JP7232485B1 - Information processing device, information processing system, information processing method and program - Google Patents

Abstract

An object of the present invention is to provide an appropriate answer to a question of an inquirer. An information processing apparatus (100) sequentially provides answers to a series of questions using a question and answer data DB (200) that stores information about features of questions and answers. The information processing apparatus 100 extracts features related to the first sentence of the series of questions based on the first sentence corresponding to the current question, and determines an answer to the current question based on the features and the question and answer data DB 200. An answer determination unit 130 is provided. If the answer to the current question cannot be determined based on the characteristics of the first sentence, the answer determination unit 130 determines the second sentence and the first sentence corresponding to the answers to the past questions before the current question. Based on this, each feature regarding the first sentence and the second sentence is extracted, and the answer to the current question is determined based on each feature and the question answer data DB 200 . [Selection drawing] Fig. 2

Description

The present invention relates to an information processing device, an information processing system, an information processing method, and a program for providing answers to questioners' questions.

Conventionally, there exists a technology for providing answers to questions asked by questioners. For example, a technology has been proposed that uses artificial intelligence to automatically determine and output an answer to a question (see, for example, Patent Document 1).

Japanese Patent Publication No. 2019-528512

With the conventional technology described above, it is possible to automatically determine an answer to a question, but depending on the content of the question, it may be difficult to determine an appropriate answer. For example, if an additional question such as "Why?"

An object of the present invention is to provide an appropriate answer to the question of the questioner.

One aspect of the present invention is an information processing device that sequentially provides answers to a series of questions using a database that stores information about features of questions and answers. This information processing device extracts features related to the first sentence of a series of questions based on the first sentence corresponding to the current question, and determines an answer to the current question based on the features and the database. Prepare. If the answer to the current question cannot be determined based on the characteristics of the first sentence, the determination unit determines based on the second sentence and the first sentence corresponding to the answers to the past questions before the current question. Features related to the first sentence and the second sentence are extracted, and an answer to the current question is determined based on these features and the database.

ADVANTAGE OF THE INVENTION According to this invention, an appropriate answer can be provided with respect to a questioner's question.

FIG. 1 is a diagram showing a simplified example of use of an information processing apparatus. FIG. 2 is a block diagram illustrating a functional configuration example of an information processing apparatus. FIG. 3 is a diagram schematically showing moving images stored in the moving image DB. FIG. 4 is a diagram showing in simplified form the contents of the question-and-answer data stored in the question-and-answer data DB. FIG. 5 is a block diagram showing a configuration example of a feature extractor used when extracting features using AI. FIG. 6 is a diagram showing an example of the relationship between an input sentence and feature amounts. FIG. 7 is a diagram showing in simplified form the contents of the question-and-answer data stored in the question-and-answer data DB. FIG. 8 is a diagram schematically showing an example of extraction processing for extracting feature amounts from new questions. FIG. 9 is a diagram schematically showing an example of extraction processing for extracting feature amounts from new questions. FIG. 10 is a flowchart showing an example of conversation response processing in the information processing device. FIG. 11 is a block diagram showing a functional configuration example of an information processing system. FIG. 12 is a block diagram illustrating a functional configuration example of an information processing system.

Embodiments of the present invention will be described below with reference to the accompanying drawings.

[Usage example of information processing device]
FIG. 1 is a diagram showing a simplified usage example when using an information processing apparatus 100. As shown in FIG. The information processing device 100 is a conversation response device that reproduces a prepared moving image of a respondent U2 in response to a question posed by a questioner U1. That is, a previously prepared moving image of the respondent U2 is displayed on the display unit 141, and audio information associated with the previously prepared moving image of the respondent U2 is output from the audio output unit 142. FIG. For example, by displaying the respondent U2 life-size on a relatively large display device, it is possible to create an atmosphere of having a conversation with the respondent U2.

Respondent U2 is previously photographed and recorded as a moving image of respondent U2. That is, in the present embodiment, as an answer to the question from the questioner U1, the answer part of the pre-recorded answer movie is reproduced instead of generating a new answer movie by image processing technology every time the question is asked. can do. Here, for example, it is conceivable to generate a new answer image of the answerer U2 by image processing. However, the answer image of the respondent U2 generated by image processing may have an awkward expression or an awkward answer. On the other hand, in the present embodiment, it is possible to play back the reply video of the respondent U2, which has been shot in advance. It is possible to prevent it from slipping. In addition, by playing back an answer moving image of answerer U2 that was actually shot, a realistic response can be realized.

[Configuration example of information processing device]
FIG. 2 is a block diagram showing a functional configuration example of the information processing apparatus 100. As shown in FIG.

The information processing apparatus 100 includes an input unit 110, a voice recognition unit 120, an answer determination unit 130, an answer response unit 140, a question/answer data DB200, and a video DB300.

The input unit 110 is an input unit that receives the voice uttered by the questioner U1 in real time. For example, as the input unit 110, a voice input device such as a microphone or an input device dedicated to voice recognition can be used. Further, for example, an input device such as a camera capable of acquiring an image, a microphone capable of inputting sound, and the like, such as an imaging device, may be used. Also, for example, an input device capable of inputting characters, such as a keyboard, may be used. Note that FIG. 2 shows an example in which the information processing device 100 is provided with the input unit 110 , but an input device separate from the information processing device 100 may be used as the input unit 110 . That is, by registering an input device such as a microphone in the information processing system in advance, it can be made to function as an input unit of the information processing apparatus 100 . An example of providing an input device separate from the information processing apparatus 100 is shown in FIG.

The input unit 110 outputs the voice information of the question to the voice recognition unit 120 when receiving the question from the questioner U1. Note that the input unit 110 accepts a question (for example, an additional question) from the questioner U1 while the answer video is being played back from the answer response unit 140 (that is, during answering).

The speech recognition section 120 converts the content of the question asked by the questioner U1 into text data based on the speech information output from the input section 110, and outputs the converted text data to the answer determination section 130. It should be noted that a publicly known speech recognition technology can be employed as a method of conversion into text data by the speech recognition unit 120 . For example, an API (Application Programming Interface) can be used. Note that the speech recognition unit 120 is realized by, for example, a processing circuit such as a CPU (Central Processing Unit). Note that when an input device capable of inputting characters is used as the input unit 110, the speech recognition unit 120 can be omitted. In this case, input unit 110 directly outputs text data generated by character input to answer determination unit 130 .

The answer determination unit 130 determines an answer to the question of the questioner U1 based on the text data output from the input unit 110 or the speech recognition unit 120, and transmits answer information on the determined answer to the answer response unit. output to 140. For example, the answer determination unit 130 extracts features related to the question of the questioner U1 based on the text data output from the input unit 110 or the speech recognition unit 120. FIG. As the feature of the question, for example, words included in the question may be extracted as shown in FIG. 4 and the like, or numerical feature amounts may be extracted as shown in FIG. 7 and the like. Then, the answer determination unit 130 extracts questions similar to the features of the extracted question from the question/answer data DB 200, and outputs answer information related to answers associated with the extracted questions to the answer response unit 140. . In this way, the answer determination unit 130 searches for a question similar to the characteristics of the question asked by the questioner U1 from the question-and-answer data DB 200 prepared in advance, and provides answer information related to the answer associated with the searched question. Output to unit 140 . The process of extracting the characteristics of these questions and the process of determining answers will be described in detail with reference to FIGS. 4 to 9 and the like. The answer determination unit 130 is implemented by, for example, a processing circuit such as a CPU.

The answer response unit 140 is an output unit that provides the questioner U1 with an answer video to the question of the questioner U1. Specifically, the answer response unit 140 extracts the playback portion in the video DB 300 based on the answer information output from the answer determination unit 130, displays the extracted video on the display unit 141, and displays the video on the display unit 141. The associated audio information is output from the audio output unit 142 . In this way, the answer responding section 140 functions as a reproducing section that reproduces the received reply. For example, an image display device such as a display capable of outputting images and sounds can be used as the answer response unit 140 . Also, for example, an output device capable of outputting at least one of images and sounds may be used. 1 and 2 show an example in which the display unit 141 and the voice output unit 142 are provided in the information processing apparatus 100 as an example of the answer response unit 140. You may use as the response part 140. FIG. An example of providing an output device separate from the information processing device 100 is shown in FIG.

The storage unit 150 is a storage medium that stores various information. For example, the storage unit 150 stores various types of information (for example, control programs, question and answer data DB 200, video DB 300) necessary for the speech recognition unit 120, the answer determination unit 130, and the answer response unit 140 to perform various processes. be done. Various storage media such as ROM (Read Only Memory), RAM (Random Access Memory), HDD (Hard Disk Drive), and SSD (Solid State Drive) can be used as the storage unit 150 .

The question-and-answer data DB 200 is a database that stores combinations of prepared questions and answers. The question-and-answer data DB 200 will be described in detail with reference to FIG.

The moving image DB 300 is a database that stores previously prepared moving images of respondent U2. Note that the moving images stored in the moving image DB 300 will be described in detail with reference to FIG.

[Video configuration example]
FIG. 3 is a diagram schematically showing a moving image 310 stored in the moving image DB 300. As shown in FIG. In FIG. 3, the length of the playback time from the start position S1 to the end position E1 of the moving image 310 is indicated by a rectangle extending in the horizontal direction.

A moving image 310 stored in the moving image DB 300 can be reproduced from each part by specifying the position of the moving image. In FIG. 3, the moving image portion from the moving image positions S11 to S12, the moving image portion from the moving image positions S21 to S22, the moving image portion from the moving image positions S31 to S32, and part of the image of each of these moving image portions The relationship with is simplified and shown. Each of these animated portions will be described in detail with reference to FIG.

For example, when answer information specifying moving image positions S11 to S12 is output from the answer determining unit 130, the answer responding unit 140 can reproduce the moving image portion from the moving image positions S11 to S12.

[Configuration example of question and answer data DB]
FIG. 4 is a diagram showing in simplified form the contents of the question-and-answer data stored in the question-and-answer data DB 200. As shown in FIG.

The question-and-answer data DB 200 is a database for storing various kinds of information for extracting answers to questions from the questioner U1 from the video 310 of the answerer U2 prepared in advance.

Specifically, a question 201 , an answer 202 , a feature 203 of the question, and a position 204 of the answer moving image are associated and stored in the question/answer data DB 200 .

The question 201 is information indicating the contents of an assumed question. This question can be generated in advance by an administrator or the like on the information processing apparatus 100 side based on the content of the moving image 310 of the respondent U2, and can be stored. In addition, artificial intelligence (AI) or the like may be used to estimate the question from the moving image 310 of the respondent U2 and store the question.

The answer 202 is information indicating the content of the answer to the question 201 . This answer can be generated in advance by an administrator or the like on the information processing apparatus 100 side based on the content of the moving image 310 of the respondent U2, and can be stored. An answer may be extracted from the moving image 310 of the answerer U2 using artificial intelligence or the like and stored.

The question feature 203 is information indicating a word or the like that is a feature of the sentence corresponding to the question 201 . For example, the question 201 "Where are you from?" features the words "where are you from" and "where are you from?" The characteristics of this question can be generated in advance by an administrator or the like on the information processing apparatus 100 side based on the content of the question 201 and stored. In addition, using artificial intelligence or the like, the feature of the question may be extracted and stored.

The answer moving image position 204 is information indicating the recorded portion in the moving image 310 stored in the moving image DB 300 . For example, the position 204 "00:23:43-00:24:15" of the answer video corresponding to the question 201 "Where are you from?" handle. Also, the position 204 "01:19:01-01:19:06" of the answer video corresponding to the question 201 "What is your favorite food?" corresponding to the part. Also, the position 204 "01:19:12-01:20:13" of the answer video corresponding to the question 201 "Why do you like rice balls?" corresponds to

[Answer extraction example]
Next, a description will be given of an extraction method by which the answer determination unit 130 extracts an answer to the question of the questioner U1 using the question/answer data DB 200. FIG.

Here, a case where the questioner U1 asks the question "What is your favorite food?" will be described as an example. Thus, when the questioner U1 issues the question "What is your favorite food?", the input unit 110 acquires the voice "What is your favorite food?" What is your favorite food?” is output to the speech recognition unit 120.

The speech recognition unit 120 converts the speech information “What is your favorite food?” output from the input unit 110 into text data “What is your favorite food?” Is it?” is output to the answer determination unit 130.

The answer determination unit 130 extracts features of the question based on the text data “What is your favorite food?” output from the speech recognition unit 120 . The features of this question are extracted based on preset words. For example, each word included in a sentence corresponding to a question is extracted as a feature of the question. Note that a known sentence analysis technique can be used as a method for extracting words contained in sentences.

In the example shown in FIG. 4, each word "What is your favorite food?" which constitutes the text data "What is your favorite food?" is extracted as a feature.

Next, the answer determining unit 130 searches the question features 203 for words that are the same as or similar to the extracted question features “like, food, what”. In the example shown in FIG. 4, as indicated by a dotted-line rectangle 211, the same feature “like, food, what” as the extracted question feature “like, food, what” is stored in the question feature 203. It is For this reason, the answer determination unit 130 selects the answer 202 "I like rice balls" and the answer video position 204 "01:19:01" as answers to the extracted question characteristics "like, food, what". -01:19:06”.

Extraction processing for searching for words that are the same as or similar to the question feature of the questioner U1 from the question feature 203 will now be described. For example, among the combinations of words stored in the question features 203, if there is a combination of words that matches all of the features of questioner U1's question, the The answer 202 corresponding to the combination and the position 204 of the answer moving image are extracted. Further, for example, if there is a combination of words that partly matches the question feature of the questioner U1 among the word combinations stored in the question feature 203, the part matches. The answer 202 corresponding to each word combination and the position 204 of the answer animation are extracted. In this case, the condition may be that a predetermined percentage or more of the words match. For example, 60% or more words may be matched as a condition. For example, if the question feature of the questioner U1 is "likes, food, there is", "likes, food" of the question feature 203 "likes, food, what" matches. In this case, 66% of the words match, so the answer 202 and the position 204 of the answer moving image corresponding to the question feature 203 "like, food, what" are extracted. Note that if there are multiple combinations in which words match at a predetermined ratio or more, the answer 202 and the position 204 of the answer moving image corresponding to the combination with the highest ratio are extracted. In this way, the answer determination unit 130 determines the current determine the answer to the question.

On the other hand, if there is no word that matches the characteristics of questioner U1's question among the combinations of words stored in the question characteristics 203, or if some of the characteristics of questioner U1's question match However, it is also envisioned that the match is less than a predetermined percentage. In this case, it is determined that there is no answer to the characteristics of questioner U1's question, 'What do you like, what is food?'. In this way, when there is no answer to the characteristics of questioner U1's question, as will be described later, it is possible to generate a new question and extract an answer using the answer before the current question. It is possible.

Next, the answer determining unit 130 sets the extracted answer 202 “I like rice balls” and the position 204 of the answer moving image “01:19:01-01:19:06” as answer information to the answer responding unit 140. output to

The answer response unit 140 responds to the questioner based on the answer 202 “I like rice balls” output from the answer determination unit 130 and the position 204 “01:19:01-01:19:06” of the answer video. The answer "I like rice balls." to the question "What is your favorite food?" from U1 is output. In this case, the response response unit 140 may output the voice information “I like rice balls.” A moving image portion corresponding to "01:19:01-01:19:06" may be extracted and reproduced. Thus, the answer "I like rice balls." to the question "What is your favorite food?" from the questioner U1 is output.

[Conversation example considering context]
Here, while the answer "I like rice balls." to the question "What is your favorite food?" is also expected to be issued. For example, it is assumed that the questioner U1 asks the question "Why?" while the answer response unit 140 is outputting the answer "I like rice balls." In this case, the answer determining unit 130 extracts the feature "why" from the question "why?". Also, it is assumed that the feature 203 of the question does not have a feature that is the same as or similar to the feature "why?".

In this way, if the question feature 203 does not have a feature identical or similar to the question feature “why?” extracted from the question “why?”, the answer determination unit 130 A new question is generated using the answers output before "Is it?". Specifically, the answer determining unit 130 synthesizes the immediately preceding answer “I like rice balls” with the current question “Why?” ?” is generated as a new question.

Similarly, the above-described feature extraction process and answer extraction process are sequentially executed for the new question "I like rice balls. Why?" In this example, it is assumed that "I like rice balls" is extracted as a characteristic of the question "I like rice balls."

Specifically, the answer determination unit 130 extracts the feature "I like rice balls, why" from the text data "I like rice balls. Why?". In addition, the answer determining unit 130 searches the question features 203 for words that are the same as or similar to the extracted question features “rice ball, like, why”. In the example shown in FIG. 4, as indicated by a dotted-line rectangle 212, the feature “Why, I like rice balls” similar to the extracted question feature “I like rice balls, why” is stored in the question features 203. ing. For this reason, the answer determination unit 130 selects the answer 202 as an answer to the characteristic of the extracted question, "I like rice balls, why?" Maybe it's because it reminds me of the rice ball that was made for me."

Next, the answer determination unit 130 selects the extracted answer 202, "Maybe because it reminds me of the rice balls that my mother made while saving little by little the rationd rice during the wartime when there weren't many things." , the position 204 of the answer moving image “01:19:12-01:19:13” is output to the answer response unit 140 . The output method from the answer response unit 140 is the same as the output method described above.

In addition, using the new question “I like rice balls. Why?” generated by combining the previous answer “I like rice balls” and the question “Why?” Even if the feature extraction process and the answer extraction process described above are executed, it is assumed that the answer to the current question will not be extracted. In such a case, a new question is generated by synthesizing the previous answer, the answer from the time before last, and the question "Why?" Execute extraction processing and answer extraction processing. However, even if a new question generated by combining the previous answer, the answer before the previous one, and the current question "Why?" is used, it is assumed that the answer to the current question will not be extracted. . In such a case, a new question is generated by synthesizing each answer from the previous answer to the three previous answers and the current question "Why?" are used to execute the above-described feature extraction processing and answer extraction processing. That is, until the answer to the current question is extracted, each answer from the previous answer to the previous answer N times (where N is an integer equal to or greater than 2) is synthesized to generate a new question. Using the new question, the feature extraction process and answer extraction process described above are executed.

However, if we go back too far in the past based on this question, it is possible that the answer will be very different from the content of this question. Therefore, a threshold TH1 may be set, and each answer from the answer one time before to the answer one time before the threshold TH may be synthesized. Here, the threshold TH1 is a value that can be appropriately set based on experimental data or the like. For example, a value of about 2 to 5 can be used as the threshold TH1. As the threshold TH1, the threshold TH1 may be set to a high value when answers with a high degree of similarity continue. That is, it is possible to change the threshold TH1 according to the content of the answer to be synthesized. Instead of using the threshold TH1, the number of characters in a sentence corresponding to a new question generated by synthesis may be used as a reference to determine an answer that goes back to the past as a synthesis target. For example, the process of generating a new question may be stopped on condition that the number of characters in a sentence corresponding to a new question generated by synthesis exceeds 100 characters.

Here, for example, the question "What is your favorite food?" is issued by the questioner U1, and while the answer "I like rice balls." Why?" is assumed. In this case, with the conventional technology, it is assumed that it is difficult to generate an appropriate answer to the additional question "Why?" because the answer is determined using only the additional question "Why?" be done. That is, besides the additional question "why?", we need other elements that can be used to generate an answer. Therefore, it is conceivable that "I do not understand the content of the question. Please try again." is output as an answer to the additional question in order to prompt the input of other elements.

On the other hand, in this embodiment, when the questioner U1 issues an additional question "Why?", the previous answer "I like rice balls." ?” to generate a new question “I like rice balls. Why?” For this reason, as an answer to the additional question, it is possible to output "I wonder if it reminds me of the omusubi that my mother made while saving little by little rice from the ration during the wartime when there was nothing." Become.

It is also assumed that the user will have a conversation about a nearby ramen shop. For example, the question "Tell me about a nearby ramen shop" is issued by the questioner U1, and the answer "There are three ramen shops nearby. Do you want to display them?" It is assumed that an additional question "Where is the most popular?" is issued by person U1. In this case, with conventional technology, it is difficult to generate an appropriate answer to the additional question "Where is the most popular?" because the answer is determined using only the additional question "Where is the most popular?" is assumed to be That is, besides the additional question "Where is the most popular?", we need other factors that can be used to generate an answer. Therefore, it is conceivable that "I do not understand the content of the question. Please try again." is output as an answer to the additional question in order to prompt the input of other elements.

On the other hand, in this embodiment, when the additional question "Where is the most popular?" and a new question "There are three ramen shops nearby. Do you want to display them? Where is the most popular?" It is possible to answer using Therefore, it is possible to output "store A has the highest evaluation, but store B is the closest" as an answer to the additional question.

[Example of extracting answers using AI]
In the above, an example of executing the feature extraction process and the answer extraction process using words or the like as the feature of the question from the questioner U1 has been shown. Here, it is also conceivable to use artificial intelligence to extract the features of the question from questioner U1, and to use these features to extract answers. Therefore, an example using AI will be shown below.

[Example of feature extraction using AI]
FIG. 5 is a block diagram showing a configuration example of the feature extractor 131 used when extracting features using AI. It is assumed that feature extractor 131 is included in answer determination section 130 . It should be noted that the learning shown in this embodiment means finding the regularity behind the data based on a large amount of data. Learning data generated by learning shown in this embodiment is generated by various learning algorithms.

The feature extractor 131 extracts a feature quantity 136 from an input sentence 135 using learning data learned in advance. That is, the feature extractor 131 extracts a feature amount relating to the sentence corresponding to the question from the questioner U1. The feature amounts shown here are obtained by digitizing the question sentences so that the computer can handle the question sentences. Typical examples of this quantification include word2vec, which performs quantification focusing on the co-occurrence relationship of words, deep learning, and the like. Also, by learning using word2vec, deep learning, etc., it is possible to obtain numerical values that include sentence features as feature amounts. For example, various sentences are learned in advance using word2vec, deep learning, or the like to generate learning data, and this learning data is used for extracting feature amounts. For example, it is possible to generate learning data by reading each sentence such as a sentence about food, a sentence about a vehicle, a sentence about emotion, etc. as teacher data. Regarding feature amounts extracted using this learning data, similar words (or meanings of words) contained in sentences have the property of being similar numerical feature amounts. Therefore, the feature extractor 131 extracts the feature amount of the sentence using the property. Thus, in the present embodiment, it is possible to use learning data previously learned by deep learning, word2vec, or the like for extracting feature amounts.

[Example of relationship between input sentence and feature quantity]
FIG. 6 is a diagram showing an example of the relationship between the input sentence 161 and the feature quantity 162. As shown in FIG. For example, when the feature extractor 131 receives the input sentence 161 "Do you like apples?"

Note that FIG. 6 shows, as a simplified example, an example of extracting a feature quantity 162 consisting of six combinations of numerical values for one input sentence 161 in order to facilitate the explanation. However, in practice, feature quantities 162 consisting of a large number (for example, about 1000 sets) of combinations of numerical values are extracted for one input sentence 161 .

The feature values extracted in this manner include values within a predetermined range (eg, 0 to 1, -1 to 1) related to some category. Note that FIGS. 6 to 9 show examples in which values of 0 to 1 are extracted in order to facilitate the explanation.

In the example shown in FIG. 6, the first numerical value indicates a numerical value related to "food", the second numerical value indicates a numerical value related to "vehicle", and the third numerical value indicates a numerical value related to "emotion". , the fourth number indicates a number associated with "action", the fifth number indicates a number associated with "place", and the sixth number indicates a number associated with "question". Note that the closer the value is to 0, the lower the relevance, and the closer the value is to 1, the higher the relevance.

For example, among the feature values of the input sentence "Do you like apples?" Do you like?" can be presumed to be highly related to food and emotions. In addition, among the feature values of the input sentence “Do you like apples?”, the numerical value related to “question” is 0.3. It can be assumed that there is On the other hand, among the feature quantities of the input sentence "Do you like apples?", since each numerical value related to "vehicle", "movement", and "place" is 0, the input sentence "Do you like apples?" It can be assumed that there is no relation to vehicle, action, or place.

As described above, in practice, feature quantities 162 composed of a large number (for example, about 1000 sets) of combinations of numerical values are extracted for one input sentence 161. It is difficult for humans to read the meaning. However, each numerical value indicates some relevance.

[Configuration example of question and answer data DB]
FIG. 7 is a simplified diagram showing the contents of the question-and-answer data stored in the question-and-answer data DB 200. As shown in FIG. Note that the example shown in FIG. 7 is a partial modification of FIG. 4, and the feature 205 of the question is different from that in FIG. For this reason, the same reference numerals as those in FIG. 4 are assigned to the features other than the feature 205 of the question, and the description thereof is omitted.

The question and answer data DB 200 stores questions 201, answers 202, question features 205, and answer moving image positions 204 in association with each other.

The question feature 205 is information indicating the feature amount extracted for the sentence corresponding to the question 201 . These feature quantities can be extracted using the feature extractor 131 as described above. It should be noted that the feature amount related to the answer 202 may be held in the question/answer data DB 200 . As a result, it is possible to reduce the amount of calculation in the process of extracting the feature amount of a new question, which will be described later.

[Answer extraction example]
Next, a description will be given of an extraction method by which the answer determination unit 130 extracts an answer to the question of the questioner U1 using the question/answer data DB 200. FIG.

Here, as in the above example, the case where the questioner U1 asks the question "Do you have a favorite food?" will be described as an example. In this case, voice input by the input unit 110 in response to questioner U1's question "Do you have any favorite food?" It is the same as each processing performed.

The feature extractor 131 of the answer determination unit 130 extracts the feature amount "0.6 0 0.7 0 0 0.39" based on the text data "Do you have any favorite food?" output from the speech recognition unit 120. to extract

Next, the answer determination unit 130 searches for a feature amount that is the same as or close to the extracted feature amount “0.6 0 0.7 0 0 0.39” from the feature 205 of the question. In the example shown in FIG. 7, as indicated by a dotted-line rectangle 213, the extracted feature amount "0.6 0 0.7 0 0 0.39" is close to the feature amount "0.6 0 0.7 0 0 0.4” is stored in the question feature 205 . Therefore, the answer determination unit 130 determines the answer 202 "I like rice balls" and the answer video position 204 " 01:19:01-01:19:06”.

Here, an extraction process for searching the question features 205 for feature amounts that are the same as or close in numerical value to the question feature amount of questioner U1 will be described. In this extraction process, when the question feature quantity of the questioner U1 completely matches each feature quantity stored in the question feature 205, the completely matching feature quantity is extracted from the question feature 205. extracted. On the other hand, if none of the feature quantities stored in the question feature 205 completely match the feature quantity of questioner U1's question, at least some of them are the same. , is extracted from the feature 205 of the question such that the difference value from other values is within a predetermined value. For example, among the six sets of feature amounts of the question of questioner U1, at least four sets of values are the same as the corresponding values of the question feature 205, and the other sets of values correspond to the question feature 205. Similarity is determined when the difference value from the value is within a predetermined value, for example, within 0.1. For example, in the above-described example, the question feature amount "0.6 0 0.7 0 0 0.39" of the questioner U1 and the question feature 205 "0.6 0 0.7 0 0 0.4" , the values in the first five columns are identical, and the difference value of the values in the last six columns is 0.01. Therefore, both are determined to be similar.

Further, for example, the difference values between the six sets of feature amounts of the question of questioner U1 and the corresponding values of the question feature 205 are sequentially calculated for each column, and the sum of the six sets of difference values for each column is calculated. If the total value of the six sets of difference values for each column is within a predetermined value, for example, within 0.2, it may be determined that they are similar. For example, in the above-described example, the question feature amount "0.6 0 0.7 0 0 0.39" of the questioner U1 and the question feature 205 "0.6 0 0.7 0 0 0.4" , the sum of the difference values in the first five columns is zero because each value in the first five columns is the same. Also, since the difference value of the values in the last six columns is 0.01, the sum of the six sets of difference values for each column is 0.01. In this case, both are determined to be similar. Note that if there are a plurality of feature amounts determined to be similar among the feature amounts stored in the question feature 205, the combination with the smallest total value of the six sets of difference values for each column The answer 202 corresponding to , and the position 204 of the answer moving image are extracted. In this way, the answer determination unit 130 determines an answer to the current question based on the degree of matching (or similarity) between the feature amount related to the question sentence of the questioner U1 and the feature amount stored in the feature 205 of the question. decide.

Next, the answer determining unit 130 outputs the extracted answer 202 “I like rice balls” and the answer moving image position 204 “01:19:01-01:19:06” to the answer responding unit 140. . The output method from the answer response unit 140 is the same as the output method described above.

[Conversation example considering context]
Here, while the answer "I like rice balls." to the question "Do you have a favorite food?" It is also assumed that For example, it is assumed that the questioner U1 asks the question "Why?" while the answer response unit 140 is outputting the answer "I like rice balls." In this case, the answer determination unit 130 assumes that there is no feature quantity in the feature 205 of the question that is the same as or close in value to the feature quantity extracted from the question “Why?”.

In this way, if there is no feature quantity that is the same as or numerically close to the feature quantity from the question "Why?" A new question is generated using the answers output before "Why?". Specifically, the answer determining unit 130 synthesizes the immediately preceding answer “I like rice balls” with the current question “Why?” ?” is generated as a new question.

FIG. 8 is a diagram schematically showing an example of extraction processing for extracting feature amounts from new questions. As shown in FIG. 8A, for each feature value of each sentence "I like rice balls." are added to extract the feature quantity of a new question “I like rice balls. Why?”. FIG. 8B shows the result of adding the numerical values in each column.

Specifically, as indicated by a dotted-line rectangle 171, the numerical value “0.8” is calculated by adding the numerical value “0.8” and the numerical value “0” relating to the feature amount “food”. Further, as indicated by a dotted-line rectangle 172, the numerical value “0.7” is calculated by adding the numerical value “0.7” and the numerical value “0” regarding the “emotion” of the feature quantity. Further, as indicated by a dotted-line rectangle 173, the numerical value “0.29” is calculated by adding the numerical value “0” and the numerical value “0.29” relating to the feature amount “questionable”. For each column other than these, since the numerical value related to the feature amount is "0", the numerical value "0" is calculated as the added value for each column. Note that it is assumed that there may be columns in which the added value exceeds 1 among the numerical values in each column. In this way, for a column whose added value exceeds 1, the added value of that column is converted to 1 and used. Note that the feature extractor 131 may be used to extract feature amounts from new questions.

Next, the answer determination unit 130 searches for a feature amount that is the same as or close to the feature amount “0.8 0 0.7 0 0 0.29” obtained by adding the numerical values in each column from the question feature 205 . In the example shown in FIG. 7, as indicated by a dotted-line rectangle 214, the feature amount "0.8 0 0.7 0 0 0.29" obtained by adding the numerical values in each column and the feature amount "0.8 0 0.7 0 0 0.3” is stored in the question features 205 . For this reason, the answer determination unit 130 determines the answer 202, "A wartime era without goods, rations Maybe it's because I'm saving rice little by little and it reminds me of the rice balls my mother made for me."

Using the synthesized and generated new question "I like rice balls. Why?" Maybe it's because I remember the rice balls that my mother made by saving little by little." .

Next, the answer determination unit 130 selects the extracted answer 202, "Maybe because it reminds me of the rice balls that my mother made while saving little by little the rationd rice during the wartime when there weren't many things." , the position 204 of the answer moving image “01:19:12-01:19:13” is output to the answer response unit 140 . As a result, the answer to the question from questioner U1, "Why?" Naa" is played.

In addition, using the new question “I like rice balls. Why?” generated by combining the previous answer “I like rice balls” and the question “Why?” It is conceivable that the answer may not be extracted even if the feature quantity extraction process and the answer extraction process are executed. In such a case, a new question is generated by synthesizing the previous answer, the answer before last, and the question "Why?" , and the answer extraction process. Likewise, if no answer is extracted, the previous answer is used in sequence to execute the above-described feature quantity extraction process and answer extraction process. However, as described above, the threshold TH1, the number of characters in the sentence corresponding to the new question, or the like may be used to limit the answers that go back to the past to be synthesized.

[Modified example of feature amount addition processing]
An example of calculating the feature amount of the new question by adding each column of the feature amount of each sentence constituting the new question has been described above. Here, it is assumed that past answers far from the current question on the time axis will be less relevant to the current question than the current question. Therefore, when calculating the feature amount of a new question, it is conceivable to give greater importance to the features of the most recent answer by reducing the degree of influence of an answer in the past that is farther from the current question. Therefore, below, an example of calculating the feature amount of a new question by decreasing the degree of influence for answers in the past that are farther from the current question will be shown.

FIG. 9 is a diagram schematically showing an example of extraction processing for extracting feature amounts from new questions. FIG. 9 shows an example of feature amount extraction processing when a new question is generated by synthesizing the previous answer, the answer before last, and the current question.

As shown in FIG. 9(A), each of the sentences "I'll ride in the car.", "I like rice balls.", "Why?" ?” is used to extract the feature quantity. It is assumed that "I will ride in a car" is the answer from the last time, "I like rice balls" is the answer from the previous time, and "Why?" is the question from this time. In this case, the feature amount of a new question is calculated by reducing the degree of influence of a past answer that is farther from the current question. Specifically, as shown in FIG. 9B, the previous answer is multiplied by a coefficient of 0.9, the answer before the previous one is multiplied by a coefficient of 0.8, and the numerical values in each column after multiplication are The addition is performed to extract the feature quantity of a new question "I ride in a car. I like rice balls. Why?". Note that the coefficients shown here are only examples, and other numerical values may be used.

Specifically, as indicated by a dotted-line rectangle 181, the numerical value "0.72" is calculated by multiplying the numerical value "0.8" of the previous question regarding the feature amount "food" by a coefficient of 0.9. In addition, since the other numerical value related to the feature quantity “food” is “0”, the numerical value related to the feature quantity “food” is calculated as “0.72”. In addition, as indicated by dotted-line rectangles 182 to 185, the feature amount of the new question is calculated similarly using the coefficients for the other feature amounts. FIG. 9C shows the addition result of adding the numerical values in each column.

Regarding the features of the new questions calculated in this way, ``I like rice balls'' + ``Why'' + ``Get in a car'' It becomes possible to By calculating the feature amount of the new question in this way, it is possible to synthesize a more appropriate question that emphasizes the most recent answer. In the example shown in FIG. 9, it is possible to derive an answer or the like corresponding to "What do you want to eat while riding a vehicle?"

In this way, the answer determination unit 130, based on each sentence about the first answer (previous answer) to N-th answer (answer N times before) (where N is an integer of 2 or more), Calculate the feature amount. Then, the answer determination unit 130 reduces the degree of influence of the feature amount regarding each sentence in response to the first answer to the Nth answer. For example, each numerical value that constitutes the feature amount for the first answer is multiplied by a coefficient of 0.9, each numerical value that constitutes the feature amount for the second answer is multiplied by a coefficient of 0.8, and a feature amount for the third answer is constituted. By multiplying each numerical value by a coefficient of 0.7, the degree of influence of the feature amount related to each sentence is reduced from the first answer to the third answer. In addition, by similarly multiplying each numerical value constituting the feature amount for each sentence after the fourth answer by a predetermined coefficient (value less than 0.7), the degree of influence of the feature amount for each sentence is reduced. . However, when the threshold TH1 of N times is set, each answer from the answer one time before to the answer one time before the threshold TH1 is used as a composition target.

Note that this example shows an example in which each of the feature amounts relating to the first to N-th answers is multiplied by a coefficient. Here, as described above, each feature amount consists of a large number (for example, about 1000 sets) of combinations of numerical values. Therefore, when each feature amount is multiplied by a coefficient, the computational load increases. is assumed. Therefore, in order to reduce the computational processing load, an attempt may be made to determine the answer without multiplying the feature amount related to the first answer by a coefficient. If the question cannot be determined using the feature amount related to the first answer, each feature amount after the second answer is multiplied by a coefficient to try to determine the answer. In order to further reduce the computational processing load, an attempt may be made to determine an answer without multiplying each feature amount relating to the first answer and the second answer by a coefficient. Then, if it is not possible to determine the current question using each feature amount related to the first and second answers, even if an attempt is made to determine the answer by multiplying the feature amount of the third and subsequent answers by a coefficient. good. In this way, if the feature amount related to the first answer (or the first and second answers) cannot be determined for this question, It is also possible to try to determine the answer with less impact. As a result, it is possible to appropriately determine the answer to the current question, and to reduce the computational processing load associated with the process of determining the answer.

[Example of operation of information processing device]
FIG. 10 is a flowchart showing an example of conversation response processing in the information processing apparatus 100. As shown in FIG. Also, this conversation response processing is executed based on a program stored in the storage unit 150 . Also, this conversation response processing is always executed in each control cycle. Also, this conversation response processing will be described with reference to FIGS. 1 to 9 as appropriate.

In step S501, the speech recognition unit 120 determines whether or not the questioner U1 has posed a question. When questioner U1 has posed a question, input unit 110 outputs the voice of the question uttered by questioner U1 to voice recognition unit 120, and proceeds to step S502. On the other hand, when the questioner U1 has not issued a question, monitoring is continued.

In step S502, the voice recognition unit 120 converts the voice of the question output from the input unit 110 into text data.

In step S503, the answer determination unit 130 extracts the features of the question sentence of the questioner U1 based on the text data output from the speech recognition unit 120, and determines the answer to the question of the questioner U1 based on the features. do. As described above, when using a word as a feature of the question sentence, the answer determination unit 130 extracts the word included in the question sentence of the questioner U1 as a feature. In this case, using the feature (one or more words) extracted from the question sentence of the questioner U1, the answer to the question of the questioner U1 is extracted from the question answer data DB 200 (see FIG. 4), and this The extracted answer is determined as an answer to questioner U1's question. Further, when using a numerical feature amount as the feature of the question sentence, the answer determination unit 130 extracts the feature amount (see FIG. 6) from the question sentence of the questioner U1. In this case, using the feature amount extracted from the question sentence of the questioner U1, the answer to the question of the questioner U1 is extracted from the question answer data DB 200 (see FIG. 7), and the extracted answer is used as the question determined as an answer to the question of person U1. Note that when a new question is generated in step S506, the answer determination unit 130 attempts to determine an answer to questioner U1's current question using the new question. That is, the answer determination unit 130 tries to extract an answer to the new question, and if an answer to the new question is extracted, the extracted answer is used as an answer to the questioner U1's current question. decide.

In step S<b>504 , the answer determination unit 130 determines whether or not the answer to the current question of the questioner U<b>1 has been extracted from the question/answer data DB 200 . If the answer to the current question can be extracted from the question/answer data DB 200, the process proceeds to step S505. On the other hand, if the answer to the current question could not be extracted from the question/answer data DB 200, the process proceeds to step S506.

In step S505, the answer response unit 140 executes a reproduction process of reproducing the moving image portion corresponding to the answer to the current question extracted in step S503.

In step S506, answer determination unit 130 synthesizes the current question and the answer to the previous question to generate a new question. Based on this new question, if the answer to the current question cannot be extracted from the question/answer data DB 200, the process of step S506 is repeated. In this case, until the answer to the current question can be extracted from the question/answer data DB 200, each answer from the previous answer to the previous answer N times (where N is an integer of 2 or more) is sequentially synthesized. Generate new questions. However, even if a new question is generated by sequentially synthesizing each answer up to the answer before the threshold value TH1, if the answer to the current question cannot be extracted from the question answer data DB 200, the questioner will receive a different answer. You may make it prompt the new question by expression.

[Configuration example of information processing system]
In the above, an example in which voice recognition processing, answer determination processing, answer response processing, and the like are executed in the information processing apparatus 100 has been shown, but all or part of each of these processes may be executed in another device. In this case, an information processing system is configured by each device that executes part of each of these processes. For example, using a server, a device that can be used by the questioner U1 (for example, a smartphone, a tablet terminal, a personal computer), various information processing devices such as a server that can be connected via a predetermined network such as the Internet, and various electronic devices At least part of each process can be executed. Examples of these are shown in FIGS.

A part (or all) of the information processing system capable of executing the functions of the information processing apparatus 100 may be provided by an application that can be provided via a predetermined network such as the Internet. This application is, for example, SaaS (Software as a Service).

FIG. 11 is a block diagram showing a functional configuration example of the information processing system 600. As shown in FIG.

The information processing system 600 includes an information processing device 610, an input device 620, an output device 630, a storage device 640, and the like. Note that the input device 620 corresponds to the input unit 110 shown in FIG. 2, the output device 630 corresponds to the answer response unit 140 shown in FIG. 2, and the storage device 640 corresponds to the storage unit 150 shown in FIG. . The speech recognition unit 120 and the answer determination unit 130 in the information processing device 610 correspond to the speech recognition unit 120 and the answer determination unit 130 shown in FIG. However, the answer determination unit 130 shown in FIG. 11 extracts the moving image part from the moving image DB 300 based on the answer information regarding the determined answer, and executes the reproduction process of outputting the moving image part to the output device 630 . In this example, one input device 620, one output device 630, and one storage device 640 are connected to the information processing device 610. However, a plurality of devices may be connected to the information processing device 610 as these devices. may be used as In this case, a plurality of storage devices 640 may store and use different answer animations.

FIG. 12 is a block diagram showing a functional configuration example of the information processing system 700. As shown in FIG.

The information processing system 700 includes a network 701, an information processing device 710, electronic devices 720 and 730, and the like. For example, an information processing device 710 and electronic devices 720 and 730 are connected via a network 701 . It should be noted that communication using wired communication or communication using wireless communication is performed between these devices. Further, as for communication between these devices, direct communication may be performed between the devices other than communication via the network 701 .

A network 701 is a network such as a public line network or the Internet. Further, each device constituting the information processing system 700 is connected to the network 701 by either a communication method using wireless communication or a communication method using wired communication, or both methods.

The information processing device 710 corresponds to the information processing device 100 shown in FIG. Each unit in the information processing apparatus 100 corresponds to each unit of the same sight shown in FIG. The information processing device 710 can be, for example, a server capable of providing various types of information.

The communication unit 711 uses at least one of wired communication and wireless communication to exchange various types of information with other devices. For example, the communication unit 711 executes reception processing for receiving question information regarding questions from the electronic devices 720 and 730, transmission processing for transmitting answer information regarding answers to the questions to the electronic devices 720 and 730, and the like.

Electronic devices 720 and 730 are fixed or portable information processing devices possessed by interrogators U1 and U3, such as smart phones, tablet terminals, smart watches, and personal computers. Also, the electronic devices 720 and 730 are devices capable of wired communication or wireless communication with the information processing apparatus 710 . Further, the electronic devices 720 and 730 execute transmission processing for transmitting question information regarding questions to the information processing device 710, reception processing for receiving answer information regarding answers to the questions from the information processing device 710, and the like. Further, for example, the electronic devices 720 and 730 can display the answers to the questions on the display units 721 and 731 and output the answers by voice based on the answer information.

For example, questioner U1 can use electronic device 720 to ask respondent U2 a question, and use electronic device 720 to view and hear answers from respondent U2. Further, for example, the questioner U3 can use the electronic device 730 to ask the respondent U2 a question, and use the electronic device 730 to view and listen to the reply from the respondent U2.

In addition, in this embodiment, as a questioning method, an example in which the questioner U1 asks a question by voice is shown, but the questioning method is not limited to this. For example, the questioner U1 may input the question as text information or as an image. Also, a telegram from another device may be input as a question from the questioner U1.

In addition, in the present embodiment, as a method of responding from the information processing apparatus 100, by reproducing the answer part of the moving image 310, an example of displaying an image and outputting an audio is shown, but the method is not limited to this. For example, when reproducing the answer part of the moving image 310, the characters corresponding to the voice may be displayed and output together with the image display, or the image display may be omitted and only the voice output may be performed. Further, for example, the answer to questioner U1 may be output as text information (for example, character display, voice output). Alternatively, the answer to the questioner U1 may be output as a telegram to another device, and the device may output the answer to the questioner U1.

Further, in the present embodiment, an example in which one moving image 310 is stored in the information processing apparatus 100, 710, etc., and the answer portion of the moving image 310 is played back has been described, but the present invention is not limited to this. For example, a plurality of answer videos may be stored in the information processing device 100, 710, etc., and the question from the questioner may be answered using the answer video desired by the questioner in response to a request from the questioner. . In addition, in the present embodiment, an example in which words and feature amounts of the question sentence are used as the feature of the question is shown, but the present invention is not limited to this. Other features that can be extracted from the question may be used.

[Effect of this embodiment]
Thus, according to the present embodiment, focusing on the continuity of the conversation between the question from the questioner U1 and the answer from the information processing device 100, it is possible to realize a conversation that takes into consideration the context of the conversation. . Specifically, based on the text of the question from the questioner U1 and the question in the question answer data DB 200 prepared in advance, a similar feature to the question from the questioner U1 is searched, and the corresponding answer part is searched. Play video. Further, by accepting an input from the input unit 110 during answer reproduction, the questioner U1 can ask a question even during answering. As a result, the questioner U1 can have a natural conversation by interrupting an answer and asking a question. That is, it is possible to realize a question-answering system that considers the context of the conversation and can respond in consideration of the contents of the conversation up to that point.

In addition, in this embodiment, in response to a question from questioner U1, an answer is searched for from an answer video prepared in advance by answerer U2 using the characteristics of the question, and a part corresponding to the answer is reproduced. system can be realized.

Further, in the present embodiment, when the answer to the current question cannot be extracted based on the features of the current question, the features of the answer to the previous question and the features of the current question are extracted, and these features are can be used to extract answers to questions. In addition, if the answer to the current question cannot be extracted based on this feature, respond to the answers to each question from 1 time before the current question to N times (where N is an integer of 2 or more) before the current question. It is possible to extract the features related to each sentence and the features related to the current question, and use these features to extract the answer to the question. As a result, even when the questioner U1 issues an additional question such as "Why?", it is possible to extract an appropriate answer from the content of the additional question. For example, by accepting input from the input unit 110 even during answer reproduction, the questioner U1 can ask a question even in the middle of answering. In this case, even if the question is interrupted by the answer, an appropriate answer to the question can be output, so natural conversation can be realized. In other words, it is possible to give a natural response considering the context of the content of the conversation.

Moreover, in the conventional question answering system, if additional questions were asked to the answers, there was a possibility that the answers would not be able to be given normally. In other words, the conventional question-answering system may only be able to respond with one question and one answer. On the other hand, in this embodiment, even if an additional question is asked in response to the answer, the answer can be appropriately answered. That is, it becomes possible to answer further questions in response to the previous answers.

In addition, conventional question-answering systems generate answers using learning data learned from a huge amount of question-answer data. On the other hand, in the present embodiment, it is possible to extract the answer to the current question by using the features related to the answer before the previous one. will be able to generate That is, in this embodiment, an appropriate answer can be realized even with a small amount of data.

[Configuration example of the present embodiment and its effect]
The information processing apparatus 100 is an information processing apparatus that sequentially provides answers to a series of questions using a question-and-answer data DB 200 (an example of a database) that stores information about features of questions and answers. The information processing apparatus 100 extracts features related to the first sentence of the series of questions based on the first sentence corresponding to the current question, and determines an answer to the current question based on the features and the question and answer data DB 200. An answer determination unit 130 (an example of a determination unit) is provided. If the answer to the current question cannot be determined based on the characteristics of the first sentence, the answer determination unit 130 determines the second sentence and the first sentence corresponding to the answers to the past questions before the current question. Based on this, each feature regarding the first sentence and the second sentence is extracted, and the answer to the current question is determined based on each feature and the question answer data DB 200 .

According to this configuration, even if the answer to this question cannot be determined using the features related to the current question, the answer to the current question is determined using the features related to the answers to the previous past question (features related to the second sentence). can do. As a result, even when an additional question such as "Why?" is issued by the questioner U1, an appropriate answer can be provided to the questioner U1 based on the content of the additional question. That is, it is possible to provide an appropriate answer to the question of questioner U1.

If the answer to the current question cannot be determined based on the characteristics of the first sentence (the sentence about the current question), the answer determination unit 130 determines the answer (the first answer) to the question one time before the current question. Extract the second sentence and features related to the first sentence and the second sentence based on the first sentence, and determine the answer to the current question based on these features and the question answer data DB 200 try.

According to this configuration, even if the answer to this question cannot be determined using the features related to the current question, the features related to the answer (first answer) to the question one time before the current question (features related to the second sentence) are used. can determine the answer to this question.

The answer determination unit 130 is the answer to each question from 1 time before the current question to N-1 (where N is an integer of 3 or more) times before. If the answer to this question cannot be determined using the sentence and the first sentence (sentence related to this question), the answers to each question from 1 to N times before this question (first answer to N-th answers) and based on the first sentence, extract each feature related to each sentence and the first sentence, and based on these features and the question and answer data DB 200, this time Attempt to determine the answer to the question.

According to this configuration, even if the answer to this question cannot be determined using the features related to the current question, using each feature related to the answers to each question from 1 to N times before the current question can determine the answer to

The answer determination unit 130 calculates the feature amount for each sentence based on each sentence corresponding to the first to Nth answers, and calculates the feature amount for each sentence as it goes back from the first answer to the Nth answer. The degree of influence of the quantity is reduced, and an attempt is made to determine the answer to the current question based on the feature quantity with the degree of influence reduced and the question-answer data DB 200 .

According to this configuration, it is possible to determine the answer to the current question by using the feature amount whose influence degree is lowered according to going back from the first answer to the Nth answer. In order to reduce the computational processing load, an attempt may be made to determine the answer without multiplying the feature amount relating to the first answer (or the first answer and the second answer) by a coefficient. In this case, if the feature amount related to the first answer (or the first answer and the second answer) cannot be determined for this question, each feature amount after the second answer (or after the third answer) is multiplied by a factor to attempt to determine the answer. As a result, it is possible to appropriately determine the answer to the current question, and to reduce the computational processing load associated with the process of determining the answer.

The question-and-answer data DB 200 is a database in which one or more words included in a question and answer information relating to the answer to the question are stored in association with each other. The answer determination unit 130 extracts one or more words included in the first sentence as features related to the first sentence, and based on the degree of matching between the one or more words and the words stored in the question and answer data DB 200 to determine the answer to this question.

According to this configuration, one or more words included in the sentence (first sentence) of the current question can be used as features to determine the answer to the current question.

The question-and-answer data DB 200 is a database in which feature values relating to questions and answer information relating to answers to the questions are stored in association with each other. The answer determination unit 130 extracts the feature amount related to the first sentence as the feature amount related to the first sentence, and based on the degree of matching between the feature amount and the feature amount stored in the question/answer data DB 200, determines the answer to the current question. to decide.

According to this configuration, it is possible to determine the answer to the current question using the feature amount related to the sentence (first sentence) of the current question.

The information processing apparatus 100 includes an answer response unit 140 that reproduces and outputs a portion corresponding to the answer determined by the answer determination unit 130 from the answer moving image in which the answers from the respondent U2 to the series of questions are recorded. further provide.

According to this configuration, since it is possible to play back the response video of the respondent U2 that has been shot in advance, it is possible to prevent the respondent U2 from having an awkward facial expression or an awkward response from the respondent U2. can be prevented. In addition, by playing back an answer moving image of answerer U2 that was actually shot, a realistic response can be realized.

The information processing system 600 is an information processing system that sequentially provides answers to a series of questions using the question-and-answer data DB 200 that stores information about the characteristics of questions and answers. The information processing system 600 includes an input device 620 (an example of an input unit) for inputting a series of questions from the questioner U1, and based on the first sentence corresponding to the current question among the series of questions, , and based on the feature and the question/answer data DB 200, the answer determination unit 130 (an example of the determination unit) that determines the answer to the current question, and the output device 630 that outputs the answer determined by the answer determination unit 130 (an example of an output unit). If the answer to the current question cannot be determined based on the characteristics of the first sentence, the answer determination unit 130 determines the second sentence corresponding to the answer to the past question before the current question and the first sentence. , and based on these features and the question/answer data DB 200, the answer to the current question is determined.

According to this configuration, even if the answer to this question cannot be determined using the features related to the current question, the answer to the current question is determined using the features related to the answers to the previous past question (features related to the second sentence). can do. That is, it is possible to realize an information processing system 600 capable of providing an appropriate answer to the question of questioner U1.

The information processing method according to the present embodiment is an information processing method for sequentially providing answers to a series of questions using the question-and-answer data DB 200 that stores information about the characteristics of questions and answers. This information processing method extracts features related to the first sentence of a series of questions based on the first sentence corresponding to the current question, and determines an answer to the current question based on the features and the question/answer data DB 200. decision processing (steps S503, S504, S506). In this determination process (steps S503, S504, S506), if the answer to the current question cannot be determined based on the characteristics of the first sentence, the second question corresponding to the answer to the past question before the current question is selected. Based on the sentence and the first sentence, features of the first and second sentences are extracted, and the answer to the current question is determined based on these features and the question/answer data DB 200 . Also, the program according to the present embodiment is a program that causes a computer to execute each of these processes. In other words, the program according to the present embodiment is a program that causes a computer to implement each function executable by the information processing apparatus 100 .

According to this configuration, even if the answer to this question cannot be determined using the features related to the current question, the answer to the current question is determined using the features related to the answers to the previous past question (features related to the second sentence). can do. That is, it is possible to realize an information processing method and program capable of providing an appropriate answer to the question of questioner U1.

Each processing procedure shown in this embodiment is an example for realizing this embodiment. Often, part of each processing procedure may be omitted or other processing procedures may be added.

Each process shown in this embodiment is executed based on a program for causing a computer to execute each processing procedure. Therefore, the present embodiment can also be understood as an embodiment of a program that realizes the function of executing each process and a recording medium that stores the program. For example, the program can be stored in the storage device of the information processing device by update processing for adding a new function to the information processing device. This enables the updated information processing apparatus to perform each process shown in this embodiment.

Although the embodiments of the present invention have been described above, the above embodiments merely show a part of application examples of the present invention, and the technical scope of the present invention is not limited to the specific configurations of the above embodiments. do not have.

100, 610, 710 information processing device, 110 input unit, 120 speech recognition unit, 130 answer determination unit, 140 answer response unit, 150 storage unit, 200 question and answer data DB, 300 video DB, 400 information processing system, 410 network, 430, 440 electronic equipment, 600, 700 information processing system, 620 input device, 630 output device, 640 storage device, 701 network, 711 communication unit, 720, 730 electronic equipment

Claims (8)

An information processing device that sequentially provides answers to a series of questions using a database that stores information about the characteristics of questions and answers,
a determination unit for extracting a feature related to the first sentence based on the first sentence corresponding to the current question among the series of questions, and determining an answer to the current question based on the feature and the database; ,
The decision unit
If the answer to the current question cannot be determined based on the characteristics of the first sentence, the second sentence and the first sentence corresponding to the first answer, which is the answer to the question one question before the current question. Extract each feature related to the first sentence and the second sentence based on and try to determine the answer to the current question based on each feature and the database,
Each sentence and the first If the answer to the current question cannot be determined even by using the sentence, each of the first to N-th answers, which are the answers to the questions from 1 to N times before the current question Based on the sentence and the first sentence, each feature related to each sentence and the first sentence is extracted, and an attempt is made to determine an answer to the current question based on each feature and the database.
Information processing equipment. The information processing device according to claim 1 ,
The determining unit calculates a feature amount related to each sentence based on each sentence corresponding to the first to Nth answers, and calculates each of the sentences corresponding to the first answer to the Nth answer. Attempt to determine the answer to the current question based on the feature amount with the reduced influence and the database,
Information processing equipment. The information processing device according to claim 1 or 2 ,
The database is a database in which one or more words included in the question and answer information related to the answer to the question are stored in association with each other,
The determination unit extracts one or more words included in the first sentence as features related to the first sentence, and based on the degree of matching between the one or more words and words stored in the database , to determine the answer to the current question;
Information processing equipment. The information processing device according to claim 1 or 2 ,
The database is a database in which the feature amount related to the question and the answer information related to the answer to the question are stored in association with each other,
The determining unit extracts a feature amount related to the first sentence as a feature related to the first sentence, and responds to the current question based on a degree of matching between the feature amount and the feature amount stored in the database. determine the
Information processing equipment. The information processing device according to claim 1 or 2 ,
An answer response unit that reproduces and outputs a portion corresponding to the answer determined by the determination unit from among the response videos in which the answers from the respondent to the series of questions are recorded,
Information processing equipment. An information processing system that sequentially provides answers to a series of questions using a database that stores information about question features and answers,
an input unit for inputting the series of questions;
a determination unit that extracts features related to the first sentence based on the first sentence corresponding to the current question among the series of questions, and determines an answer to the current question based on the features and the database;
An output unit for outputting the answer determined by the determination unit,
The decision unit
If the answer to the current question cannot be determined based on the characteristics of the first sentence, the second sentence and the first sentence corresponding to the first answer, which is the answer to the question one question before the current question. Extract each feature related to the first sentence and the second sentence based on and try to determine the answer to the current question based on each feature and the database,
Each sentence and the first If the answer to the current question cannot be determined even by using the sentence, each of the first to N-th answers, which are the answers to the questions from 1 to N times before the current question Based on the sentence and the first sentence, each feature related to each sentence and the first sentence is extracted, and an attempt is made to determine an answer to the current question based on each feature and the database.
Information processing system. An information processing method for sequentially providing answers to a series of questions using a database in which information about question characteristics and answers is stored , which is executed by a computer ,
Determination processing of extracting features related to the first sentence based on the first sentence corresponding to the current question among the series of questions, and determining an answer to the current question based on the features and the database ,
In the decision process,
If the answer to the current question cannot be determined based on the characteristics of the first sentence, the second sentence and the first sentence corresponding to the first answer, which is the answer to the question one question before the current question. Extract each feature related to the first sentence and the second sentence based on and try to determine the answer to the current question based on each feature and the database,
Each sentence and the first If the answer to the current question cannot be determined even by using the sentence, each of the first to N-th answers, which are the answers to the questions from 1 to N times before the current question Based on the sentence and the first sentence, each feature related to each sentence and the first sentence is extracted, and an attempt is made to determine an answer to the current question based on each feature and the database.
Information processing methods. A program to be executed by a computer that sequentially provides answers to a series of questions using a database that stores information about question characteristics and answers,
a determination process for extracting a feature related to the first sentence based on the first sentence corresponding to the current question among the series of questions, and determining an answer to the current question based on the feature and the database; let the computer run
In the decision process,
If the answer to the current question cannot be determined based on the characteristics of the first sentence, the second sentence and the first sentence corresponding to the first answer, which is the answer to the question one question before the current question. Extract each feature related to the first sentence and the second sentence based on and try to determine the answer to the current question based on each feature and the database,
Each sentence and the first If the answer to the current question cannot be determined even by using the sentence, each of the first to N-th answers, which are the answers to the questions from 1 to N times before the current question Based on the sentence and the first sentence, each feature related to each sentence and the first sentence is extracted, and an attempt is made to determine an answer to the current question based on each feature and the database.
program.

Images