified for an uttered inquiry and that the response has been output, or that the user has performed an operation to end the inquiry. can be done.
If the end condition of the inquiry process is not satisfied, NO is determined in step S107, and the process proceeds to step S1.
On the other hand, if the end conditions for the automatic response process are met, a determination of YES is made in step S107, and the automatic response process ends.

As described above, in the information processing system 1 according to the present embodiment, the query extraction filter generated from the attributes included in the assumed query is used to specify the answer candidate based on the attribute relevance. a second identification process for identifying answer candidates based on the relevance of the attributes using the identification process and an answer extraction filter generated from the attributes of the answers prepared for the envisioned inquiry; are executed in parallel, and the answer specified in the first specific process or the second specific process is output as the final answer.
Therefore, it is possible to identify an answer to an inquiry based on the content of the assumed inquiry and the attribute of the answer prepared for the assumed inquiry, without specifying answer candidates by language processing.
Therefore, when performing an automatic response to an utterance, it is possible to specify a more appropriate response content while shortening the processing time.

[Modification 1]
In each of the above-described embodiments, the case where multiple tables are generated as search data has been described as an example, but the present invention is not limited to this.
In other words, the format of the search data is not particularly limited, and as long as the data (words representing attributes, answer numbers, etc.) can be traced to identify the answer, it can be composed of a single table. , pointers, etc., to form data groups in which relationships are defined.

[Modification 2]
In the second embodiment, narrowing down process 2 has been described as identifying answer candidates based on attribute relevance based on the latest utterance and the previous utterance, but the present invention is not limited to this.
That is, the narrowing process 2 can identify answer candidates based on attribute relevance based on the latest utterance and an arbitrary number of past utterances preceding the latest utterance.
As a result, the number of past utterances to be reflected in the narrowing process 2 can be appropriately set according to the statistical data or the application to which the information processing system 1 is applied.
In addition, as the narrowing process 2, a plurality of processes (that is, multiple layers of narrowing down process 2) can be executed in parallel.
As a result, even if the direction of the user's utterance for an inquiry changes, or if the user utters a word that does not conform to the guidance from the system, the refinement results (refinement table) based on the most recent predetermined utterances ) can be utilized to continue the narrowing process.
Therefore, since it is possible to accurately reflect the user's intention and narrow down the answers, it is possible to quickly specify an appropriate answer to the inquiry according to the user's preference.

[Modification 3]
In the first and second embodiments, the speech recognition processing unit 153 includes a first speech recognition unit 153-1 to an n-th speech recognition unit 153-n, and executes a plurality of speech recognition processes in parallel. was described as an example, but it is not limited to this. For example, the speech recognition processing unit 153 outputs the results of speech recognition processing performed by one speech recognition engine (for example, a speech recognition engine that performs general-purpose dictation), and responds to the speech recognition results by language processing. and the process of identifying answer candidates based on attribute relevance may be executed in parallel. For example, narrowing down processes 1 to 3 of the second embodiment and process of identifying answer candidates by language processing can be executed in parallel for one speech recognition result.
In addition, for one speech recognition result, in addition to identifying answer candidates by language processing and answer candidates based on attribute relevance, the content of dialogue that induces utterances with attributes effective for answer identification ( message) can also be specified.
At this time, AI (Artificial Intelligence) may be used to respond to the user's utterance for inquiry with a dialogue that induces an utterance of an answer or an attribute effective for specifying the answer.
In this case, while suppressing an increase in processing load, it is possible to obtain more useful information than the processing result of a single speech recognition process.
In addition to general-purpose dictation, the types of speech recognition processing used at this time include those aimed at dictation for specific fields, speech recognition by extracting words, and various other types of speech recognition. Treatment can be used.

[Modification 4]
In the second embodiment, speech recognition processing for extracting words using an extraction filter is executed for the user's utterance, and based on the extracted words, narrowing down processes 1 to 3 are performed to answer based on the relevance of attributes. Although the case of narrowing down the candidates for is described as an example, it is not limited to this.
That is, in addition to the form in which the user utters an utterance containing a word for narrowing down the answer and inputs the word by voice recognition, the form in which the word is input by operating the screen can be used.
For example, a word for an inquiry to be input in narrowing process 1 is input by screen operation, or a word for further narrowing down the answer candidates narrowed down by narrowing process 1 or narrowing process 2 is input by screen operation. can do.
As a form of inputting words by operating the screen, the display (output unit 816) of the terminal device 10 scrolls (moves and displays) words (words representing attributes) that are related to the answer candidates in the window. Then, the user can select a word by mouse click or tap operation. In addition, as another form of inputting words by operating the screen, on the display (output unit 816) of the terminal device 10, words (words representing attributes) related to answer candidates are entered in a window using a pull-down menu or the like. A form in which a list is displayed and the user selects a word by mouse click or tap operation, or a form in which a list of words is displayed in a table format and the user selects a word by turning on a check box attached to each word. and so on.

As words to be selected in the narrowing process 1, words stored in each table included in the connector tables can be displayed. In addition, as words to be selected in narrow-down processing 2 or narrow-down processing 3, words stored in each table included in narrow-down table type 1 or narrow-down table type 2 (that is, words after narrowing down by narrow-down processing ) can be displayed.
When a word is input by operating the screen, the user can use the candidate words displayed on the screen as hints to select an effective word for specifying the answer.
It is also possible to display words (words representing attributes) related to answer candidates on the screen in one of the forms described above, and to select a word by uttering one of the words by the user. be.
Furthermore, in addition to the form of inputting words by screen operation, it is also possible to adopt a form of inputting words by keyboard operation or the like.
In this case, when the user inputs arbitrary text (words, sentences, etc.), by extracting words included in the input text, the answer candidates are narrowed down based on attribute relevance. can be done.

[Modification 5]
In the above-described embodiment and modifications, the case of accepting inquiries by voice has been described, but the present invention is not limited to this.
In other words, it is possible to receive inquiries in the form of text data, such as inquiries by chat.
Therefore, in the above-described embodiments and modifications, the configuration for performing speech recognition processing on voice inquiries can be appropriately replaced with a configuration for acquiring text data representing the content of the inquiry.

The information processing system 1 configured as described above includes a connector DB 174 , a voice recognition processing section 153 , a narrowing processing section 155 and an integration processing section 156 .
The connector DB 174 stores first attribute information regarding envisioned inquiries.
The connector DB 174 stores second attribute information regarding answers to envisioned inquiries.
The speech recognition processing unit 153 acquires the content of the inquiry.
The narrowing processing unit 155 and the integration processing unit 156 generate a response (answer or dialogue message).
As a result, it is possible to specify the answer to the inquiry from the content of the assumed inquiry and the attribute of the answer prepared for the assumed inquiry.
Therefore, when an automatic response to an inquiry is made, it is possible to specify a more appropriate response content while shortening the processing time.

The voice recognition processing unit 153 acquires a voice recognition result obtained by performing voice recognition processing on an utterance for an inquiry.
As a result, when an inquiry is made by voice, it is possible to specify a more appropriate response content while shortening the processing time.

The narrowing-down processing unit 155 represents attribute information included in the speech recognition result of the speech recognition process, first attribute information and second attribute information, and attributes expanded from the first attribute information and second attribute information. A response containing an answer to the inquiry is identified based on the extended attribute information.
This makes it possible to easily identify answers that match the uttered inquiry even when unexpected expressions are used, such as when expressions fluctuate or words with similar meanings are used. becomes.

The information processing system 1 also includes a speech recognition processing unit 153 , a language processing unit 154 , a narrowing processing unit 155 and an integration processing unit 156 .
The voice recognition processing unit 153 acquires a voice recognition result obtained by performing voice recognition processing on an utterance for an inquiry.
The language processing unit 154 identifies, by language processing, answer candidates that match the inquiry from the semantic content represented by the speech recognition result of the speech recognition processing unit 153 .
The narrowing down processing unit 155 extracts words included in the inquiry from the speech recognition results of the speech recognition processing unit 153, and based on the relevance between the attribute information based on the words included in the inquiry and the attribute information based on the answers to the inquiry. , to identify candidate answers that match the query.
The integration processing unit 156 identifies a response including an answer to the inquiry based on the answer candidates identified by the language processing unit 154 and the answer candidates identified by the narrowing processing unit 155 .
Accordingly, it is possible to specify an answer to an inquiry by using not only a specification method based on language processing but also a specification method based on matching of words representing attributes.
Therefore, when performing an automatic response to an utterance, it is possible to specify a more appropriate response content while shortening the processing time.

The narrow-down processing unit 155 identifies a response including an answer to the inquiry based on attribute information based on the answer to the inquiry and expanded attribute information representing attributes expanded from the attribute information based on the answer to the inquiry.
This makes it possible to easily identify answers that match the uttered inquiry even when unexpected expressions are used, such as when expressions fluctuate or words with similar meanings are used. becomes.

If the accuracy of the answer candidate acquired by the language processing unit 154 is equal to or higher than the set threshold, the integration processing unit 156 regards the answer candidate acquired by the language processing unit 154 as an answer, and If the accuracy of the obtained answer candidate is less than the set threshold, the answer candidate specified by the narrowing processing unit 155 is set as the answer.
As a result, when an inquiry is uttered, it is possible to obtain a more accurate speech recognition result at high speed and specify an appropriate answer.

The information processing system 1 also includes a speech recognition processing unit 153 , a narrowing processing unit 155 and an integration processing unit 156 .
Based on the first attribute information about the assumed inquiry and the second attribute information about the answer to the assumed inquiry, the narrowing processing unit 155 selects one or more associated with the answer to the assumed inquiry. Generate search data composed of attributes.
The speech recognition processing unit 153 acquires the content of the inquiry.
The narrow-down processing unit 155 and the integration processing unit 156 specify a response including an answer to the content of the inquiry based on the attribute information of the content of the inquiry and the search data.
As a result, when an inquiry is made, it is possible to specify an answer at high speed using the search data.

The voice recognition processing unit 153 acquires a voice recognition result obtained by performing voice recognition processing on an utterance for an inquiry.
As a result, when an inquiry is made by voice, it is possible to identify the answer at high speed using the search data.

The search data is prepared before the content of the inquiry is obtained, and when the content of the inquiry is obtained, the generated search data is referenced to identify the answer to the content of the inquiry.
As a result, since the search data is prepared before the inquiry is made, it becomes possible to specify the answer at high speed when the inquiry is uttered.

The search data includes response specifying information representing attributes that can uniquely specify an answer, and response group specifying information representing attributes that can specify multiple answer candidates.
This makes it possible to appropriately determine whether the words contained in the utterance for the inquiry represent attributes that can uniquely identify an answer or attributes that can identify multiple candidate answers. A process can be performed to identify the answer.

The search data includes expanded attribute information representing attributes expanded from the attributes contained in the response specifying information and the response group specifying information.
This makes it possible to easily identify answers that match the uttered inquiry even when unexpected expressions are used, such as when expressions fluctuate or words with similar meanings are used. becomes.

The narrowing-down processing unit 155, when the content of the inquiry acquired by the speech recognition processing unit 153 identifies a plurality of answers to the inquiry, selects one associated with the answer to the assumed inquiry included in the search data. Or narrow down the attributes to attributes that uniquely identify answers to the content of the inquiry, or attributes that identify fewer answers to the content of the inquiry.
Based on the narrowed-down attribute, the narrow-down processing unit 155 executes, as a response, dialogue for guiding additional inquiry content for specifying the answer to the inquiry content.
As a result, even if the user does not make an inquiry with appropriate content, it is possible to receive further inquiries and specify appropriate answers.

The narrowing-down processing unit 155 performs a first narrowing-down process of specifying the answer to the content of the latest inquiry for the entire original search data, and search data whose attributes are narrowed down from the original search data. As a target, a second narrowing-down process of specifying answers based on the contents of inquiries made a predetermined number of times in the past is executed in parallel.
As a result, it is possible to increase the possibility that an appropriate answer is specified as an answer candidate based on attribute relevance.

The narrowing-down processing unit 155 performs a third narrowing-down process of specifying answers based on the content of a predetermined number of inquiries in the past different from the second narrowing-down process as the first narrowing-down process and the second narrowing-down process. Run in parallel.
As a result, even if the directionality of the content of the user's inquiry changes, or if the user makes an inquiry that does not conform to the guidance from the system, the results of narrowing down by the most recent predetermined number of inquiries (narrowing-down table) can be used to continue the narrowing process.
Therefore, since it is possible to accurately reflect the user's intention and narrow down the answers, it is possible to quickly specify an appropriate answer to the inquiry according to the user's preference.

The narrow-down processing unit 155 displays on the screen words representing attributes contained in the original search data or words representing attributes narrowed down from the original search data, and selects the words displayed on the screen. , to narrow down the answers to the content of the inquiry.
As a result, the user can use the candidate words displayed on the screen as hints to select effective words for specifying the answer.

The information processing system 1 also includes a voice recognition processing unit 153 and a narrowing down processing unit 155 .
The voice recognition processing unit 153 acquires a voice recognition result obtained by performing voice recognition processing on an utterance for an inquiry.
The narrow-down processing unit 155 extracts first attribute information related to an assumed inquiry, second attribute information related to an answer to the assumed inquiry, and expanded attributes related to the first attribute information and the second attribute information. The attribute information included in the speech recognition result of the speech recognition processing unit 153 is filtered by a filter including information as an element.
The narrow-down processing unit 155 identifies responses including answers to the inquiry utterances based on the filtering results.
As a result, it is possible to quickly extract words representing attributes from utterances for an inquiry and to specify answer candidates based on the relevance of the attributes.

The narrow-down processing unit 155 identifies the first attribute information or the second attribute information related to the word extracted from the utterance for inquiry, and associates it with the first attribute information or the second attribute information. Identify the answer to the envisioned query as the answer to the utterance for the query.
This makes it possible to easily identify answers related to words included in the utterance for inquiry.

It should be noted that the present invention is not limited to the above-described embodiments, and includes modifications, improvements, and the like within the scope of achieving the object of the present invention.
For example, in the above-described embodiment, an example in which the information processing system 1 is configured as a system including a plurality of information processing apparatuses has been described, but the present invention is not limited to this. That is, the present invention may be implemented by providing the functions of the information processing system 1 in a single information processing apparatus.

Also, the functions of the server 20 may be distributed and implemented in a plurality of servers, the functions of the server may be partially implemented in the terminal device 10, and the functions of the terminal device 10 may be partially implemented in the server. good too.
For example, in each of the above-described embodiments, the speech recognition processing is executed in the server 20, but the speech recognition processing may be executed in the terminal device 10, and the speech recognition result executed in another system etc. may be acquired (received, etc.) and used for processing.
Further, in each of the above-described embodiments, the search data can be sequentially updated in accordance with the update of the response content table or the update (addition, correction, deletion, etc.) of words representing attributes based on operation results.

Further, it is possible to implement the present invention by appropriately combining the configurations described in the above-described embodiments and modifications.
The series of processes described above can be executed by hardware or by software.
In other words, the functional configurations of FIGS. 3, 4, 9, 16, and 17 are merely examples and are not particularly limited. That is, it is sufficient if the information processing system 1 has a function capable of executing the above-described series of processes as a whole. It is not limited to examples 16 and 17.
Also, one functional block may be composed of hardware alone, software alone, or a combination thereof.

When executing a series of processes by software, a program constituting the software is installed in a computer or the like from a network or a recording medium.
The computer may be a computer built into dedicated hardware. Also, the computer may be a computer capable of executing various functions by installing various programs, such as a general-purpose personal computer.

A storage medium for storing the program is constituted by a removable medium distributed separately from the apparatus main body, or a storage medium preinstalled in the apparatus main body. Removable media are composed of, for example, magnetic disks, optical disks, magneto-optical disks, flash memories, or the like. Optical discs are composed of, for example, CD-ROMs (Compact Disk-Read Only Memory), DVDs (Digital Versatile Disks), Blu-ray Discs (registered trademark), and the like. The magneto-optical disk is composed of an MD (Mini-Disk) or the like. The flash memory is, for example, a USB (Universal Serial Bus) memory or an SD card. Further, the storage medium pre-installed in the apparatus body is composed of, for example, a ROM, a hard disk, or a flash memory in which programs are stored.

In this specification, the steps of writing a program recorded on a recording medium are not necessarily processed chronologically according to the order, but may be executed in parallel or individually. It also includes the processing to be performed.
Further, in this specification, the term "system" means an overall device composed of a plurality of devices, a plurality of means, or the like.

1 information processing system, 10 terminal device, 20 server, 30 network, 800 information processing device, 811 CPU, 812 ROM, 813 RAM, 814 bus, 815 input unit, 815a operation input unit, 815b voice input unit, 816 output unit, 817 storage unit, 818 communication unit, 819 drive, 820 imaging unit, 831 removable medium, 51 speech reception unit, 52 speech data transmission unit, 53 response reception unit, 54 response output unit, 151 speech data reception unit, 152 preprocessing unit , 153 speech recognition processing unit, 153-1 first speech recognition unit, 153-2 second speech recognition unit, 154 language processing unit, 155 narrowing processing unit, 156 integration processing unit, 171 speech recognition dictionary database (speech recognition dictionary DB ), 172 word dictionary database (word dictionary DB), 173 response content database (response content DB), 174 connector database (connector DB), 501 dictation speech recognition unit, 502 first recognition result specifying unit, 503 first recognition result output Section 511 Word Extraction Speech Recognition Section 512 Second Recognition Result Identification Section 513 Second Recognition Result Output Section

Claims (10)

An information processing system including one or more information processing devices,
a first attribute information storage means in which first attribute information related to an assumed inquiry is stored in advance ;
a second attribute information storage means in which second attribute information related to answers to the assumed inquiries is stored in advance ;
inquiry content acquisition means for acquiring the content of an inquiry;
response specifying means for specifying a response including an answer to the inquiry based on the attribute information of the content of the inquiry and the first attribute information and the second attribute information;
with
The attribute information, the first attribute information, and the second attribute information included in the contents of the inquiry are represented by words, and the first attribute information includes words contained in the assumed inquiry, and , contains words that are related to the content of the assumed inquiry and are not included in the assumed inquiry,
The response identifying means searches for a word representing the first attribute information and a word representing the second attribute information having relevance to a word representing attribute information included in the content of the inquiry, thereby identify the answer,
An information processing system , wherein a process of specifying a response including an answer to the inquiry is performed in parallel with a process of specifying the answer to the inquiry using language processing. 2. The information processing system according to claim 1, wherein said inquiry content acquiring means acquires a speech recognition result obtained by subjecting an utterance for inquiry to speech recognition processing. The response identifying means determines attribute information included in the contents of the inquiry, the first attribute information and the second attribute information, and attributes expanded from the first attribute information and the second attribute information. 3. The information processing system according to claim 1, wherein a response including an answer to the inquiry is specified based on the extended attribute information represented. An information processing system including one or more information processing devices,
voice recognition means for acquiring voice recognition results obtained by performing voice recognition processing on an utterance for an inquiry;
language processing means for specifying, by language processing, candidate answers that match the inquiry from the semantic content expressed by the speech recognition results of the speech recognition means;
extracting words included in the inquiry from the speech recognition result of the speech recognition means; attribute processing means for identifying suitable answer candidates;
response specifying means for specifying a response including an answer to the inquiry based on the answer candidate specified by the language processing means and the answer candidate specified by the attribute processing means;
with
The attribute information based on the words included in the inquiry and the attribute information based on the answers to the inquiry are represented by words,
The attribute processing means searches for words representing attribute information based on answers to the inquiry that are related to words representing attribute information based on the words included in the inquiry, thereby finding answer candidates that match the inquiry. identify,
Information characterized in that the process of specifying an answer candidate that matches the query by the language processing means and the process of specifying an answer candidate that matches the query by the attribute processing means are executed in parallel. processing system. The attribute processing means generates a response including an answer to the inquiry based on attribute information based on the answer to the inquiry and expanded attribute information representing attributes expanded from the attribute information based on the answer to the inquiry to similar words. 5. The information processing system according to claim 4, wherein the information processing system is specified. When the probability of the answer candidate acquired by the language processing means is equal to or higher than a set threshold, the response specifying means sets the answer candidate acquired by the language processing means as an answer, and the language processing means 6. The information processing system according to claim 4, wherein when the accuracy of the obtained answer candidate is less than a set threshold, the answer candidate specified by the attribute processing means is set as the answer. . An information processing method executed in an information processing system including one or more information processing devices,
an inquiry content acquisition step for acquiring the content of an inquiry;
A response that specifies a response including an answer to the inquiry based on attribute information of the content of the inquiry, first attribute information related to the assumed inquiry, and second attribute information related to the assumed answer to the inquiry. a specific step;
including
The first attribute information and the second attribute information are stored in advance,
The attribute information, the first attribute information, and the second attribute information included in the contents of the inquiry are represented by words, and the first attribute information includes words contained in the assumed inquiry, and , contains words that are related to the content of the assumed inquiry and are not included in the assumed inquiry,
In the response identifying step, by searching for a word representing the first attribute information and a word representing the second attribute information having relevance to a word representing attribute information possessed by the content of the inquiry, identify the answer,
An information processing method , wherein the process of specifying a response including an answer to the inquiry is performed in parallel with the process of specifying the answer to the inquiry using language processing. An information processing method executed in an information processing system including one or more information processing devices,
a speech recognition step of obtaining a speech recognition result obtained by performing speech recognition processing on an utterance for an inquiry;
a language processing step of specifying, by language processing, answer candidates that match the inquiry from the semantic content represented by the speech recognition result in the speech recognition step;
Words included in the query are extracted from the voice recognition result in the voice recognition step, and based on the relationship between attribute information based on the words included in the query and attribute information based on answers to the query, an attribute processing step of identifying candidate matching answers;
a response identification step of identifying a response including an answer to the inquiry based on the answer candidate identified in the language processing step and the answer candidate identified in the attribute processing step;
including
The attribute information based on the words included in the inquiry and the attribute information based on the answers to the inquiry are represented by words,
The attribute processing step searches for words representing attribute information based on answers to the inquiry that are related to words representing attribute information based on words contained in the inquiry, thereby finding answer candidates that match the inquiry. identify,
Information characterized in that the process of identifying answer candidates that match the query in the language processing step and the process of identifying answer candidates that match the query in the attribute processing step are executed in parallel. Processing method. to the computer,
an inquiry content acquisition function for acquiring the content of an inquiry;
A response that specifies a response including an answer to the inquiry based on attribute information of the content of the inquiry, first attribute information related to the assumed inquiry, and second attribute information related to the assumed answer to the inquiry. a specific function;
to realize
The first attribute information and the second attribute information are stored in advance,
The attribute information, the first attribute information, and the second attribute information included in the contents of the inquiry are represented by words, and the first attribute information includes words contained in the assumed inquiry, and , contains words that are related to the content of the assumed inquiry and are not included in the assumed inquiry,
The response specifying function searches for a word representing the first attribute information and a word representing the second attribute information having relevance to a word representing attribute information included in the content of the inquiry, thereby responding to the inquiry. identify the answer,
A program , wherein the process of specifying a response including an answer to the inquiry is performed in parallel with the process of specifying the answer to the inquiry using language processing. to the computer,
a speech recognition function that obtains a speech recognition result obtained by subjecting an utterance for an inquiry to speech recognition processing;
a language processing function that identifies, by language processing, answer candidates that match the inquiry from the semantic content expressed by the speech recognition results of the speech recognition function;
Words included in the query are extracted from the voice recognition result of the voice recognition function, and based on the relationship between attribute information based on the words included in the query and attribute information based on the answers to the query, an attribute processing function that identifies suitable candidate answers;
a response specifying function for specifying a response including an answer to the inquiry based on the answer candidate specified by the language processing function and the answer candidate specified by the attribute processing function;
to realize
The attribute information based on the words included in the inquiry and the attribute information based on the answers to the inquiry are represented by words,
The attribute processing function searches for words representing attribute information based on answers to the inquiry that are related to words representing attribute information based on the words included in the inquiry, thereby finding answer candidates that match the inquiry. identify,
A program characterized in that a process in which the language processing function identifies answer candidates that match the query and a process in which the attribute processing function identifies answer candidates that match the query are executed in parallel. .

Images