WO2022209313A1

WO2022209313A1 - Information processing device, information processing method, and information processing program

Info

Publication number: WO2022209313A1
Application number: PCT/JP2022/004789
Authority: WO
Inventors: 一憲荒木; 文規本間
Original assignee: ソニーグループ株式会社
Priority date: 2021-03-29
Filing date: 2022-02-08
Publication date: 2022-10-06
Also published as: JPWO2022209313A1

Abstract

An information processing device according to the present disclosure comprises: a first selection unit for selecting a first QA pair that corresponds to a search query from a first data group that includes a plurality of QA pairs which are the combinations of question information corresponding to a response and response information that indicates the response; and a second selection unit for selecting, on the basis of the first QA pair selected by the first selection unit, response information that corresponds to the search query as target response information from a second data group that includes a plurality of QA pairs which are different from the first data group.

Description

Information processing device, information processing method and information processing program

The present disclosure relates to an information processing device, an information processing method, and an information processing program.

Conventionally, there are known services that provide answers in response to input such as questions from users. For example, an artificial intelligence algorithm obtained by performing supervised learning using the correspondence relationship between input patterns and answer information stored in an answer database as teacher data automatically generates answer information according to the user's input. (for example, Patent Document 1).

Patent No. 6624539

However, with conventional technology, it is not always possible to select an appropriate response to a search query. For example, in the prior art, the user's next input is predicted, and the prediction result is automatically displayed to the user as the next input candidate, but what is predicted and displayed is the user's input, There is room for improvement in response to the answer to the user. Therefore, it is desirable to select appropriate responses to search queries.

Therefore, the present disclosure proposes an information processing device, an information processing method, and an information processing program capable of selecting an appropriate response to a search query.

In order to solve the above problems, an information processing apparatus according to one embodiment of the present disclosure provides a first data group including a plurality of QA pairs that are a combination of question information corresponding to a response and response information indicating the response, , a first selection unit that selects a first QA pair corresponding to a search query, and a plurality of QA pairs different from the first data group based on the first QA pair selected by the first selection unit a second selection unit that selects the response information corresponding to the search query from the two data groups as target response information.

FIG. 3 is a diagram showing an example of information processing according to an embodiment of the present disclosure; FIG. 1 is a diagram illustrating a configuration example of an information processing system according to an embodiment of the present disclosure; FIG. 1 is a diagram illustrating a configuration example of an information processing device according to an embodiment of the present disclosure; FIG. It is a figure which shows an example of the response log storage part which concerns on embodiment of this indication. 4 is a diagram illustrating an example of a log QA storage unit according to an embodiment of the present disclosure; FIG. FIG. 11 illustrates an example of a manual QA storage unit according to an embodiment of the present disclosure; FIG. It is a figure which shows an example of a merge process. 1 is a diagram illustrating a configuration example of a terminal device according to an embodiment of the present disclosure; FIG. 4 is a flow chart showing a processing procedure of the information processing device according to the embodiment of the present disclosure; FIG. 4 is a sequence diagram showing processing procedures of the information processing system according to the embodiment of the present disclosure; It is a figure which shows the system configuration|structure outline|summary of an information processing system. 4 is a flowchart showing processing of a first processing pattern; It is a figure which shows an example of a 2nd processing pattern. 9 is a flowchart showing processing of a second processing pattern; It is a figure which shows an example of a 3rd process pattern. 10 is a flowchart showing processing of a third processing pattern; It is a figure which shows an example of the 1st form regarding the application of a system. FIG. 10 is a diagram showing an example of a second mode of application of the system; FIG. 10 is a diagram showing an example of a third mode of application of the system; It is a figure which shows an example of the information for a display. It is a figure which shows an example of the 1st form regarding service provision. It is a figure which shows an example of the 2nd form regarding service provision. It is a figure which shows an example of the display by the side of an operator. It is a figure which shows an example of a display on the customer side. It is a figure which shows an example of the 3rd form regarding service provision. It is a figure which shows an example of domain deployment. It is a figure which shows an example of generation|occurrence|production of manual QA. FIG. 10 is a diagram showing an example of a case where an image is used as a response; It is a figure which shows an example in the case of making a response by voice. 1 is a hardware configuration diagram showing an example of a computer that implements functions of an information processing apparatus; FIG.

Below, embodiments of the present disclosure will be described in detail based on the drawings. The information processing apparatus, information processing method, and information processing program according to the present application are not limited to this embodiment. Further, in each of the following embodiments, the same parts are denoted by the same reference numerals, thereby omitting redundant explanations.

The present disclosure will be described according to the order of items shown below.
1. Embodiment 1-1. Outline of information processing according to embodiment of present disclosure 1-1-1. Background and Effects 1-2. Configuration of information processing system according to embodiment 1-3. Configuration of Information Processing Apparatus According to Embodiment 1-3-1. Merge example 1-4. Configuration of terminal device according to embodiment 1-5. Information processing procedure according to embodiment 1-5-1. Procedure of processing related to information processing apparatus 1-5-2. Procedure of processing related to information processing system 1-6. Overview of system configuration 1-7. Processing example 1-7-1. First processing pattern 1-7-2. Second processing pattern 1-7-3. Third processing pattern 1-8. Adaptation example 1-9. Example of service provision 1-9-1. Display example 1-9-2. Application configuration example 1-10. Example of domain expansion 1-11. Manual QA generation example (human in the loop)
1-12. Examples of information types 1-12-1. Image 1-12-2. Voice 2. Other Embodiments 2-1. Other configuration examples 2-2. Others 3. Effects of the present disclosure 4 . Hardware configuration

[1. embodiment]
[1-1. Overview of information processing according to the embodiment of the present disclosure]
FIG. 1 is a diagram illustrating an example of information processing according to an embodiment of the present disclosure. Information processing according to the embodiment of the present disclosure is implemented by an information processing system 1 (see FIG. 2) including an information processing device 100 (see FIG. 3) and a terminal device 10 (see FIG. 8).

The information processing device 100 is an information processing device that executes information processing according to the embodiment. In FIG. 1, the information processing apparatus 100 combines question information (also simply referred to as "question") corresponding to a response and response information (also simply referred to as "response") indicating the response (also referred to as "answer"). (hereinafter also referred to as “QA pairs”) is used. The first data group LQA is a log QA (Question-Answer) generated from past response history (response log), which will be described later. The question here may be anything as long as it can be associated with a response (also referred to as an "answer"). That is, the question here may be anything as long as it triggers a response following the question. In other words, questions are not limited to interrogative questions or inquiries, but are concepts that include various questions as long as they provoke reactions (responses) from other subjects.

The first data group LQA includes a QA pair P11 that is a combination of question Q11 and response A11, a QA pair P12 that is a combination of question Q12 and response A12, and a QA pair P13 that is a combination of question Q13 and response A13. , including multiple QA pairs such as

In addition, in FIG. 1, the information processing apparatus 100 uses the second data group MQA including a plurality of combinations (QA pairs) of questions and responses corresponding to the questions. The second data group MQA is manual QA created by humans. The second data group MQA is created manually (manually) so as to satisfy the conditions regarding responses. For example, the second data group MQA is manually created so as not to contain personal information. Also, for example, the second data group MQA is manually created so as not to depend on the context. The creation of manual QA such as the second data group MQA will be described later.

The second data group MQA includes a QA pair P21 that is a combination of question Q21 and response A21, a QA pair P22 that is a combination of question Q22 and response A22, and a QA pair P23 that is a combination of question Q23 and response A23. , including multiple QA pairs such as Here, since the responses such as the responses A21 to A23 included in the second data group MQA are manually created so as to satisfy the response conditions, they do not include responses that depend on personal information or context.

From here, the information processing apparatus 100 selects a response (also referred to as a “target response”) corresponding to a search query (simply referred to as a “query”) using the first data group LQA and the second data group MQA. explain. In FIG. 1, when selecting a QA pair from the second data group MQA, a processing pattern using question information (hereinafter also referred to as "first processing pattern") will be described. Patterns for selecting pairs are not limited to this. Patterns other than the first processing pattern will be described later.

　Fig. 1 shows a case where the search query QE1 is the character information (character string) "I went to a chiropractor due to an accident. Will I be compensated?" The search query QE1 is input by a person such as an operator or a customer of the service (QA service) provided by the information processing apparatus 100 (also referred to as a “user”). Note that the input mode may be any form such as character input using a keyboard or touch panel, input by voice, or the like.

The information processing device 100 uses the first data group LQA and the search query QE1 to select a QA pair (also referred to as "first QA pair") corresponding to the search query QE1 (step S1). The information processing apparatus 100 selects the first QA pair from the first data group LQA by searching the first data group LQA using the search query QE1.

The information processing device 100 compares the search query QE1 with each question in the first data group LQA, and selects a QA pair of questions similar to the search query QE1 among the questions in the first data group LQA as a first QA pair. select. The information processing device 100 compares a vector obtained by vector-converting the search query QE1 (also referred to as a "query vector") and a vector obtained by vector-converting each question in the first data group LQA (also referred to as a "question vector"). . Then, the information processing apparatus 100 selects, from the first data group LQA, QA pairs of questions with question vectors similar to the query vector of the search query QE1 as first QA pairs. In FIG. 1, in order to simplify the explanation, the information processing apparatus 100 selects, from the first data group LQA, the QA pair of the question whose question vector is most similar to the query vector of the search query QE1 as the first QA pair. .

For example, the information processing apparatus 100 takes text (character information) as input and converts the character information into a vector using a model (vector conversion model) or the like that outputs a vector (embedded expression) expressing the text. Vector conversion of text (character information) will be described later. The information processing device 100 converts the search query QE1 into a query vector using a vector conversion model. The information processing device 100 converts each question in the first data group LQA into a question vector using a vector conversion model.

It should be noted that the QA pair information (questions, etc.) in the first data group LQA may be converted into vectors in advance. The information processing apparatus 100 calculates the cosine similarity between the query vector of the search query QE1 and the question vector of each question in the first data group LQA, and selects the QA pair of the question of the question vector having the highest cosine similarity. Select as 1QA pair. Note that the cosine similarity is merely an example, and various types of information may be used to measure the closeness between vectors without being limited to the cosine similarity. In FIG. 1, the information processing device 100 selects the QA pair P12, which is the combination of the question Q12 and the response A12, as the first QA pair.

Then, the information processing apparatus 100 selects a QA pair (also referred to as a "second QA pair") from the second data group MQA using the second data group MQA and the first QA pair (step S2). In FIG. 1 , the information processing apparatus 100 searches the second data group MQA using the question Q12 (also referred to as “first question Q12”) of the QA pair P12 selected as the first QA pair, thereby obtaining the second data Select the second QA pair from the group MQA.

In FIG. 1, the information processing apparatus 100 selects QA pairs from the second data group MQA by processing the first processing pattern. The information processing device 100 compares the first question Q12 with each question in the second data group MQA, and selects a QA pair of questions similar to the first question Q12 among the questions in the second data group MQA as the second QA. Select as a pair. The information processing device 100 compares the vector obtained by vector-transforming the first question Q12 and the vector (question vector) obtained by vector-transforming each question in the second data group MQA. Then, the information processing apparatus 100 selects QA pairs of questions with question vectors similar to the vector of the first question Q12 from the second data group MQA as second QA pairs. In FIG. 1, in order to simplify the explanation, the information processing apparatus 100 selects the QA pair of the question whose question vector is most similar to the vector of the first question Q12 from among the second data group MQA as the second QA pair. .

The information processing device 100 converts each question in the second data group MQA into a question vector using a vector conversion model. The QA pair information (questions, etc.) in the second data group MQA may be converted into vectors in advance. The information processing apparatus 100 calculates the cosine similarity between the vector of the first question Q12 and the question vector of each question in the second data group MQA, and selects the QA pair of the question of the question vector having the highest cosine similarity as the first QA pair. Select as 2QA pair. In FIG. 1, the information processing apparatus 100 selects the QA pair P21, which is the combination of the question Q21 and the response A21, as the second QA pair. Then, the information processing apparatus 100 selects the response A21 of the QA pair P21 selected as the second QA pair as the target response of the search query QE1. In FIG. 1, the information processing apparatus 100 selects the response A21, which is the character information "manipulative treatment due to an accident will be compensated", as the target response of the search query QE1. For example, the information processing apparatus 100 provides the selected response A21 to the user.

[1-1-1. Background and effects, etc.]
In the conventional automatic response system, QA pairs are created by people, so the coverage rate such as the utterance coverage rate for queries is low. Therefore, it is possible to increase these QA pairs by using response logs such as past manned chat logs. However, there is a possibility that personal information and context-dependent information may be included in responses such as answer sentences in that case, and information that should not be output to customers, etc., may be output. There is a possibility that it will not be possible to respond appropriately due to such reasons.

Therefore, it is conceivable for the system to filter this personal information and context-dependent information by creating regular expressions and dictionaries, but the problem is that creating rules for this is extremely costly and time-consuming.

On the other hand, in the information processing system 1 described above, QA pairs created manually and satisfying response conditions such as personal information and context dependence are combined with QA pairs generated based on response logs such as manned chat logs. As a result, the information processing system 1 improves the coverage rate by using the Q sentence (question) of the log QA sentence, and suppresses the possibility that personal information and context-dependent information will be output. Quality can be guaranteed.

For example, in the information processing system 1, a first data group LQA, which is log QA generated from a past response history (response log), and a second data group MQA, which does not include responses corresponding to personal information, context dependence, etc. Use As a result, the information processing system 1 uses the questions of the first data group LQA to improve the coverage rate for queries, and uses the responses of the second data group MQA to improve the output of personal information and context-dependent information. can reduce the possibility of Thus, the information processing system 1 can select an appropriate response to the search query. Note that the second data group MQA is not limited to manual creation, and may be automatically created without human intervention as long as it can be created so as to include only QA pairs that satisfy conditions regarding responses.

[1-2. Configuration of information processing system according to embodiment]
The information processing system 1 shown in FIG. 2 will be described. As shown in FIG. 2 , the information processing system 1 includes a terminal device 10 and an information processing device 100 . The terminal device 10 and the information processing device 100 are communicably connected by wire or wirelessly via a predetermined communication network (network N). FIG. 2 is a diagram illustrating a configuration example of an information processing system according to the embodiment; Note that the information processing system 1 shown in FIG. 2 may include a plurality of terminal devices 10 and a plurality of information processing apparatuses 100 .

The information processing apparatus 100 selects a first QA pair corresponding to the search query from a first data group including a plurality of QA pairs, and based on the first QA pair, from the second data group, response information corresponding to the search query. as the target response information. Also, the information processing device 100 is a computer that transmits various types of information to the terminal device 10 . The information processing device 100 is an information processing device that is used to provide various services such as response assistance to an operator.

The terminal device 10 is a computer used by the user. For example, the terminal device 10 receives an input of a search query by the user. The terminal device 10 transmits the search query to the information processing device such as the information processing device 100 . In addition, the terminal device 10 may accept an input of a search query by voice.

The terminal device 10 is a device used by the user. The terminal device 10 receives input from the user. The terminal device 10 receives voice input by user's utterance and input by user's operation. The terminal device 10 displays information according to the user's input. The terminal device 10 may be any device as long as it can implement the processing in the embodiments. For example, the terminal device 10 is a device such as a smart phone, a smart speaker, a television, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. may

[1-3. Configuration of information processing device according to embodiment]
Next, the configuration of the information processing apparatus 100, which is an example of an information processing apparatus that executes information processing according to the embodiment, will be described. FIG. 3 is a diagram illustrating a configuration example of an information processing device according to an embodiment of the present disclosure.

As shown in FIG. 3, the information processing device 100 has a communication section 110, a storage section 120, and a control section . The information processing apparatus 100 includes an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from an administrator of the information processing apparatus 100, and a display unit (for example, a liquid crystal display, etc.) for displaying various information. may have.

The communication unit 110 is implemented by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to the network N (see FIG. 2) by wire or wirelessly, and transmits and receives information to and from another information processing device such as the terminal device 10 . Also, the communication unit 110 may transmit and receive information to and from a device used by a customer (such as the device DV in FIG. 24).

The storage unit 120 is implemented by, for example, a semiconductor memory device such as RAM (Random Access Memory) or flash memory, or a storage device such as a hard disk or optical disk. The storage unit 120 according to the embodiment has a response log storage unit 121, a log QA storage unit 122, a manual QA storage unit 123, and a content information storage unit 124, as shown in FIG.

The storage unit 120 stores various information other than the above. The storage unit 120 stores information of various applications (programs) that implement response selection processing, display of selected information, and the like. For example, the information processing apparatus 100 can execute an application for providing response candidates by activating an application for presenting response candidates (also referred to as a "response candidate providing application").

The response log storage unit 121 according to the embodiment stores information about response logs. For example, the response log storage unit 121 stores a history of past conversations (communications) between an operator and a customer. FIG. 4 is a diagram illustrating an example of a response log storage unit according to an embodiment of the present disclosure; FIG. 4 shows an example of the response log storage unit 121 according to the embodiment. In the example shown in FIG. 4, the response log storage unit 121 includes items such as "Operator", "Dialogue", and "Customer".

"Operator" indicates identification information for identifying the operator who performed the corresponding dialogue. FIG. 4 shows an example in which an operator ID such as "OP1" for identifying an operator is stored in "Operator", but a name or the like may be used as long as the operator can be identified.

"Dialogue" indicates the content of the dialogue. FIG. 4 shows a case where "dialog" stores character information indicating who spoke what and when. Any information may be stored in "dialog" as long as the contents of the dialogue can be grasped.

"Customer" indicates identification information for identifying the customer who has conducted the corresponding dialogue. FIG. 4 shows abstract information indicating a customer such as "Mr. A", but "Customer" stores information that can identify the customer, such as customer ID and name.

In FIG. 4, the response log storage unit 121 stores information when the operator (operator OP1) identified by the operator "OP1" interacted with the customer "Mr. A". For example, the utterance “This is ABC Insurance. For example, the utterance "I'm looking at the website now, but I don't understand the application procedure" was made by C (customer) at 12:00:10 on November 25, 2020. indicates

It should be noted that the response log storage unit 121 may store various types of information, not limited to the above, depending on the purpose. In this way, the response log stored in the response log storage unit includes information such as when, who said what (input). For example, the information processing apparatus 100 processes the response log stored in the response log storage unit 121 to generate log QA as shown in the log QA storage unit 122 . Note that the log QA shown in the log QA storage unit 122 may be performed by a device (information generation device) other than the information processing device 100 . In this case, the information processing device 100 receives log QA from the information generating device and stores the received log QA in the log QA storage unit 122 .

The log QA storage unit 122 according to the embodiment stores various information related to log QA, which is an example of the first data group. The log QA storage unit 122 stores various information about QA pairs generated from response logs. FIG. 5 is a diagram illustrating an example of a log QA storage unit according to an embodiment of the present disclosure; FIG. 5 shows an example of the log QA storage unit 122 according to the embodiment. In the example shown in FIG. 5, the log QA storage unit 122 includes items such as “Q (question)” and “A (answer)”.

In FIG. 5, the log QA storage unit 122 stores a Q (question) such as "I'm looking at the website now, but I don't understand the application procedure." A QA pair or the like associated with A (answer) such as "Can I have it?" is stored. Each QA pair stored in the log QA storage unit 122 is generated by processing the response log. For example, the information processing apparatus 100 analyzes each sentence (character information) in the response log and generates a QA pair from the response log based on the analysis result. For example, the information processing apparatus 100 extracts a question sentence and a response sentence from the response log by analyzing the response log appropriately using techniques related to morphological analysis, semantic analysis, etc., and extracts the extracted question sentence and the response. and sentences may be used as a QA pair to generate a log QA (first data group). The above is only an example, and the process of processing the response log to generate the log QA (first data group) can be any kind of process if the log QA (first data group) can be generated from the response log. It may be processing.

"Q (question)" indicates a question. "A (answer)" indicates a response (answer). It should be noted that the log QA storage unit 122 may store various types of information, not limited to the above, depending on the purpose. For example, the log QA storage unit 122 may store a QA pair ID that identifies each QA pair. FIG. 5 shows a case where personal information such as "Mr. A" is included in the response in the log QA storage unit 122 . Thus, the first data group can include personal information and the like.

The manual QA storage unit 123 according to the embodiment stores various information related to manual QA, which is an example of the second data group. The manual QA storage unit 123 stores various types of information regarding manually generated QA pairs. FIG. 6 is a diagram illustrating an example of a manual QA storage unit according to the embodiment; The manual QA storage unit 123 shown in FIG. 6 includes items such as "category", "Q (question)", and "A (answer example)".

In FIG. 6, the manual QA storage unit 123 stores, for the category "compensation details", a Q (question) such as "Will the cost of treatment in the event of a traffic accident be compensated?" is stored.

"Category" indicates the category to which the corresponding QA pair applies. "Q (question)" indicates a question. "A (answer)" indicates a response (answer). Note that the manual QA storage unit 123 may store various types of information, not limited to the above, depending on the purpose. For example, the manual QA storage unit 123 may store a QA pair ID that identifies each QA pair.

The content information storage unit 124 according to the embodiment stores various types of information regarding content displayed on the terminal device 10 . For example, the content information storage unit 124 stores information about content displayed by an application (also referred to as an “app”) installed in the terminal device 10 . Note that the above is merely an example, and the content information storage unit 124 may store various types of information according to the content for which response candidates are displayed. The content information storage unit 124 stores various kinds of information necessary for providing content to the terminal device 10, displaying response candidates on the terminal device 10, and the like.

Return to Figure 3 and continue the explanation. The control unit 130, for example, uses a CPU (Central Processing Unit), an MPU (Micro Processing Unit), etc. to store programs stored inside the information processing apparatus 100 (for example, an information processing program according to the present disclosure, etc.) as RAM (Random Access Memory) or the like as a work area. Also, the control unit 130 is implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the control unit 130 has an acquisition unit 131, a conversion unit 132, a first selection unit 133, a second selection unit 134, a generation unit 135, and a transmission unit 136. implements or performs the information processing functions and operations described in . Note that the internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be another configuration as long as it performs information processing described later. Moreover, the connection relationship between the processing units of the control unit 130 is not limited to the connection relationship shown in FIG. 3, and may be another connection relationship.

The acquisition unit 131 acquires various types of information. The acquisition unit 131 acquires various types of information from an external information processing device such as the terminal device 10 . The acquisition unit 131 acquires various types of information from the terminal device 10 as information input to the terminal device 10 . The acquisition unit 131 acquires from the terminal device 10 the search query input to the terminal device 10 . Acquisition unit 131 acquires various types of information from storage unit 120 .

The conversion unit 132 performs conversion processing. The conversion unit 132 converts the search query into a format that the first selection unit 133 uses for searching. The conversion unit 132 converts the QA pair information into a format that the second selection unit 134 uses for searching. For example, the conversion unit 132 converts character information into vectors.

The conversion unit 132 takes text (character information) as an input and converts the character information into a vector using a vector conversion model that outputs a vector (embedded expression) representing the text. The conversion unit 132 converts the text into a vector using any vectorization method. For example, the conversion unit 132 converts the text into a vector using BoW (Bag of Words), BERT (Bidirectional Encoder Representations from Transformers), or the like. Also, the conversion unit 132 may perform vector conversion processing using a technique used in the field of machine learning or natural language processing. Various models may be used as the vector conversion model. For example, any network configuration such as DNN (Deep Neural Network) can be adopted for the vector conversion model. For example, the vector transformation model may be a model having a network configuration of Transformers other than BERT. For example, the vector conversion model may be CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), LSTM (Long short-term memory), or the like. That is, the vector conversion model can employ any configuration as long as input information can be converted into a vector.

The conversion unit 132 converts the search query QE1 into a query vector using a vector conversion model. The conversion unit 132 converts the QA pairs of the first data group LQA into vectors. For example, the conversion unit 132 converts each question in the first data group LQA into a question vector using a vector conversion model. For example, the conversion unit 132 converts each response in the first data group LQA into a response vector using a vector conversion model.

The conversion unit 132 converts the QA pairs of the second data group MQA into vectors. For example, the conversion unit 132 converts each question in the second data group MQA into a question vector using a vector conversion model. For example, the conversion unit 132 converts each response in the second data group MQA into a response vector using a vector conversion model.

The first selection unit 133 selects various information. For example, the first selection unit 133 selects various information based on information from other information processing devices such as the terminal device 10 . For example, the first selection unit 133 selects various information based on information stored in the storage unit 120 .

The first selection unit 133 selects various information based on the various information acquired by the acquisition unit 131 . The first selection unit 133 selects various information based on the information converted by the conversion unit 132 . The first selection unit 133 determines which of the first processing pattern, the second processing pattern, and the third processing pattern is used to select the first QA pair. For example, the first selection unit 133 determines which processing pattern is used to select the first QA pair according to settings or user designation. The first selection unit 133 selects the first QA pair according to the processing pattern specified by the user from among the first processing pattern, the second processing pattern, and the third processing pattern. The first selection unit 133 selects the first QA pair according to a preset processing pattern among the first processing pattern, the second processing pattern, and the third processing pattern.

The first selection unit 133 selects a first QA pair corresponding to a search query from a first data group including a plurality of QA pairs that are combinations of question information corresponding to responses and response information indicating the responses. The first selection unit 133 selects a first QA pair from a first data group including a plurality of QA pairs that are combinations of question information, which is text information, and response information. The first selection unit 133 selects a first QA pair from a first data group including a plurality of QA pairs that are combinations of question information, response information that is character information, and the like.

The first selection unit 133 selects the first QA pair from the first data group including multiple QA pairs based on the response log. The first selection unit 133 selects a first QA pair from a first data group including a plurality of QA pairs generated by processing response logs. The first selection unit 133 selects the first QA pair by searching the first data group using the search query. The first selection unit 133 selects the first QA pair based on the similarity between the question information of each QA pair in the first data group and the search query.

The first selection unit 133 uses the vector transformed by the transformation unit 132 to select the first QA pair. The first selection unit 133 selects, from the first data group, QA pairs of questions with question vectors similar to the query vector of the search query as first QA pairs. The first selection unit 133 calculates the cosine similarity between the query vector of the search query and the question vector of each question in the first data group, and selects the QA pair of the question of the question vector with the highest cosine similarity as the first QA. Select as a pair.

The second selection unit 134 selects various information. For example, the second selection unit 134 selects various information based on information from other information processing devices such as the terminal device 10 . For example, the second selection unit 134 selects various information based on information stored in the storage unit 120 .

The second selection unit 134 selects various information based on the various information acquired by the acquisition unit 131 . The second selection unit 134 selects various information based on the information converted by the conversion unit 132 . The second selection section 134 selects various information based on the information selected by the first selection section 133 . For example, the second selection unit 134 determines to select target response information with the same processing pattern as the first selection unit 133 . The second selection unit 134 selects target response information according to the processing pattern performed by the first selection unit 133 from among the first processing pattern, the second processing pattern, and the third processing pattern.

The second selection unit 134 selects response information corresponding to the search query from a second data group including a plurality of QA pairs different from the first data group, based on the first QA pair selected by the first selection unit 133. , as the target response information. The second selection unit 134 selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and response information that satisfies conditions regarding responses.

The second selection unit 134 selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and response information that does not include personal information. The second selection unit 134 selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and context-independent response information.

The second selection unit 134 selects the second QA pair based on the question. The second selection unit 134 determines a second QA pair corresponding to the question information of the first QA pair by searching the second data group using the question information of the first QA pair, and selects the second QA pair from the second data group. The response information of the 2QA pairs is selected as the target response information.

The second selection unit 134 selects the second QA pair based on the response. The second selection unit 134 selects target response information from the second data group by searching the second data group using the response information of the first QA pair.

The second selection unit 134 selects the second QA pair based on the combination of questions and responses. The second selection unit 134 determines the second QA pair corresponding to the first QA pair by searching the second data group using the first QA pair, and selects the response information of the second QA pair from the second data group. , as the target response information.

The second selection unit 134 uses the vector converted by the conversion unit 132 to select target response information. The second selection unit 134 selects, from the second data group MQA, QA pairs of questions whose question vectors are similar to the vectors of questions of the first QA pairs as second QA pairs. The second selection unit 134 calculates the cosine similarity between the vector of the question of the first QA pair and the question vector of each question in the second data group MQA, and the QA pair of the question of the question vector with the highest cosine similarity as the second QA pair. The second selection unit 134 selects the response of the QA pair selected as the second QA pair as the target response of the search query.

The generation unit 135 generates various types of information. The generation unit 135 generates various information based on the various information acquired by the acquisition unit 131 . The generator 135 generates information indicating the response selected by the second selector 134 .

The generation unit 135 generates display information indicating the target response information. The generation unit 135 generates display information for displaying a list of first response information, which is response information selected using only the first data group, and target response information. The generation unit 135 generates display information for displaying a list of the second response information, which is the response information selected using only the second data group, and the target response information. When there is a plurality of pieces of target response information selected by the second selection unit 134, the generating unit 135 generates display information for displaying the plurality of pieces of target response information side by side.

The generating unit 135 appropriately uses various techniques to generate various information such as screens (contents) to be provided to external information processing devices. The generation unit 135 generates a screen (content) and the like to be provided to the terminal device 10 . For example, the generation unit 135 generates a screen (content) and the like to be provided to the terminal device 10 based on information stored in the storage unit 120 . The generating unit 135 generates each content such as the content CT1. The generation unit 135 may generate a screen (content) or the like by any process as long as the screen (content) or the like to be provided to an external information processing apparatus can be generated. For example, the generation unit 135 generates a screen (content) to be provided to the terminal device 10 by appropriately using various techniques related to image generation, image processing, and the like. For example, the generation unit 135 generates a screen (content) to be provided to the terminal device 10 using various technologies such as Java (registered trademark) as appropriate. Note that the generation unit 135 may generate a screen (content) to be provided to the terminal device 10 based on CSS, JavaScript (registered trademark), or HTML format. Also, for example, the generation unit 135 may generate screens (contents) in various formats such as JPEG (Joint Photographic Experts Group), GIF (Graphics Interchange Format), and PNG (Portable Network Graphics).

The transmitting unit 136 functions as a providing unit that provides the display information generated by the generating unit 135. The transmission unit 136 transmits various information. The transmission unit 136 transmits various types of information to an external information processing device. The transmission unit 136 provides various types of information to an external information processing device. For example, the transmission unit 136 transmits various information to other information processing devices such as the terminal device 10 . The transmitter 136 provides information stored in the storage 120 . Transmitter 136 transmits information stored in storage 120 .

The transmission unit 136 provides various information based on information from other information processing devices such as the terminal device 10. The transmitting section 136 provides various information based on the information stored in the storage section 120 .

The transmission unit 136 transmits the information selected by the second selection unit 134. The transmitter 136 transmits the information generated by the generator 135 . The transmission unit 136 transmits the display information to the terminal device 10 . The transmission unit 136 transmits content to the terminal device 10 .

[1-3-1. Merge example]
Note that, for example, similar QA pairs may be merged and used. For example, similar QA pairs in log QA generated from response logs may be merged. This point will be described with reference to FIG. FIG. 7 is a diagram illustrating an example of merge processing. The merge processing shown in FIG. 7 is executed by the information processing apparatus 100. FIG. For example, the conversion unit 132 executes the merge processing shown in FIG.

The example shown in FIG. 7 shows an example of merging four QA pairs into one QA pair. The four QA pairs in FIG. 7 are also combinations of questions and their responses regarding screen operations. Thus, the four QA pairs in FIG. 7 are questions of similar content. In this way, multiple QA pairs may be merged into one if the content of multiple QA pairs overlaps.

For example, the degree of overlap of each QA pair may be calculated using general machine learning/natural language processing techniques such as cosine similarity of documents (character information) (BoW, BERT embedded expressions, etc.). For example, if the vectors of documents are equal to or larger than a certain threshold, they may be regarded as the same sentence. For example, a QA pair whose degree of overlap indicated by cosine similarity or the like is equal to or greater than a threshold is targeted for merging. For example, the information processing apparatus 100 calculates cosine similarity between questions of each QA pair, and merges QA pair groups whose cosine similarity is equal to or higher than a predetermined threshold into one QA pair. In addition, the information processing apparatus 100 calculates the cosine similarity between sentences (documents) that combine the questions and responses of each QA pair, and divides the QA pair group whose cosine similarity is equal to or higher than a predetermined threshold into one QA pair. May be merged.

In the example of FIG. 7, the cosine similarity between each of the four QA pairs is greater than or equal to the predetermined threshold, so the four QA pairs are merged. The information processing apparatus 100 performs a process of merging four QA pairs (step S10).

For example, the response (A sentence) of the QA pair after merging is determined based on the A sentences of the four QA pairs before merging. For example, the A sentence of the QA pair after merging is the frequently used A sentence in four QA pairs before merging. For example, the information processing apparatus 100 calculates cosine similarity between A sentences, and classifies them as the same A sentence if they are equal to or greater than a certain threshold. Then, the information processing apparatus 100 may randomly select one sentence from the cluster with the largest number of A sentences belonging to it, and use it as a post-merge response (A sentence).

The questions (Q sentences) of the QA pair after merging may also be determined in the same way as the responses (A sentences). For example, Q sentences of the QA pair after merging are frequently used in four QA pairs before merging. For example, the information processing apparatus 100 calculates the cosine similarity between Q sentences, and classifies them as the same Q sentences if they are equal to or greater than a certain threshold. Then, the information processing apparatus 100 may randomly select one sentence from the cluster with the largest number of Q sentences belonging to it, and use it as a post-merge question (Q sentence).

[1-4. Configuration of terminal device according to embodiment]
Next, the configuration of the terminal device 10, which is an example of an information processing device that executes information processing according to the embodiment, will be described. FIG. 8 is a diagram illustrating a configuration example of a terminal device according to an embodiment of the present disclosure.

As shown in FIG. 8, the terminal device 10 has a communication section 11, an input section 12, a display section 13, a storage section 14, a control section 15, and an audio output section 16.

The communication unit 11 is implemented by, for example, a NIC, a communication circuit, or the like. The communication unit 11 is connected to a predetermined communication network (network) by wire or wirelessly, and transmits and receives information to and from an external information processing device. For example, the communication unit 11 is connected to a predetermined communication network by wire or wirelessly, and transmits and receives information to and from the information processing device 100 .

The input unit 12 accepts inputs by various user operations. For example, the input unit 12 may receive various operations from a user such as an operator via a display surface (for example, the display unit 13) using a touch panel function. The input unit 12 may also receive various operations from buttons provided on the terminal device 10 or from a keyboard or mouse connected to the terminal device 10 . The input unit 12 may receive user input by voice via a microphone or the like. The input unit 12 receives various operations by user's speech.

The display unit 13 functions as a providing unit that provides information to the user who uses the terminal device 10 by displaying information. The display unit 13 is a display screen of a tablet terminal or the like realized by, for example, a liquid crystal display or an organic EL (Electro-Luminescence) display, and is a display device for displaying various information.

The display unit 13 displays applications. The display unit 13 displays content. The display unit 13 displays various information received from the information processing device 100 . The display unit 13 displays the content CT1 from the information processing device 100 .

The storage unit 14 is realized by, for example, a semiconductor memory device such as RAM (Random Access Memory) or flash memory, or a storage device such as a hard disk or optical disk. The storage unit 14 stores, for example, various information received from the information processing device 100 . The storage unit 14 stores, for example, information regarding an application (for example, an information output application, etc.) installed in the terminal device 10, such as a program.

The control unit 15 is a controller. For example, various programs stored in a storage device such as the storage unit 14 inside the terminal device 10 are executed by the CPU, MPU, or the like using the RAM as a work area. Realized. For example, the various programs include programs for applications that perform information processing (for example, information output applications). Also, the control unit 15 is a controller, and is implemented by an integrated circuit such as ASIC or FPGA, for example.

As shown in FIG. 6, the control unit 15 includes an acquisition unit 151, a transmission unit 152, a reception unit 153, and a processing unit 154, and implements or executes the information processing functions and actions described below. . Note that the internal configuration of the control unit 15 is not limited to the configuration shown in FIG. 6, and may be another configuration as long as it performs information processing described later. Further, the connection relationship between the processing units of the control unit 15 is not limited to the connection relationship shown in FIG. 6, and may be another connection relationship.

The acquisition unit 151 acquires various types of information. For example, the acquisition unit 151 acquires various types of information from an external information processing device. For example, the acquisition unit 151 stores the acquired various information in the storage unit 14 . Acquisition unit 151 acquires input information accepted by input unit 12 . Acquisition unit 151 acquires an input search query.

The transmission unit 152 transmits various information to an external information processing device via the communication unit 11. The transmission unit 152 transmits various types of information to the information processing device 100 . The transmission unit 152 transmits various information stored in the storage unit 14 to an external information processing device. The transmission unit 152 transmits various information acquired by the acquisition unit 151 to the information processing apparatus 100 . The transmitting unit 152 transmits the search query acquired by the acquiring unit 151 to the information processing device 100 . Transmitter 152 transmits input information accepted by input unit 12 to information processing apparatus 100 .

The receiving unit 153 receives information from the information processing device 100 via the communication unit 11 . The receiving unit 153 receives information provided by the information processing device 100 . The receiving unit 153 receives content from the information processing device 100 . The receiving unit 153 receives the content CT1.

The processing unit 154 executes various types of processing. The processing unit 154 displays various information via the display unit 13 . For example, the processing unit 154 controls display on the display unit 13 . The processing unit 154 outputs various kinds of information as voice through the voice output unit 16 . For example, the processing unit 154 controls audio output of the audio output unit 16 .

The processing unit 154 outputs the information received by the receiving unit 153. The processing unit 154 outputs content provided from the information processing device 100 . The processing unit 154 causes the content received by the receiving unit 153 to be displayed on the display unit 13 or output as audio by the audio output unit 16 . The processing unit 154 displays content via the display unit 13 . The processing unit 154 outputs the contents as audio through the audio output unit 16 .

Note that each process performed by the control unit 15 described above may be realized by, for example, JavaScript (registered trademark). Further, when processing such as information processing by the control unit 15 described above is performed by a predetermined application, each unit of the control unit 15 may be realized by the predetermined application, for example. For example, processing such as information processing by the control unit 15 may be realized by control information received from an external information processing device. For example, when the display process described above is performed by a predetermined application (for example, an information output application, etc.), the control unit 15 may have an application control unit that controls the predetermined application or a dedicated application, for example.

The audio output unit 16 is realized by a speaker that outputs audio, and is an output device for outputting various types of information as audio. The audio output unit 16 audio-outputs the content provided from the information processing apparatus 100 . For example, the audio output unit 16 outputs audio corresponding to information displayed on the display unit 13 .

[1-5. Information processing procedure according to the embodiment]
Next, various information processing procedures according to the embodiment will be described with reference to FIGS. 9 and 10. FIG.

[1-5-1. Procedure of processing related to information processing device]
First, using FIG. 9, the flow of processing related to the information processing apparatus according to the embodiment of the present disclosure will be described. FIG. 9 is a flow chart showing the processing procedure of the information processing device according to the embodiment of the present disclosure. Specifically, FIG. 9 is a flow chart showing the procedure of information processing by the information processing apparatus 100 .

As shown in FIG. 9, the information processing device 100 selects a first QA pair corresponding to a search query from a first data group including multiple QA pairs (step S101). Then, based on the selected first QA pair, the information processing apparatus 100 selects response information corresponding to the search query from the second data group including a plurality of QA pairs different from the first data group as target response information. Select (step S102).

[1-5-2. Procedure of processing related to information processing system]
Next, with reference to FIG. 10, the flow of processing related to the information processing system according to the embodiment of the present disclosure will be described. FIG. 10 is a sequence diagram showing processing procedures of the information processing system according to the embodiment of the present disclosure.

As shown in FIG. 10, the terminal device 10 accepts input of a search query by a user such as an operator (step S201). Then, the terminal device 10 transmits the search query to the information processing device 100 (step S202).

The information processing device 100 executes selection processing for selecting a target response corresponding to the search query acquired from the terminal device 10 (step S203). The information processing apparatus 100 selects a first QA pair corresponding to the search query from the first data group, and selects a target response corresponding to the search query from the second data group based on the selected first QA pair.

The information processing apparatus 100 generates display information (content) indicating the selected target response (step S204). Then, the information processing apparatus 100 provides the generated display information (step S205). The information processing device 100 transmits display information to the terminal device 10 . Then, the terminal device 10 displays the display information acquired from the information processing device 100 (step S206).

[1-6. System configuration overview]
Here, with reference to FIG. 11, a system configuration overview conceptually showing each function in the information processing system 1 will be described. FIG. 11 is a diagram showing an overview of the system configuration of the information processing system.

The information processing system 1 includes a UI section that receives a search query from the user U and provides the user U with an answer (response) corresponding to the search query. In the device configuration of the information processing system 1 shown in FIG. 2, for example, the terminal device 10 has the function of the UI section.

The information processing system 1 also includes a document vector conversion unit that, when acquiring text (character information), returns a vector (embedded expression) representing the text. For example, when the document vector conversion unit acquires the text "Hello", it outputs a vector such as "(0,0.2,0.1,0.4,-0.3,0.1,...)".

In the device configuration of the information processing system 1 shown in FIG. 2, for example, the information processing device 100 has the function of a document vector conversion unit. For example, the conversion unit 132 of the information processing apparatus 100 has the function of a document vector conversion unit.

The information processing system 1 includes a first QA (database) that stores log-based QA pairs. In the device configuration of the information processing system 1 shown in FIG. 2, for example, the information processing device 100 has the first QA (database). For example, the log QA storage unit 122 of the information processing device 100 corresponds to the first QA (database).

The information processing system 1 includes a second QA (database) that stores QA pairs based on manual creation. In the device configuration of the information processing system 1 shown in FIG. 2, for example, the information processing device 100 has a second QA (database). For example, the manual QA storage unit 123 of the information processing device 100 corresponds to the second QA (database).

In addition, the information processing system 1 uses the information of the first QA (database) and the second QA (database) and the function of the document vector conversion unit to generate an answer sentence (target response) corresponding to the search query received by the UI unit. includes a search part for searching (selecting) For example, the information processing device 100 has the function of a search unit. For example, the first selection unit 133 and the second selection unit 134 of the information processing device 100 have the function of the search unit.

As described above, if the information processing system 1 has the functions of the UI unit, the document vector conversion unit, the search unit, and the information of the first QA (database) and the second QA (database), the device configuration is shown in FIG. 2, any device configuration can be adopted. Other examples of the device configuration will be described later.

[1-7. Processing example]
From here, examples of various processes will be described with reference to FIGS. 12 to 16. FIG. Note that description of the same points as in FIG. 1 will be omitted as appropriate. In the flowchart below, processing will be described based on the system configuration example of the information processing system 1 shown in FIG. 11 . In the following, a case where the information processing system 1 performs processing will be described as an example. may be performed by any device capable of

[1-7-1. First processing pattern]
First, an example of the processing flow according to the first processing pattern shown in FIG. 1 will be described with reference to FIG. FIG. 12 is a flow chart showing the processing of the first processing pattern.

As shown in FIG. 12, the information processing system 1 converts the text of the search query into a vector (document vector) (step S301). For example, the information processing system 1 converts the character string of the search query into a vector (query vector).

The information processing system 1 searches for (the vector of) Q sentences of the log QA and nearby Q sentences by cosine similarity (step S302). For example, the information processing system 1, based on the cosine similarity between a vector of search query character strings (query vector) and a vector of log QA (first data group) questions (question vector), Select 1 QA pair.

The information processing system 1 searches Q sentences (vectors of) of manual QA for Q sentences near the searched Q sentence (vector of) (step S303). For example, the information processing system 1 selects the second QA pair based on the cosine similarity between the vector of the question of the first QA pair and the vector (question vector) obtained by vectorizing the question of manual QA (second data group). do.

The information processing system 1 returns the A sentence linked to the searched Q sentence to the UI unit (step S304). For example, the information processing system 1 provides the UI unit with the response of the selected second QA pair.

[1-7-2. Second processing pattern]
Patterns other than the first processing pattern will now be described. First, a processing pattern using response information when selecting a QA pair from the second data group MQA (hereinafter also referred to as “second processing pattern”) will be described with reference to FIG. FIG. 13 is a diagram showing an example of the second processing pattern. The same reference numerals are assigned to the same points as those in FIG. 1, and the description thereof is omitted as appropriate.

In FIG. 13, the information processing device 100 selects the QA pair P12, which is the combination of the question Q12 and the response A12, as the first QA pair for the search query QE1, as in step S1 of FIG.

Then, the information processing apparatus 100 selects a QA pair (second QA pair) from the second data group MQA using the second data group MQA and the first QA pair (step S12). In FIG. 13 , the information processing apparatus 100 searches the second data group MQA using the response A12 (also referred to as “first response A12”) of the QA pair P12 selected as the first QA pair, thereby obtaining the second data Select the second QA pair from the group MQA.

In FIG. 13, the information processing apparatus 100 selects QA pairs from the second data group MQA by the processing of the second processing pattern. The information processing apparatus 100 compares the first response A12 with each response in the second data group MQA, and selects a QA pair of responses similar to the first response A12 among the responses in the second data group MQA as the second QA. Select as a pair. The information processing apparatus 100 compares the vector obtained by vector-converting the first response A12 and the vector (response vector) obtained by vector-converting each response in the second data group MQA. Then, the information processing apparatus 100 selects, from the second data group MQA, a QA pair of responses of response vectors similar to the vector of the first response A12 as a second QA pair.

The information processing device 100 converts each response in the second data group MQA into a response vector using the vector conversion model. The QA pair information (responses, etc.) in the second data group MQA may be converted into vectors in advance. The information processing apparatus 100 calculates the cosine similarity between the vector of the first response A12 and the response vector of each response in the second data group MQA, and selects the response QA pair of the response vector with the highest cosine similarity as the first Select as 2QA pair. In FIG. 13, the information processing device 100 selects the QA pair P21, which is the combination of the question Q21 and the response A21, as the second QA pair.

Although FIG. 13 shows the case of selecting the second QA pair, the information processing apparatus 100 selects the response of the response vector similar to the vector of the first response A12 as the target response without selecting the second QA pair. You may Then, the information processing apparatus 100 selects the response A21 of the QA pair P21 selected as the second QA pair as the target response of the search query QE1.

Next, an example of the processing flow related to the second processing pattern shown in FIG. 3 will be described using FIG. FIG. 14 is a flow chart showing the processing of the second processing pattern.

As shown in FIG. 14, the information processing system 1 converts the text of the search query into a vector (step S401). For example, the information processing system 1 converts the character string of the search query into a vector (query vector).

The information processing system 1 searches Q sentences (vectors thereof) of the log QA and nearby Q sentences by cosine similarity (step S402). For example, the information processing system 1, based on the cosine similarity between a vector of search query character strings (query vector) and a vector of log QA (first data group) questions (question vector), Select 1 QA pair.

The information processing system 1 searches A sentences (vectors of) of manual QA for nearby A sentences for (vectors of) A sentences linked to the searched Q sentences (step S403). For example, the information processing system 1 selects the second QA pair based on the cosine similarity between the vector of the response of the first QA pair and the vector (response vector) obtained by vectorizing the response of manual QA (second data group) do.

The information processing system 1 returns the searched sentence A to the UI unit (step S404). For example, the information processing system 1 provides the UI unit with the response of the selected second QA pair.

[1-7-3. Third processing pattern]
Next, a processing pattern (hereinafter also referred to as “third processing pattern”) using QA pair information when selecting a QA pair from the second data group MQA will be described with reference to FIG. FIG. 15 is a diagram showing an example of the third processing pattern. For example, in the third processing pattern, the information processing apparatus 100 combines QA pair questions and responses. The same reference numerals are assigned to the same points as those in FIG. 1, and the description thereof is omitted as appropriate.

In FIG. 15, the information processing device 100 selects the QA pair P12, which is the combination of the question Q12 and the response A12, as the first QA pair for the search query QE1, as in step S1 of FIG.

Then, the information processing apparatus 100 selects a QA pair (second QA pair) from the second data group MQA using the second data group MQA and the first QA pair (step S22). In FIG. 15 , the information processing apparatus 100 searches the second data group MQA using the QA pair P12 (also referred to as “first QA pair P12”) selected as the first QA pair, thereby obtaining Select the second QA pair.

In FIG. 15, the information processing apparatus 100 selects a QA pair from the second data group MQA by the processing of the third processing pattern. The information processing apparatus 100 compares the first QA pair P12 with each QA pair in the second data group MQA, and selects a QA pair similar to the first QA pair P12 among the QA pairs in the second data group MQA as the second QA pair. Select as a pair. The information processing apparatus 100 compares the vector obtained by vector-transforming the first QA pair P12 and the vector (QA pair vector) obtained by vector-transforming each QA pair in the second data group MQA. For example, a QA pair vector (QA pair vector) is a vector that combines the question vector and the response vector. The QA pair vector may be a vector obtained by concatenating a question vector obtained by vectorizing the question of the QA pair and a response vector obtained by vectorizing the response of the QA pair. Then, the information processing apparatus 100 selects the QA pair of the QA pair vector similar to the vector of the first QA pair P12 from the second data group MQA as the second QA pair.

The information processing device 100 converts each QA pair in the second data group MQA into a QA pair vector using a vector conversion model. The QA pair information in the second data group MQA may be converted into vectors in advance. The information processing apparatus 100 calculates the cosine similarity between the vector of the first QA pair P12 and each QA pair vector in the second data group MQA, and selects the QA pair of the QA pair vector having the highest cosine similarity as the second QA pair. Select as In FIG. 15, the information processing device 100 selects the QA pair P21, which is the combination of the question Q21 and the response A21, as the second QA pair. Then, the information processing apparatus 100 selects the response A21 of the QA pair P21 selected as the second QA pair as the target response of the search query QE1.

Next, an example of the processing flow according to the third processing pattern shown in FIG. 3 will be described using FIG. FIG. 16 is a flow chart showing the processing of the third processing pattern.

As shown in FIG. 16, the information processing system 1 converts the text of the search query into a vector (step S501). For example, the information processing system 1 converts the character string of the search query into a vector (query vector).

The information processing system 1 searches for (the vector of) Q sentences of the log QA and neighboring Q sentences by cosine similarity (step S502). For example, the information processing system 1, based on the cosine similarity between a vector of search query character strings (query vector) and a vector of log QA (first data group) questions (question vector), Select 1 QA pair.

The information processing system 1 searches for (a vector of) a combined sentence (a vector of) the Q sentence and the A sentence of the log QA for a neighboring QA pair from (the vector of) the combined sentence of the Q sentence and the A sentence of the manual QA. (Step S503). For example, the information processing system 1 selects the second QA pair based on the cosine similarity between the vector of the first QA pair and the vector (QA pair vector) obtained by vectorizing each QA pair of manual QA (second data group). select.

The information processing system 1 returns the A sentence of the searched QA pair to the UI unit (step S504). For example, the information processing system 1 provides the UI unit with the response of the selected second QA pair.

[1-8. Adaptation example]
Application examples of the information processing system 1 described above will now be described with reference to FIGS. 17 to 19. FIG. 17 to 19 show an example applied to an automatic response system in a call center. FIG. 17 is a diagram showing an example of a first form regarding application of the system. Also, FIG. 18 is a diagram showing an example of a second mode of application of the system. Moreover, FIG. 19 is a figure which shows an example of the 3rd form regarding the application of a system. The application form of the information processing system 1 is not limited to the first to third forms shown in FIGS. 17 to 19, and any form can be adopted.

17 to 19, the side that provides the service of the automatic response system at the call center is called the service side SV, and the side that receives the service of the automatic response system at the call center is shown as the customer side CS. The chatbot BT having an automatic response function shown in FIGS. 17 to 19 has various functions used for automatic responses to customers, such as the function of selecting responses corresponding to search queries provided by the information processing apparatus 100 described above. have.

First, the first form will be described with reference to FIG. In the first form shown in FIG. 17, the chatbot BT basically communicates (dialogues) with customers through chat. In the first form shown in FIG. 17, the chatbot BT automatically responds when it is clear what the customer wants to ask, such as when the question can be solved by FAQ (Frequently Asked Questions) search.

On the other hand, if there are many OOD (Out-of-Domain) inputs whose intentions cannot be interpreted, and if the customer's wishes cannot be sorted out, the operator will respond. In this case, the chatbot BT displays, on the terminal device 10 used by the operator, response candidates to be returned next from the conversation history/contents. The operator confirms the conversation history and response candidates, and confirms the chatbot's response. In addition, when the content that the chatbot BT tried to respond to is inappropriate (if the content is NG), the operator returns a response directly to the customer.

Next, the second embodiment will be described using FIG. 18, but the description of the same points as in FIG. 17 will be omitted as appropriate. The second form shown in FIG. 18 shows a form in which information for assisting the operator is displayed by a real FAQ display, a navigation tool, or the like at the time of answering. In the second mode shown in FIG. 18, the operator communicates (dialogues) with the customer by voice or chat. In this case, the chatbot BT displays, on the terminal device 10 used by the operator, answer candidates/related FAQs to be returned next from the conversation history/contents. The operator confirms the conversation history and response candidates and interacts with the customer.

Next, the third embodiment will be described using FIG. 19, but the description of the same points as in FIG. 17 will be omitted as appropriate. The third form shown in FIG. 19 shows a form of semi-manned service provision. In the third form shown in FIG. 19, the chatbot BT basically communicates (dialogues) with customers through chat. The third form shown in FIG. 19 is the same as a manned chatbot in terms of customer experience.

The chatbot BT displays, on the terminal device 10 used by the operator, candidate responses to be returned next from the conversation history/contents. The operator confirms the conversation history and response candidates, and confirms the chatbot's response. In addition, when the content that the chatbot BT tried to respond to is inappropriate (if the content is NG), the operator returns a response directly to the customer.

[1-9. Example of service provision]
An example of service provision will now be described. A case of providing information to an operator in an automatic response system in a call center will be described below as an example.

[1-9-1. Display example]
First, various information and the like will be described based on the screen image provided to the operator shown in FIG. FIG. 20 is a diagram showing an example of display information. The information processing system 1 generates content CT1, which is an example of list information shown in FIG.

First, each component in the content CT1 in FIG. 20 will be described. An input field IN1 in the content CT1 is an area for receiving an input of a search query by a user such as an operator. For example, the operator inputs a character string to be searched for in the input field IN1, and the information processing system 1 receives the character string input by the operator as a search query.

First-type candidates FA1-FA4, second-type candidates SA1-SA4, and third-type candidates TA1-TA4 shown below the input field IN1 indicate response candidates corresponding to the search query input in the input field IN1. First-type candidates FA1, FA2, FA3, and FA4 indicate response candidates selected using only log QA (first data group). When describing the first type candidates FA1, FA2, FA3, FA4, etc. without distinguishing them, they may be described as "first type candidates FA". An answer candidate (first type candidate FA) selected using only the log QA (first data group) may be referred to as a first type candidate, a first type answer candidate, or the like. In FIG. 20, each first type candidate FA is indicated by a rectangular frame, and each first type candidate FA is character information indicating the content of the response.

For example, the information processing system 1 uses only the first data group LQA, searches the first data group LQA using the search query, and selects the QA pair response selected as the first type response candidate. Although FIG. 20 shows a case where four first-type candidate FAs are selected and displayed as a list, the number of first-type candidate FAs to be selected is not limited to four, and may be three or less. and may be 5 or more. The first-type candidates FA are arranged in descending order of similarity to the search query. For example, the order of arrangement of the first type candidates FA may be arranged in the order of neighbors at the time of nearest neighbor calculation.

For example, the information processing system 1 selects, from among the QA pairs of the first data group LQA, the responses of the QA pairs of questions whose similarity to the search query is equal to or greater than a predetermined threshold as the first type candidate FA. Then, the information processing system 1 generates display information including the first type candidate FA. That is, the first type candidate FA is a log QA response, and may include personal information and context-dependent content.

Second-type candidates SA1, SA2, SA3, and SA4 indicate response candidates selected using only manual QA (second data group). When the second-type candidates SA1, SA2, SA3, SA4, etc. are described without distinction, they may be described as "second-type candidates SA." A response candidate (second type candidate SA) selected using only manual QA (second data group) may be referred to as a second type candidate, a second type response candidate, or the like. In FIG. 20, each second-type candidate SA is indicated by a rectangular frame, and each second-type candidate SA is character information indicating the content of the response.

For example, the information processing system 1 uses only the second data group MQA, searches the second data group MQA using the search query, and selects the QA pair response selected as the second type of response candidate. Although FIG. 20 shows a case where four second-type candidate SAs are selected and displayed as a list, the number of second-type candidate SAs to be selected is not limited to four, and may be three or less. and may be 5 or more. The second-type candidate SAs are arranged in descending order of similarity to the search query. For example, the order of arrangement of the second type candidates SA may be the order of neighbors at the time of nearest neighbor calculation.

For example, the information processing system 1 selects, from among the QA pairs of the second data group MQA, the responses of the QA pairs of questions whose degree of similarity to the search query is equal to or greater than a predetermined threshold as the second type candidate SA. Then, the information processing system 1 generates display information including the second type candidate SA. That is, the second type candidate SA is a manual QA response and does not contain personal information or context-dependent content.

Third-type candidates TA1, TA2, TA3, and TA4 indicate response candidates selected using both the above-described log QA (first data group) and manual QA (second data group). When the third type candidates TA1, TA2, TA3, TA4, etc. are not distinguished, they may be described as "third type candidate TA". An answer candidate (third-type candidate TA) selected using only manual QA (second data group) may be referred to as a third-type candidate, a third-type answer candidate, or the like. In FIG. 20, each third-type candidate TA is indicated by a rectangular frame, and each third-type candidate TA is character information indicating the content of the response.

For example, the information processing system 1 uses both the first data group LQA and the second data group MQA. Specifically, the information processing system 1 searches the first data group LQA using the search query to select the first QA pair, and searches and selects the second data group MQA using the selected first QA pair. The response of the second QA pair obtained is selected as a response candidate of the third type.

Although FIG. 20 shows a case where four third-type candidate TAs are selected and displayed as a list, the number of selected third-type candidate TAs is not limited to four, and may be three or less. and may be 5 or more. The third-type candidate TAs are arranged in descending order of similarity to the search query. For example, the third-type candidate TAs may be arranged in the order of neighbors at the time of nearest neighbor calculation.

For example, the information processing system 1, out of the QA pairs of the second data group MQA, the response of the QA pair of the question whose similarity with the first QA pair selected from the first data group LQA is equal to or higher than a predetermined threshold, Select as 3 types of candidate TAs. Then, the information processing system 1 generates display information including the third type candidate TA. That is, the third type candidate TA is a manual QA response and does not contain personal information or context-dependent content.

Note that FIG. 20 is merely an example, and the arrangement of the first-type candidates FA, second-type candidates SA, and third-type candidates TA may be arbitrary. For example, the third type candidate TA, the second type candidate SA, and the first type candidate FA may be arranged in this order from the left. Further, for example, the information processing system 1 may allow the user to select the first type candidate FA, the second type candidate SA, and the third type candidate TA from a pull-down or the like, and display the response candidates of the selected type.

[1-9-2. Application configuration example]
Next, an application configuration example will be described with reference to FIGS. 21 and 22. FIG. Various applications shown below are provided by the information processing system 1 . FIG. 21 is a diagram showing an example of a first form regarding service provision. For example, FIG. 21 shows a case where a function of providing response candidates is implemented as an application separate from the customer response application. The customer response application referred to here is a system for communicating directly with a customer (customer). Also, FIG. 22 is a diagram showing an example of a second mode of service provision. For example, FIG. 22 shows a case where a function of providing response candidates is implemented as the same application as the customer service application. The terminal device 10 shown in FIGS. 21 and 22 is used by an operator who provides customer service. 20 are assigned the same reference numerals, and the description thereof is omitted as appropriate.

First, the example of FIG. 21 will be described. As described above, FIG. 21 shows a case where the first application AP1, which is an application that provides response candidates to the operator, and the second application AP2, which is a customer response application, are separate applications. The content displayed in the first application AP1 is the same as the content CT1 in FIG. Note that FIG. 21 shows each of the first type candidate FA, the second type candidate SA, and the third type candidate TA in such a manner that a reference numeral is written in each rectangular frame.

In the second application AP2, text information of the dialogue between the operator and the customer is displayed in chronological order, similar to the so-called message application. FIG. 21 shows a state in which a character string TX1 input by the operator and a subsequent character string TX2 input by the customer are displayed. The operator inputs a desired character string or the like in the input field IN2 for inputting a message, and designates (eg, clicks) a button BT1 labeled "send" to transmit the character string (message) input in the input field IN2. to the customer. Since these points are the same as those of a general message application, detailed description thereof will be omitted.

From here, an operation example of the operator in FIG. 21 will be described. For example, the operator copies the character string TX2, which is the question or the like entered by the customer, and pastes it into the input field IN1.

In response to the input of the character string TX2 into the input field IN1, the first application AP1 displays response candidates corresponding to the character string TX2. In FIG. 21, the first application AP1 executes search (selection) processing for each of the first to third type candidates using the character string TX2 as a search query. Then, the first application AP1 displays each search (selection) result as first type candidates FA1 to FA4, second type candidates SA1 to SA4, and third type candidates TA1 to TA4 corresponding to the character string TX2.

After confirming each of the displayed first type candidates FA1 to FA4, second type candidates SA1 to SA4, and third type candidates TA1 to TA4, the operator selects the most appropriate response to the character string TX2. to select. For example, the operator copies the most suitable response candidate among the displayed first type candidates FA1 to FA4, second type candidates SA1 to SA4, and third type candidates TA1 to TA4, and fills in the input field IN2. Paste (paste). Then, the operator designates (clicks, etc.) the button BT1 to transmit to the customer a response candidate that seems to be the most appropriate.

Next, the example of FIG. 22 will be described. As described above, FIG. 22 shows the case where the third application AP3 is implemented as a single application that provides the operator with candidate responses and the customer response application. The contents displayed in the third application AP3 are the same as those in FIG. 21 except that the first application AP1 and the second application AP2 in FIG. 21 are displayed as one application and the input field IN1 is not included.

The third application AP3 automatically displays (presents) answer candidates in response to acquisition of an input sentence (corresponding to the character string TX2) such as a question entered by the customer. Then, the third application AP3 allows the operator to specify (click, etc.) the response candidate among the displayed first type candidates FA1 to FA4, second type candidates SA1 to SA4, and third type candidates TA1 to TA4. Paste (stick) to IN2. In this way, in the third application AP3, response candidates are automatically presented from the customer's input text, and the operator designates (by clicking, etc.) an arbitrary response (response) candidate from among them, thereby providing a response to the customer. It becomes possible to use it as a (answer sentence).

The processing mode of the third application AP3 is not limited to the example shown in FIG. 22, and any processing mode can be adopted. This point will be described with reference to FIG. FIG. 23 is a diagram showing an example of display on the operator side.

In the example of FIG. 23, the third application AP3 has the highest score (similarity, etc.) among the selected first type candidates FA1 to FA4, second type candidates SA1 to SA4, and third type candidates TA1 to TA4. A response candidate (hereinafter also referred to as "optimal response") is displayed so as to be distinguishable from other response candidates. In FIG. 23, the third application AP3 arranges the third type candidate TA1, which is the optimum response, at the top, and arranges the other response candidates below the third type candidate TA1. The third application AP3 also arranges and displays the third type candidate TA1, which is the optimum response, in the input field IN2.

For example, if the accuracy of the selection of response candidates is sufficiently high, the response candidate with the highest score may be automatically reflected as the response. Then, the operator only needs to make a final confirmation. For example, the third application AP3 may return the selected optimal response to the customer even if the operator does not explicitly specify (click, etc.).

A display example of the application AP11, which is a customer response application on the customer side, will be described using FIG. FIG. 24 is a diagram showing an example of display on the customer side. A device DV shown in FIG. 24 is, for example, a terminal device such as a smart phone used by a customer. As shown in FIG. 24, the display of the application AP11 is similar to that of a normal message application without displaying the information corresponding to the content CT1 shown in FIG. Example Since the display of the application AP11 is for the customer, the arrangement of the icon representing the operator and the icon representing the customer are reversed.

It should be noted that the response to the customer may be selected from the second type candidates SA or the third type candidates TA, and may not be selected from the first type candidates FA. As a result, the information processing system 1 can prevent information including personal information and context-dependent content from being provided to customers.

Also, even if the optimal response described above is automatically responded, other response candidates may be presented to the operator. This point will be described with reference to FIG. FIG. 25 is a diagram showing an example of a third mode of service provision. The example of FIG. 25 shows a case where second-type candidates SA and third-type candidates TA are to be displayed, and first-type candidates FA are excluded.

In FIG. 25, when the second application AP2 returns the second-type candidate SA1, which is the optimum response among the second-type candidates SA1-SA4 and the third-type candidates TA1-TA4, to the customer as a response to the character string TX2. indicates Then, the second application AP2 selects a predetermined number of response candidates (also referred to as "alternative response candidates") from among the response candidates other than the second-type candidate SA1, starting with the highest score, under the notation "maybe". indicate. FIG. 25 shows a case where three of the second-type candidate SA2, the third-type candidate TA1, and the third-type candidate TA2 are displayed as alternative response candidates.

　The operator who has confirmed the alternative response candidates designates (clicks, etc.) the alternative response candidate if there is one that seems more appropriate than the optimal response. In this case, the information processing system 1 feeds back information on the designated alternative response candidates. For example, the information processing system 1 uses a vector conversion model or Update similarity calculation functions.

[1-10. Example of domain expansion]
An example of domain expansion will now be described. The selection process described above can be expanded to other domains (other industries, etc.). The concept of this point will be explained using FIG. FIG. 26 is a diagram illustrating an example of domain development.

FIG. 26 shows an example of expanding information in ABC General Insurance, which is the domain "non-life insurance", to other domains such as "real estate", "life insurance", "bank" and "food". There is both log QA and manual QA for ABC General Insurance, which is the domain General Insurance.

For ABC real estate, which is the domain "real estate", there is no log QA for that domain, and there is manual QA. In this case, the information processing system 1 uses the log QA of ABC insurance with the domain "non-life insurance" as the log QA of ABC real estate with the domain "real estate". As a result, the information processing system 1 can also process ABC real estate, which is the domain "real estate" for which there is no log QA.

ABC Life Insurance, which is the domain "life insurance", shows a pattern in which there is neither log QA nor manual QA for that domain. In this case, the information processing system 1 uses the log QA of ABC insurance with the domain "non-life insurance" as the log QA of ABC life insurance with the domain "life insurance". Also, manual QA for ABC life insurance whose domain is "life insurance" is manually created by a process described later. As a result, the information processing system 1 can also process ABC life insurance, which is the domain "life insurance" for which there is no log QA.

For ABC Bank, which is the domain "bank", it shows a pattern in which there is log QA for that domain and no manual QA. In this case, the manual QA for ABC Bank with the domain "bank" is manually created by the process described below. As a result, the information processing system 1 can also process ABC Bank, which is the domain "bank" for which there is no log QA.

Also, domain expansion can be applied without relying on similarities between domains. For example, for ABC Foods, which is a domain "food" with low similarity to the domain "non-life insurance", there is no log QA for that domain, but there is manual QA. In this case, the information processing system 1 uses the log QA of ABC General Insurance, which is the domain "general insurance", as the log QA of ABC Foods, which is the domain "food". As a result, the information processing system 1 can also process ABC foods, which is the domain "food" for which there is no log QA. Thus, the information processing system 1 can be deployed to any domain regardless of the similarity of the domains.

[1-11. Manual QA generation example (human-in-the-loop)]
An example of manual QA generation will now be described. In the following, as an example of manual generation, a case of generating manual QA by human-in-the-loop shown in FIG. 27 will be described as an example. FIG. 27 is a diagram illustrating an example of manual QA generation. FIG. 27 is a diagram showing the concept of processing when generating manual QA by human-in-the-loop. In FIG. 27, the first data group LQA and the second data group MQA shown in FIG. 1 for explanation are shown as examples of log QA and manual QA, but log QA and manual QA are shown in FIG. It is not limited to things. For example, a new manual QA may be generated by performing the human-in-the-loop process shown in FIG. 27 from a state in which there is no manual QA.

First, in human-in-the-loop, combinations of QA pairs that are included in log QA and not included in manual QA are extracted. For example, the process of extracting a combination of QA pairs is performed by (the generation unit 135 of) the information processing apparatus 100 .

In FIG. 27, the information processing system 1 compares each QA pair in the first data group LQA with each QA pair in the second data group MQA, and based on the comparison result, the data included in the first data group LQA. QA pairs that are included in the second data group MQA and are not included in the second data group MQA are extracted (step S11). For example, the information processing system 1 uses the cosine distance between questions (Q sentences), and if the distance (similarity) is greater than or equal to a threshold, the QA pairs are considered to have different contents. Then, the information processing system 1 extracts, among the QA pairs in the first data group LQA, QA pairs that are separated from all QA pairs in the second data group MQA by a distance (similarity) equal to or greater than a threshold.

In FIG. 27, the information processing system 1 converts a QA pair P13, which is a combination of a question Q13 and an answer A13, from the first data group LQA into a QA pair ("uncorresponding pair ”). That is, FIG. 27 shows a case where the QA pair whose content corresponds to the QA pair P13 is not included in the second data group MQA, which is manual QA.

Then, the information processing system 1 presents the extracted uncorresponding pairs to the user U (step S12). The user U in FIG. 27 is assumed to be a person such as a supervisor (administrator) who has the authority to create a manual QA manual. For example, the display unit 13 of the terminal device 10 presents the unmatched pair to the user U. FIG. That is, the information processing system 1 presents the extracted unmatched pairs to the user U by displaying the extracted unmatched pairs on the terminal device 10 . In FIG. 27, the information processing system 1 presents to the user U the QA pair P13, which is the extracted uncorresponding pair. The user U expands the manual QA based on the presented contents.

After confirming the presented unsupported pair, the user U manually creates a QA pair corresponding to the unsupported pair and adds it to the manual QA (step S13). For example, if the unsupported pair includes a portion that does not satisfy the conditions regarding the response, such as personal information or context-dependent content, the user U corrects that portion. Then, the user U adds a QA pair (also referred to as an "additional QA pair") that has been modified so that it does not include personal information or context-dependent content, and has a response that satisfies the conditions regarding the response, to the manual QA. . In FIG. 27, the user U confirms the QA pair P13, which is an uncorresponding pair, manually creates an addition QA pair corresponding to the content of the QA pair P13, and adds it to the second data group MQA. If the response A13 of the QA pair P13 does not contain personal information or context-dependent content and satisfies the response conditions, the user U adds the QA pair P13 to the second data group MQA as a QA pair for addition. may

In the human-in-the-loop process described above, for example, the process of generating manual QA is repeated until there are no more unsupported pairs. Further, when a new QA pair is added to the log QA and the QA pair corresponds to an unsupported QA pair, the information processing system 1 notifies the user U that the unsupported QA pair has been added as a new QA pair. May notify and prompt manual QA updates. Note that the above is merely an example, and manual QA may be generated by any method as long as manual QA can be generated.

[1-12. Information type example]
In the above example, the type of question and response information included in the QA pair is text, but if the processing can be applied, it is not limited to text and can be applied to various types of information. good too. In the above example, questions (Q) and responses (A) are text (characters). , voice, etc., can also be used. This point will be described below. An example in which the type of response is other than text will be described below.

[1-12-1. image]
First, a case where the response type is an image will be described with reference to FIG. FIG. 28 is a diagram showing an example of a case where an image is used as a response. In this case, the information processing system 1 selects an image corresponding to the search query (character) as a response. For example, when the user inputs a character string "Something sad happened today", the information processing system 1 performs selection processing using the character string as a search query. For example, the information processing system 1 selects a flower image as a response to the search query "Something sad happened today." For example, the information processing system 1 selects an image to respond to based on the similarity between the search query and the question of the QA pair. In the search process, the information processing system 1 may calculate the degree of similarity between images, which is the type of response, and select an image to respond to based on the degree of similarity.

[1-12-2. audio]
Next, a case where the response type is voice will be described with reference to FIG. FIG. 29 is a diagram showing an example of a case where a voice response is made. In this case, the information processing system 1 selects the voice corresponding to the search query (character) as a response. For example, when the user inputs a character string "Something sad happened today", the information processing system 1 performs selection processing using the character string as a search query. For example, the information processing system 1 selects, as a response to the search query "Something sad happened today", a soft voice saying "Are you okay?". For example, the information processing system 1 selects a response voice based on the similarity between the search query and QA pair questions. In the search process, the information processing system 1 may calculate the degree of similarity between sounds, which is the type of response, and select the sound to respond to based on the degree of similarity.

[2. Other embodiments]
The processes according to the above-described embodiments may be implemented in various different forms (modifications) other than the above-described embodiments and modifications.

[2-1. Other configuration examples]
Although the above example shows the case where the information processing device 100 and the terminal device 10 are separate units, these devices may be integrated. That is, a device (information processing device 100 or the like) that selects a response to a search query and a device (terminal device 10 or the like) that displays the selected response may be integrated. In other words, the terminal device 10 used by a user such as an operator may be an information processing device that selects a response to a search query.

In this case, the information processing system 1 includes a terminal device 10 that also functions as an information processing device that selects a response to a search query, and a content providing device (server device) that provides content to the terminal device 10. good too.

The configuration described above is merely an example, and the information processing system 1 may have any device configuration as long as it can select a response to a search query and provide that information.

[2-2. others]
Further, among the processes described in each of the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed manually. can also be performed automatically by known methods. In addition, information including processing procedures, specific names, various data and parameters shown in the above documents and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each drawing is not limited to the illustrated information.

Also, each component of each device illustrated is functionally conceptual and does not necessarily need to be physically configured as illustrated. In other words, the specific form of distribution and integration of each device is not limited to the one shown in the figure, and all or part of them can be functionally or physically distributed and integrated in arbitrary units according to various loads and usage conditions. Can be integrated and configured.

In addition, the above-described embodiments and modifications can be appropriately combined within a range that does not contradict the processing content.

In addition, the effects described in this specification are only examples and are not limited, and other effects may be provided.

[3. Effects of the Present Disclosure]
As described above, the information processing device according to the present disclosure (for example, the information processing device 100 or the terminal device 10 in the embodiment) includes the first selection unit (first selection unit 133 in the embodiment), the second selection unit ( In the embodiment, a second selection unit 134) is provided. A first selection unit selects a first QA pair corresponding to a search query from a first data group including a plurality of QA pairs that are combinations of question information corresponding to responses and response information indicating the responses. The second selection unit selects the response information corresponding to the search query from the second data group including a plurality of QA pairs different from the first data group based on the first QA pair selected by the first selection unit. Select as response information.

Thus, the information processing apparatus according to the present disclosure can select an appropriate response to the search query by selecting the response corresponding to the search query in two steps.

Also, the first selection unit selects a first QA pair from a first data group including a plurality of QA pairs that are combinations of question information, which is character information, and response information. In this way, the information processing device can select a response for a search query of character information by selecting the first QA pair from the first data group in which the character information is question information. an appropriate response can be selected.

Also, the first selection unit selects a first QA pair from a first data group including a plurality of QA pairs that are combinations of question information, response information that is character information, and the like. In this way, the information processing device can select an appropriate response to the search query by selecting the first QA pair from the first data group having character information as response information.

Also, the first selection unit selects the first QA pair from the first data group including a plurality of QA pairs based on the response log. In this way, the information processing device can select a response based on past responses by selecting the first QA pair from the first data group generated based on the response log. An appropriate response can be selected for a search query.

Also, the first selection unit selects a first QA pair from a first data group including a plurality of QA pairs generated by processing the response log. In this way, the information processing device can select a response based on past responses by selecting the first QA pair from the first data group for processing the response log. an appropriate response can be selected.

Also, the second selection unit selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and response information that satisfies conditions regarding responses. In this way, by selecting a response from response information that satisfies the response-related conditions, the information processing apparatus can limit the responses that are actually output to those that satisfy the response-related conditions. can be selected.

Also, the second selection unit selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and response information that does not include personal information. In this way, the information processing apparatus can suppress the possibility of outputting personal information by selecting a response from response information that does not include personal information, and therefore selects an appropriate response to a search query. can do.

Also, the second selection unit selects target response information from a second data group including a plurality of QA pairs that are combinations of question information and context-independent response information. In this way, the information processing apparatus can suppress the possibility of the response depending on the context by selecting the response from the response information that does not depend on the context. can do.

Also, the first selection unit selects the first QA pair by searching the first data group using the search query. Thus, the information processing device can select an appropriate response to the search query by searching the first data group using the search query.

Also, the first selection unit selects the first QA pair based on the similarity between the question information of each QA pair in the first data group and the search query. In this way, the information processing device can select a QA pair that is a question similar to the search query, and therefore can select an appropriate response to the search query.

In addition, the second selection unit determines a second QA pair corresponding to the question information of the first QA pair by searching the second data group using the question information of the first QA pair, and from the second data group, Select the response information of the second QA pair as the target response information. In this way, the information processing device can select a response based on the similarity of questions, and therefore can select an appropriate response to the search query.

Also, the second selection unit selects target response information from the second data group by searching the second data group using the response information of the first QA pair. In this way, the information processing device can select a response based on the similarity of the responses, and therefore can select an appropriate response to the search query.

Further, the second selection unit uses the first QA pair to search the second data group to determine the second QA pair corresponding to the first QA pair, and from the second data group, the response information of the second QA pair is selected as the target response information. In this way, the information processing device can select a response based on the similarity of the combination of question and response, and therefore can select an appropriate response to the search query.

The information processing apparatus also includes a generating unit (generating unit 135 in the embodiment) that generates display information indicating target response information. In this way, the information processing apparatus can provide information on the selected response to the user by generating the display information indicating the target response information.

In addition, the generation unit generates display information for displaying a list of the first response information, which is the response information selected using only the first data group, and the target response information. In this way, the information processing apparatus generates display information that displays a list of responses selected by a plurality of different methods, so that the information on the responses selected by various methods can be provided in a state in which the user can compare them. Become.

In addition, the generation unit generates display information for displaying a list of the second response information, which is the response information selected using only the second data group, and the target response information. In this way, the information processing apparatus generates display information that displays a list of responses selected by a plurality of different methods, so that the information on the responses selected by various methods can be provided in a state in which the user can compare them. Become.

Also, when there is a plurality of pieces of target response information selected by the second selection unit, the generation unit generates display information for displaying the plurality of pieces of target response information side by side. In this manner, the information processing apparatus generates display information that displays a plurality of response candidates side by side, thereby providing a plurality of response candidates to the user in a state in which the user can select which response is preferable. .

The information processing apparatus also includes a providing unit (transmitting unit 136 in the embodiment) that provides display information generated by the generating unit. In this way, the information processing apparatus can provide the selected response information by providing the display information indicating the target response information.

[4. Hardware configuration]
The information processing apparatus (information equipment) such as the information processing apparatus 100 and the terminal apparatus 10 according to each of the embodiments described above is implemented by a computer 1000 configured as shown in FIG. 30, for example. FIG. 30 is a hardware configuration diagram showing an example of a computer 1000 that implements the functions of an information processing device. An information processing apparatus 100 according to an embodiment will be described below as an example. The computer 1000 has a CPU 1100 , a RAM 1200 , a ROM (Read Only Memory) 1300 , a HDD (Hard Disk Drive) 1400 , a communication interface 1500 and an input/output interface 1600 . Each part of computer 1000 is connected by bus 1050 .

The CPU 1100 operates based on programs stored in the ROM 1300 or HDD 1400 and controls each section. For example, the CPU 1100 loads programs stored in the ROM 1300 or HDD 1400 into the RAM 1200 and executes processes corresponding to various programs.

The ROM 1300 stores a boot program such as BIOS (Basic Input Output System) executed by the CPU 1100 when the computer 1000 is started, and programs dependent on the hardware of the computer 1000.

The HDD 1400 is a computer-readable recording medium that non-temporarily records programs executed by the CPU 1100 and data used by such programs. Specifically, HDD 1400 is a recording medium that records an information processing program according to the present disclosure, which is an example of program data 1450 .

A communication interface 1500 is an interface for connecting the computer 1000 to an external network 1550 (for example, the Internet). For example, CPU 1100 receives data from another device via communication interface 1500, and transmits data generated by CPU 1100 to another device.

The input/output interface 1600 is an interface for connecting the input/output device 1650 and the computer 1000 . For example, the CPU 1100 receives data from input devices such as a keyboard and mouse via the input/output interface 1600 . The CPU 1100 also transmits data to an output device such as a display, speaker, or printer via the input/output interface 1600 . Also, the input/output interface 1600 may function as a media interface for reading a program or the like recorded on a predetermined recording medium. Media include, for example, optical recording media such as DVD (Digital Versatile Disc) and PD (Phase change rewritable disk), magneto-optical recording media such as MO (Magneto-Optical disk), tape media, magnetic recording media, semiconductor memories, etc. is.

For example, when the computer 1000 functions as the information processing apparatus 100 according to the embodiment, the CPU 1100 of the computer 1000 implements the functions of the control unit 130 and the like by executing the information processing program loaded on the RAM 1200. The HDD 1400 also stores an information processing program according to the present disclosure and data in the storage unit 120 . Although CPU 1100 reads and executes program data 1450 from HDD 1400 , as another example, these programs may be obtained from another device via external network 1550 .

Note that the present technology can also take the following configuration.
(1)
A first selection unit that selects a first QA pair corresponding to a search query from a first data group including a plurality of QA pairs that are a combination of question information corresponding to a response and response information indicating the response;
Based on the first QA pair selected by the first selection unit, from a second data group containing a plurality of QA pairs different from the first data group, the response information corresponding to the search query, the target response a second selection unit for selecting as information;
Information processing device.
(2)
The first selection unit
The information processing apparatus according to (1), wherein the first QA pair is selected from the first data group including a plurality of QA pairs that are combinations of the question information, which is character information, and the response information.
(3)
The first selection unit
The information processing apparatus according to (1) or (2), wherein the first QA pair is selected from the first data group including a plurality of QA pairs that are combinations of the question information and the response information that is character information.
(4)
The first selection unit
The information processing apparatus according to any one of (1) to (3), wherein the first QA pair is selected from the first data group including a plurality of QA pairs based on response logs.
(5)
The first selection unit
The information processing apparatus according to (4), wherein the first QA pair is selected from the first data group including a plurality of QA pairs generated by processing a response log.
(6)
The second selection unit
Selecting the target response information from the second data group including a plurality of QA pairs that are combinations of the question information and the response information that satisfies a response-related condition (1) to (5). The information processing device described.
(7)
The second selection unit
The information processing apparatus according to (6), wherein the target response information is selected from the second data group including a plurality of QA pairs that are combinations of the question information and the response information that does not include personal information.
(8)
The second selection unit
The information processing apparatus according to (6) or (7), wherein the target response information is selected from the second data group including a plurality of QA pairs that are combinations of the question information and the context-independent response information.
(9)
The first selection unit
The information processing apparatus according to any one of (1) to (8), wherein the first QA pair is selected by searching the first data group using the search query.
(10)
The first selection unit
The information processing apparatus according to (9), wherein the first QA pair is selected based on the similarity between the question information of each QA pair in the first data group and the search query.
(11)
The second selection unit
determining a second QA pair corresponding to the question information of the first QA pair by searching the second data group using the question information of the first QA pair; The information processing device according to any one of (1) to (10), wherein the response information of 2QA pairs is selected as the target response information.
(12)
The second selection unit
Select the target response information from the second data group by searching the second data group using the response information of the first QA pair (1) to any one of (10) The information processing device described.
(13)
The second selection unit
Determine a second QA pair corresponding to the first QA pair by searching the second data group using the first QA pair, and from the second data group, the response information of the second QA pair, The information processing apparatus according to any one of (1) to (10), which is selected as the target response information.
(14)
a generation unit that generates display information indicating the target response information;
The information processing device according to any one of (1) to (13).
(15)
The generating unit
The information processing apparatus according to (14), wherein the display information for displaying a list of the first response information that is the response information selected using only the first data group and the target response information is generated.
(16)
The generating unit
The information according to (14) or (15), wherein the display information for displaying a list of the second response information, which is the response information selected using only the second data group, and the target response information is generated. processing equipment.
(17)
The generating unit
(14) to (16) according to any one of (14) to (16), wherein when there is a plurality of the target response information selected by the second selection unit, the display information for displaying the plurality of the target response information side by side is generated. Information processing equipment.
(18)
a providing unit that provides the display information generated by the generating unit;
The information processing device according to any one of (14) to (17).
(19)
Selecting a first QA pair corresponding to the search query from a first data group containing a plurality of QA pairs that are a combination of question information corresponding to the response and response information indicating the response,
executing a process of selecting the response information corresponding to the search query as target response information from a second data group including a plurality of QA pairs different from the first data group, based on the selected first QA pair information processing method.
(20)
Selecting a first QA pair corresponding to the search query from a first data group containing a plurality of QA pairs that are a combination of question information corresponding to the response and response information indicating the response,
executing a process of selecting the response information corresponding to the search query as target response information from a second data group including a plurality of QA pairs different from the first data group, based on the selected first QA pair Information processing program that allows

1 information processing system 100 information processing device 110 communication unit 120 storage unit 121 response log storage unit 122 log QA storage unit 123 manual QA storage unit 124 content information storage unit 130 control unit 131 acquisition unit 132 conversion unit 133 first selection unit 134 second 2 selection unit 135 generation unit 136 transmission unit (providing unit)
10 terminal device (information processing device)
11 communication unit 12 input unit 13 display unit (providing unit)
14 storage unit 15 control unit 151 acquisition unit 152 transmission unit 153 reception unit 154 processing unit 16 audio output unit

Claims

A first selection unit that selects a first QA pair corresponding to a search query from a first data group including a plurality of QA pairs that are a combination of question information corresponding to a response and response information indicating the response;
Based on the first QA pair selected by the first selection unit, from a second data group containing a plurality of QA pairs different from the first data group, the response information corresponding to the search query, the target response a second selection unit for selecting as information;
Information processing device.
The first selection unit
The information processing apparatus according to claim 1, wherein said first QA pair is selected from said first data group including a plurality of QA pairs that are combinations of said question information, which is character information, and said response information.
The first selection unit
The information processing apparatus according to claim 1, wherein the first QA pair is selected from the first data group including a plurality of QA pairs that are combinations of the question information and the response information that is character information.
The first selection unit
The information processing apparatus according to claim 1, wherein said first QA pair is selected from said first data group including a plurality of QA pairs based on response logs.
The first selection unit
The information processing apparatus according to Claim 4, wherein said first QA pair is selected from said first data group including a plurality of QA pairs generated by processing a response log.
The second selection unit
The information processing apparatus according to claim 1, wherein the target response information is selected from the second data group including a plurality of QA pairs that are combinations of the question information and the response information that satisfies a response condition.
The second selection unit
The information processing apparatus according to claim 6, wherein the target response information is selected from the second data group including a plurality of QA pairs that are combinations of the question information and the response information that does not include personal information.
The second selection unit
The information processing apparatus according to claim 6, wherein the target response information is selected from the second data group including a plurality of QA pairs that are combinations of the question information and the context-independent response information.
The first selection unit
The information processing apparatus according to claim 1, wherein the first QA pair is selected by searching the first data group using the search query.
The first selection unit
The information processing apparatus according to claim 9, wherein the first QA pair is selected based on the similarity between the question information of each QA pair of the first data group and the search query.
The second selection unit
determining a second QA pair corresponding to the question information of the first QA pair by searching the second data group using the question information of the first QA pair; The information processing apparatus according to claim 1, wherein the response information of 2QA pairs is selected as the target response information.
The second selection unit
The information processing apparatus according to claim 1, wherein the target response information is selected from the second data group by searching the second data group using the response information of the first QA pair.
The second selection unit
Determine a second QA pair corresponding to the first QA pair by searching the second data group using the first QA pair, and from the second data group, the response information of the second QA pair, The information processing apparatus according to claim 1, wherein the target response information is selected.
a generation unit that generates display information indicating the target response information;
The information processing apparatus according to claim 1, comprising:
The generating unit
15. The information processing apparatus according to claim 14, wherein the display information for displaying a list of the first response information, which is the response information selected using only the first data group, and the target response information is generated.
The generating unit
15. The information processing apparatus according to claim 14, wherein the display information for displaying a list of the second response information, which is the response information selected using only the second data group, and the target response information is generated.
The generating unit
15. The information processing apparatus according to claim 14, wherein when there is a plurality of pieces of target response information selected by the second selection unit, the display information for displaying the plurality of pieces of target response information side by side is generated.
a providing unit that provides the display information generated by the generating unit;
The information processing apparatus according to claim 14, comprising:
Selecting a first QA pair corresponding to the search query from a first data group containing a plurality of QA pairs that are a combination of question information corresponding to the response and response information indicating the response,
executing a process of selecting the response information corresponding to the search query as target response information from a second data group including a plurality of QA pairs different from the first data group, based on the selected first QA pair information processing method.
Selecting a first QA pair corresponding to the search query from a first data group containing a plurality of QA pairs that are a combination of question information corresponding to the response and response information indicating the response,
executing a process of selecting the response information corresponding to the search query as target response information from a second data group including a plurality of QA pairs different from the first data group, based on the selected first QA pair Information processing program that allows