JP2025040085A - Information processing device and program - Google Patents

Abstract

[Problem] To provide an appropriate answer to an input whose contents may not be included in training data. [Solution] An information processing device according to one aspect of the present disclosure includes an input acquisition unit that acquires a user input sentence from a user, a related text acquisition unit that acquires text related to the user input sentence from a publication published by a publisher, an auxiliary input generation unit that generates a question sentence by adding the user input sentence and auxiliary text requesting an answer to the user input sentence based on the text, and an answer output unit that inputs the question sentence into a trained model to generate and output an answer. [Selected Figure] Figure 1

Description

This disclosure relates to an information processing device and a program.

In recent years, interactive AI services using generative artificial intelligence (generative AI), such as ChatGPT, have become available.

For example, Patent Document 1 discloses a technology that uses such generative AI to provide answers to user questions while referencing appropriate documents.

JP 2023-76413 A

Conversational AI services are expected to be utilized in the field of education. However, since the learning data used in such conversational AI services is data publicly available on the Internet, appropriate answers may not be given to questions about content that is not publicly available on the Internet, and incorrect answers may be generated. In addition, a phenomenon called hallucination is known to occur, in which the generation AI generates answers that cannot be justified based on the data it has learned. If such incorrect answers or answers that cannot be justified are presented to the user, there is a risk that the user will not be able to learn effectively.

In view of the above, the present disclosure aims to provide a technology that can provide appropriate answers to input content that may not be included in the learning data.

An information processing device according to one aspect of the present disclosure includes an input acquisition unit that acquires a user input sentence from a user, a related text acquisition unit that acquires text related to the user input sentence from a publication published by a publisher, an auxiliary input generation unit that generates a question sentence by adding the text and auxiliary text requesting an answer to the user input sentence based on the text to the user input sentence, and an answer output unit that inputs the question sentence into a trained model to generate and output an answer.

According to one aspect of the present disclosure, text related to a user input sentence that is found in a publication can be obtained, and an answer based on that text can be generated and output, making it possible to provide an appropriate answer to an input whose content may not be included in the training data.

FIG. 1 is a block diagram showing a schematic configuration example of an education support system according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a hardware configuration of a user device according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a functional configuration of a user device according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a hardware configuration of a user information management device according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a functional configuration of a user information management device according to an embodiment of the present disclosure. FIG. 1 is a diagram illustrating an example of user information stored in a user information storage unit of a user information management device according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a hardware configuration of an answer generation device according to an embodiment of the present disclosure. FIG. 1 is a block diagram illustrating an example of a functional configuration of an answer generation device according to an embodiment of the present disclosure. FIG. 1 is a sequence diagram illustrating an example of an operation of an education support system according to an embodiment of the present disclosure. 1 is a flowchart illustrating an example of the operation of a user device according to an embodiment of the present disclosure. 1 is a flowchart illustrating an example of the operation of a user device according to an embodiment of the present disclosure. A flowchart showing an example of the operation of a user information management device according to an embodiment of the present disclosure. 1 is a flowchart illustrating an example of an operation of an answer generating device according to an embodiment of the present disclosure. FIG. 1 illustrates an example display screen including inputs and answers displayed on a user device according to an embodiment of the present disclosure.

Embodiments of the present disclosure will be described below with reference to the drawings. However, the scope of the invention is not limited to the examples shown in the drawings. In the following description, parts having the same functions and configurations are given the same reference numerals and their description will be omitted.

(Embodiment)
＜Overview＞
According to an embodiment of the present disclosure, an information processing device (e.g., a user device 10 described later) includes an input acquisition unit, a related text acquisition unit, an auxiliary input generation unit, and an answer output unit. The input acquisition unit acquires a user input sentence from a user. The related text acquisition unit acquires text related to the user input sentence from a publication published by a publisher. The auxiliary input generation unit generates a question sentence by adding the acquired text and an auxiliary sentence requesting an answer to the acquired user input sentence based on the acquired text to the acquired user input sentence. The answer output unit inputs the generated question sentence to a trained model (e.g., a large-scale language model) to generate and output an answer.

In this way, the information processing device generates a question that requests an answer based on related text from a publication that is considered to be more reliable than information on the Internet in response to a user input sentence from a user, inputs the question into a trained model, and outputs an answer. This makes it possible to provide an appropriate answer to an input whose content may not be included in the training data.

<Educational support system configuration>
FIG. 1 is a block diagram illustrating a schematic configuration example of a teaching assistance system according to an embodiment of the present disclosure.

As shown in FIG. 1, the education support system 1 includes a user device 10, a user information management device 20, an answer generation device 30, a publisher database (DB) device 40, and a web device 50. The user device 10 and the user information management device 20 are communicatively connected to each other via a network 60, the user device 10 and the answer generation device 30 are communicatively connected to each other via the network 60, and the answer generation device 30, the publisher DB device 40, and the web device 50 are communicatively connected to each other via the network 60. The network 60 may be a LAN (Local Area Network), the Internet, a VPN (Virtual Private Network), a mobile communication network, or a combination of these.

The educational support system 1 presents the user (learner) with a response to an input from the user of the user device 10.

The user device 10 has a configuration that allows it to connect to the user information management device 20, the answer generation device 30, the publisher DB device 40, and the web device 50, for example, via a dedicated application or a web browser installed on the user device 10. For example, the following processing of the user device 10 may be performed via a dedicated application or a web browser.

In order to use the educational support service (or question and answer service) provided by the educational support system 1, the user of the user device 10 pre-registers user information including a user ID and password (hereinafter also referred to as login information) in the user information management device 20.

The user device 10 transmits the user's login information input by the user of the user device 10 to the user information management device 20 via the network 60. If authentication is successful in the authentication process using the login information by the user information management device 20, the user of the user device 10 can use the educational support service.

After successful authentication, the user device 10 receives user information about the user from the user information management device 20 via the network 60. The user device 10 then accepts input of questions, etc., by voice, text, etc., from the user of the user device 10. The user device 10 then generates a search query based on the accepted input in response to the user's instructions (e.g., pressing a specific button), and queries the publisher DB device 40 and/or the web device 50 via the network 60 using the generated search query.

In this embodiment, the user device 10 first queries the publisher DB device 40 using a search query. If a positive search result is obtained (in other words, information (or text or content) related to the search query is found in the teaching materials stored in the publisher DB device 40), the user device 10 obtains information related to the search query (and therefore the user's input) from the teaching materials stored in the publisher DB device 40 as a search result. On the other hand, if a positive search result is not obtained (in other words, a negative search result is obtained or information related to the search query is not found in the teaching materials stored in the publisher DB device 40), then the answer generating device 30 queries the web device 50 using the search query, for example, via an API (Application Programming Interface). The answer generating device 30 generates an answer to the user's input based on the search result by the web device 50. Note that the above process is executed when the user has the authority to access the learning materials stored in the publisher DB device 40. If the user does not have the authority to access the learning materials stored in the publisher DB device 40, an inquiry is made to the web device 50 without making an inquiry to the publisher DB device 40.

Based on the search results for the search query, the user device 10 may generate a final input by adding text related to the search query to the received input (if a positive search result is obtained), or may use the received input as the final input (if no positive search result is obtained). The user device 10 then transmits the final input to the answer generation device 30 via the network 60. Note that in this embodiment, it is assumed that the business entity providing the answer generation device 30 is permitted, for example by contract, to access (read) electronic learning materials such as textbooks (an example of a publication, also simply referred to as learning materials) stored in the publisher DB device 40.

The user device 10 obtains the answer by receiving the answer to the final input transmitted from the answer generating device 30 via the network 60. The user device 10 then outputs the answer to the final input (for example, displays the answer to the input on a display device (display unit) provided in the user device 10).

Note that although FIG. 1 shows one user device 10, multiple user devices 10 may exist. The user device 10 may also be called an information terminal, a user terminal, a terminal, a client, a client device, etc.

The user information management device 20 is installed, for example, in a cloud environment.

The user information management device 20 receives login information of the user of the user device 10 from the user device 10 via the network 60. The user information management device 20 executes authentication processing using the received login information.

After the authentication is successful (if the received login information matches the registered login information), the user information management device 20 transmits the registered user information to the user device 10 via the network 60. The user information will be described later.

The answer generating device 30 is installed, for example, in a cloud environment.

The answer generating device 30 generates an answer to an input from a user. The answer generating device 30 receives a final input from the user device 10 via the network 60. The answer generating device 30 generates an answer to the received final input using a trained large language model (LLM: Large Language Model). The answer generating device 30 transmits the generated answer to the user information management device 20 via the network 60.

The answer generating device 30 may be a device that provides an existing conversational AI service, such as ChatGPT provided by OpenAI or Bard provided by Google (registered trademark). In this case, the user device 10 can use an API for using such a conversational AI service to transmit a final input to the answer generating device 30 and receive an answer to the final input from the answer generating device 30. Alternatively, the answer generating device 30 may provide a conversational AI service by training an LLM as described above using training data such as text data published on the web.

The publisher DB device 40 is operated by a publisher and stores electronic learning materials published by the publisher. The publisher DB device 40 may exist on a publisher basis, or may exist as a device that provides learning materials in a format that compiles learning materials from multiple publishers.

The web device 50 is, for example, a device that provides a search service provided by Google (registered trademark) or the like.

<User device configuration>
2 is a block diagram illustrating an example of a hardware configuration of a user device according to an embodiment of the present disclosure. The user device 10 may be realized by any computer or information processing device such as a tablet, a smartphone, or a notebook PC (Personal Computer).

For example, as shown in FIG. 2, the user device 10 may include a storage device 11, a processing device 12, a user interface (UI) device 13, a communication device 14, and a bus 15. The storage device 11, the processing device 12, the UI device 13, and the communication device 14 may be connected to each other via the bus 15.

The programs or instructions that realize the functions and processes of the user device 10 described above and below may be downloaded from some external device (e.g., a server) via a network or the like. Such programs or instructions may also be provided from a removable storage medium such as a CD-ROM (Compact Disc Read Only Memory) or a flash memory.

The storage device 11 is realized by a RAM (Random Access Memory), a flash memory, a hard disk drive, etc., and stores installed programs or instructions as well as files, data, etc. used to execute the programs or instructions. The storage device 11 may also include a non-transitory storage medium.

The processing device 12 may be realized, for example, by a general-purpose processor or controller (circuit), or may be realized by a dedicated processor or controller (circuit). When realized by a general-purpose processor, the processing device 12 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores. In this case, the processing device 12 executes the functions and processing of the user device 10 described above and below according to the programs or instructions stored in the storage device 11, data such as parameters used to execute the programs or instructions, etc. Such a program, when executed by the computer or processing device 12, may be a program that causes the computer or processing device 12 to function as the user device 10 according to the present disclosure. In addition, such a program, when executed by the computer or processing device 12, may be a program that causes the computer or processing device 12 to execute the method related to educational support (question answering) according to the present disclosure.

The UI device 13 may be composed of input devices such as a keyboard, mouse, camera, microphone, etc., output devices such as a display, speaker, headset, printer, etc., and input/output devices such as a haptic device, touch panel, etc., and realizes an interface between the user and the user device 10. For example, the user operates the user device 10 by operating a graphical user interface (GUI) displayed on a display or touch panel using a keyboard, mouse, etc.

The communication device 14 may be realized by various communication circuits that perform communication processing with external devices, the Internet, LAN, VPN, and other communication networks.

The above-described hardware configuration of the user device 10 is merely an example, and the user device 10 according to the present disclosure may be realized by any other suitable hardware configuration.

Figure 3 is a block diagram showing an example of the functional configuration of a user device according to one embodiment of the present disclosure.

3, the user device 10 includes an input receiving unit 101, a user information acquiring unit 102, a related text acquiring unit 103, a final input/output unit 104, an answer acquiring unit 105, and a display control unit 106. The input receiving unit 101, the user information acquiring unit 102, the related text acquiring unit 103, the final input/output unit 104, the answer acquiring unit 105, and the display control unit 106 may be realized by the processing device 12 as hardware, as software (e.g., a dedicated application), or as a combination of hardware and software.

The input acceptance unit 101 accepts input (e.g., user information, login information, questions, etc.) from a user of the user device 10. If the input is user information or login information, the input acceptance unit 101 transmits the accepted input to the user information management device 20 via the communication device 14. If the input is neither user information nor login information (e.g., if the input is a question), the input acceptance unit 101 outputs the accepted input to the related text acquisition unit 103. The input acceptance unit may be referred to as an input acquisition unit.

The user information acquisition unit 102 acquires user information by receiving user information related to the user from the user information management device 20 via the communication device 14. The user information acquisition unit 102 outputs at least a portion of the acquired user information (e.g., "age" and "authority" shown in FIG. 6) to the related text acquisition unit 103 and the final input/output unit 104.

The related text acquisition unit 103 generates one or more search queries based on the input output by the input reception unit 101 in response to a user instruction (for example, pressing a predetermined button (for example, button 1402 shown in FIG. 14)). When generating one search query, the related text acquisition unit 103 may use the input output by the input reception unit 101 as the search query as is. When generating multiple search queries, the related text acquisition unit 103 may generate multiple search queries from the input output by the input reception unit 101, taking into account spelling variations and related words. For example, the related text acquisition unit 103 may instruct the answer generation device 30 to generate a sentence similar to the input, and generate the sentence (answer) obtained from the answer generation device 30 as the search query, or may use BERT (Bidirectional Encoder Representations from Transformers) or the like to identify synonyms for each word constituting the input, and generate a sentence in which the word is replaced with the synonym as the search query.

For example, if the user's input (question) is "Why did Toyotomi Hideyoshi defeat Akechi Mitsuhide?", the related text acquisition unit 103 can use the above-mentioned method to generate multiple search queries, such as not only "Why did Toyotomi Hideyoshi defeat Akechi Mitsuhide?" but also "Why did Toyotomi Hideyoshi defeat Akechi Mitsuhide?", "Why did Hashiba Hideyoshi defeat Akechi Mitsuhide?", "Why did Hashiba Hideyoshi defeat Akechi Samanosuke?", etc.

The related text acquisition unit 103 queries the publisher DB device 40 and/or the web device 50 via the communication device 14 using the generated search query.

At this time, based on the user information output by the user information acquisition unit 102, if the "authority" shown in FIG. 6 is "yes", the related text acquisition unit 103 first uses a search query to make an inquiry to the publisher DB device 40 that is likely (reliable) to obtain an appropriate answer.

For example, the related text acquisition unit 103 may use BERT or the like to calculate the feature amounts of the search query and the sentences (texts) in the teaching materials to be searched, and when the difference between these feature amounts is smaller than a predetermined value, it may be estimated that text related to the search query (input) exists in the teaching materials (i.e., the text is related to the search query). Alternatively, as another example, the related text acquisition unit 103 may calculate the feature amounts of the search query and the sentences (texts) in the teaching materials to be searched, and when the feature amounts are larger or smaller than a predetermined value, it may be estimated that text related to the search query exists in the teaching materials (i.e., the text is related to the search query). For example, the related text acquisition unit 103 may use tf-idf (term frequency - inverse document frequency) to replace the word groups in the search query and the sentences in the teaching materials with feature vectors based on numerical values, calculate the similarity (e.g., cosine similarity) of the feature vectors, and when the cosine similarity is larger than a predetermined value (e.g., 0.8, etc.), acquire the sentences (texts) as related to the search query. Alternatively, as another example, the related text acquisition unit 103 may acquire text within the learning material that is related to the search query by using a GPT Index, LangChain, sentence-transformers model, or the like.

In addition, in the publisher DB device 40, when the teaching materials are divided by age (for 11-year-olds, for 13-year-olds, etc.) or by grade (for 5th grade elementary school students, for 1st grade junior high school students, etc.), the related text acquisition unit 103 may select teaching materials (or teaching material DB) divided by age or grade based on the user information (for example, "age" shown in FIG. 6) output by the user information acquisition unit 102. For example, the related text acquisition unit 103 may select teaching materials for the target age or grade, or may select teaching materials for the target age or grade and teaching materials for ages or grades below the target age or grade. More specifically, when the user information (target grade) is "3rd grade junior high school students," the related text acquisition unit 103 may select teaching materials for 3rd grade junior high school students, or may select teaching materials for 1st grade elementary school students to 3rd grade junior high school students.

If it is estimated that text related to the search query exists in the teaching materials stored in the publisher DB device 40, the related text acquisition unit 103 acquires the text as a search result for the search query. In this way, the related text acquisition unit 103 acquires text related to the input from the teaching materials. The related text acquisition unit 103 outputs the acquired search results together with the input output by the input receiving unit 101 to the final input output unit 104.

If it is estimated that the text related to the search query does not exist in the teaching materials stored in the publisher DB device 40, or if the "authority" shown in FIG. 6 is "no" based on the user information output by the user information acquisition unit 102, the related text acquisition unit 103 queries the web device 50 using the search query. Then, the related text acquisition unit 103 accesses the top K (e.g., K=1) web pages ranked as the results of the search by the web device 50, and extracts and acquires text related to the search query as the search result, as described above with respect to the publisher DB device 40. In this way, the related text acquisition unit 103 acquires text related to the input from the Internet. The related text acquisition unit 103 outputs the acquired search result together with the input output by the input acceptance unit 101 to the final input output unit 104.

If the query to the publisher DB device 40 and the web device 50 does not yield a positive search result for the search query, the response processing unit 302 outputs only the input output by the input receiving unit 101 to the final input output unit 104.

When a positive search result is obtained for the search query, the final input output unit 104 generates a final input (question sentence) by adding the positive search result output by the related text acquisition unit 103 and an auxiliary sentence to the input requesting an answer using the search result for the input output by the related text acquisition unit 103 (an answer based on the search result for the input) to the input output by the related text acquisition unit 103 to the input output by the related text acquisition unit 103. On the other hand, when a positive search result is not obtained for the search query, the final input output unit 104 decides to use the input output by the related text acquisition unit 103 as the final input. The final input output unit 104 transmits the final input to the answer generation device 30 via the communication device 14. The final input output unit 104 may instruct the answer generation device 30 to generate one answer for each learning material (in this case, text related to each learning material is sent to the answer generation device 30, and multiple answers can be generated for one input), or may instruct the answer generation device 30 to generate one answer for one input (in this case, text related to each learning material is compiled and sent to the answer generation device 30). In the following, "send" may be interpreted as "output" as appropriate. The final input output unit 104 may be referred to as an input sending unit.

If the search results are too long (for example, if the search results are longer than a predetermined number of characters or if the data is longer than a predetermined number of bytes), the related text acquisition unit 103 or the final input/output unit 104 may instruct the answer generation device 30 to summarize the search results and acquire a summary sentence (answer) from the answer generation device 30. Here, the answer generation device 30 generates a summary sentence by inputting the search results and a sentence requesting a summary sentence of the search results to the LLM. Then, for example, the related text acquisition unit 103 may output the summary sentence to the final input/output unit 104, and the final input/output unit 104 may generate a final input in the same manner as above by adding the summary sentence output by the related text acquisition unit 103 instead of the search results to the input output by the related text acquisition unit 103, and transmit the final input to the answer generation device 30.

In addition, the final input output unit 104 may add auxiliary text to the input output by the related text acquisition unit 103, based on the user information (e.g., "age" shown in FIG. 6) output by the user information acquisition unit 102, requesting an answer appropriate for the target age or grade level. More specifically, if the user information (grade level) is "third year junior high school student," the final input output unit 104 may add auxiliary text such as "Please provide an answer appropriate for a third year junior high school student (15 years old)" to the input output by the related text acquisition unit 103.

The answer acquisition unit 105 acquires an answer by receiving an answer to the final input transmitted by the final input output unit 104 from the answer generation device 30 via the communication device 14. The answer acquisition unit 105 outputs the acquired answer to the display control unit 106.

The display control unit 106 generates a screen including the answer output by the answer acquisition unit 105, and displays the screen on a display device (display unit) included in the UI device 13. In other words, the display control unit 106 outputs the answer (to the outside). The answer may include the teaching material that is the source (reference source) of the answer, its name, the page on which the related text exists, the related text, etc. (see FIG. 14).

<Configuration of User Information Management Device>
4 is a block diagram illustrating an example of a hardware configuration of a user information management device according to an embodiment of the present disclosure. The user information management device 20 may be realized by a computing device (computer) such as a server or an information processing device.

For example, as shown in FIG. 4, the user information management device 20 may include a storage device 21, a processing device 22, a UI device 23, a communication device 24, and a bus 25. The storage device 21, the processing device 22, the UI device 23, and the communication device 24 may be connected to each other via the bus 25.

The programs or instructions that realize the functions and processes of the user information management device 20 described above and below may be downloaded from some external device (e.g., a server) via a network or the like. Such programs or instructions may also be provided from a removable storage medium such as a CD-ROM or flash memory.

The storage device 21 is realized by RAM, flash memory, a hard disk drive, etc., and stores installed programs or instructions as well as files, data, etc. used to execute the programs or instructions. The storage device 11 may include a non-transitory storage medium.

The processing device 22 may be realized, for example, by a general-purpose processor or controller (circuit), or may be realized by a dedicated processor or controller (circuit). When realized by a general-purpose processor, the processing device 22 may be realized by one or more CPUs, GPUs, processing circuits, etc., which may be composed of one or more processor cores. In this case, the processing device 22 executes the functions and processing of the user information management device 20 described above and below according to the programs or instructions stored in the storage device 21, data such as parameters used to execute the programs or instructions, etc. Such a program may be a program that causes the computer or processing device 22 to function as the user information management device 20 according to the present disclosure when executed by the computer or processing device 22. In addition, such a program may be a program that causes the computer or processing device 22 to execute the method related to educational support (question answering) according to the present disclosure when executed by the computer or processing device 22.

The UI device 23 may be composed of input devices such as a keyboard, mouse, camera, microphone, etc., output devices such as a display, speaker, headset, printer, etc., and input/output devices such as a haptic device, touch panel, etc., and realizes an interface between the user and the user information management device 20. For example, the user operates the user information management device 20 by operating a GUI displayed on a display or touch panel using a keyboard, mouse, etc.

The communication device 24 may be realized by various communication circuits that perform communication processing with external devices, the Internet, LAN, VPN, and other communication networks.

The hardware configuration of the user information management device 20 described above is merely an example, and the user information management device 20 according to the present disclosure may be realized by any other suitable hardware configuration.

Figure 5 is a block diagram showing an example of the functional configuration of a user information management device according to one embodiment of the present disclosure.

For example, as shown in FIG. 5, the user information management device 20 includes an input acquisition unit 201, a user information registration unit 202, an authentication process execution unit 203, a user information output unit 204, and a user information storage unit 211. The input acquisition unit 201, the user information registration unit 202, the authentication process execution unit 203, and the user information output unit 204 may be realized as hardware, software, or a combination of hardware and software by the processing device 22. The user information storage unit 211 may be realized by the storage device 21.

The input acquisition unit 201 acquires input by receiving input (user information or login information) by the user of the user device 10 from the user device 10 via the communication device 24. If the input is user information, the input acquisition unit 201 outputs the input to the user information registration unit 202. If the input is login information, the input acquisition unit 201 outputs the input to the authentication process execution unit 203.

The user information registration unit 202 registers the user information by storing the user information output by the input acquisition unit 201 in the user information storage unit 211.

The authentication process execution unit 203 compares the login information output by the input acquisition unit 201 with the login information among the user information stored in the user information storage unit 211, and if they match, determines that authentication has been successful. If authentication is successful, the authentication process execution unit 203 outputs all of the user information corresponding to the login information or all of the user information other than the login information to the user information output unit 204. If authentication is unsuccessful, the authentication process execution unit 203 transmits a message to the user device 10 via the network 60 indicating that authentication has failed.

The user information output unit 204 transmits the user information output by the authentication process execution unit 203 to the user device 10 via the network 60.

The user information storage unit 211 stores user information about the user of the user device 10.

Figure 6 shows an example of user information stored in the user information storage unit 211.

As shown in FIG. 6, user information 250 may include a user ID, password, last name, first name, age, nationality, the local government where the user resides, authority, etc. User information 250 may further include any information related to the user, such as the user's grade, academic level (e.g., deviation score), etc., and is not limited to the information shown in the figure. User information 250 may not include a password, in which case there may be a separate table that stores user IDs and passwords.

In this embodiment, the "authority" shown in FIG. 6 means the access authority (read authority) to the teaching materials (i.e., copyrighted work) published by a publisher that publishes textbooks adopted in the local government where the user lives. For example, when a user (e.g., the user's guardian) registers or changes user information, the user can obtain the access authority (read authority) to the teaching materials published by the publisher in response to agreeing to the conditions required by the publisher (e.g., provision of various personal information, payment of a fee, etc.) via the user information management device 20. In the illustrated example, "authority yes" means that the user has the access authority to the teaching materials published by the publisher. In the case of "authority yes," the publisher DB device 40 corresponding to the publisher is accessed in response to the user's input (question), and the teaching materials stored in the publisher DB device 40 are referenced, and if text related to the user's input (question) is found in the teaching materials, an answer based on the related text is presented to the user. On the other hand, "authority no" means that the user does not have the access authority to the teaching materials published by the publisher. In the case of "no authority", access to the publisher DB device 40 corresponding to the publisher is not permitted in response to the user's input (question) (or, although the publisher DB device 40 corresponding to the publisher is accessed and the teaching materials stored in the publisher DB device 40 are referenced, even if text related to the user's input (question) is found in the teaching materials, an answer based on the related text is not presented to the user due to copyright restrictions). Note that the user may be able to select a publisher with which to make a contract so that the user can obtain access to teaching materials published by a publisher other than the publisher that publishes the textbook adopted by the local government in which the user lives. Also, if the school, cram school, etc. to which the user belongs has a contract with a publisher, the authority may be automatically set to "yes".

The user device 10 that receives the user information 250 from the user information management device 20 may transmit a part or all of the received user information 250 to the answer generation device 30.

<Configuration of Answer Generation Device>
7 is a block diagram showing an example of a hardware configuration of an answer generating device according to an embodiment of the present disclosure. The answer generating device 30 may be realized by a computing device (calculator, computer) such as a server or an information processing device.

For example, as shown in FIG. 7, the answer generating device 30 may include a storage device 31, a processing device 32, a UI device 33, a communication device 34, and a bus 35. The storage device 31, the processing device 32, the UI device 33, and the communication device 34 may be connected to each other via the bus 35.

The programs or instructions that realize the functions and processes of the answer generating device 30 described above and below may be downloaded from some external device (e.g., a server) via a network or the like. Such programs or instructions may also be provided from a removable storage medium such as a CD-ROM or flash memory.

The storage device 31 is realized by RAM, flash memory, a hard disk drive, etc., and stores installed programs or instructions as well as files, data, etc. used to execute the programs or instructions. The storage device 31 may also include a non-transitory storage medium.

The processing device 32 may be realized, for example, by a general-purpose processor or controller (circuit), or may be realized by a dedicated processor or controller (circuit). When realized by a general-purpose processor, the processing device 32 may be realized by one or more CPUs, GPUs, processing circuits, etc., which may be composed of one or more processor cores. In this case, the processing device 32 executes the functions and processing of the answer generating device 30 described above and below according to the programs or instructions stored in the storage device 31, data such as parameters used to execute the programs or instructions, etc. Such a program, when executed by the computer or processing device 32, may be a program that causes the computer or processing device 32 to function as the answer generating device 30 according to the present disclosure. Furthermore, such a program, when executed by the computer or processing device 32, may be a program that causes the computer or processing device 32 to execute the method related to educational support (question answering) according to the present disclosure.

The UI device 33 may be composed of input devices such as a keyboard, mouse, camera, microphone, etc., output devices such as a display, speaker, headset, printer, etc., and input/output devices such as a haptic device, touch panel, etc., and realizes an interface between the user and the answer generation device 30. For example, the user operates the answer generation device 30 by operating a GUI displayed on a display or touch panel using a keyboard, mouse, etc.

The communication device 34 may be realized by various communication circuits that perform communication processing with external devices, the Internet, LAN, VPN, and other communication networks.

The hardware configuration of the answer generating device 30 described above is merely an example, and the answer generating device 30 according to the present disclosure may be realized by any other suitable hardware configuration.

Figure 8 is a block diagram showing an example of the functional configuration of an answer generation device according to an embodiment of the present disclosure.

For example, as shown in FIG. 8, the answer generation device 30 includes an input acquisition unit 301 and an answer processing unit 302. The input acquisition unit 301 and the answer processing unit 302 may be realized by the processing device 32 as hardware, as software, or as a combination of hardware and software.

The input acquisition unit 301 acquires the final input by receiving the final input from the user device 10 via the communication device 34. The input acquisition unit 301 outputs the final input to the answer processing unit 302.

The answer processing unit 302 generates an answer to the final input by inputting the final input output by the input acquisition unit 301 into the LLM, which is a trained model. Here, if the final input includes text related to the search query (input from the user), the answer processing unit 302 generates an answer based on the text related to the search query. Note that the answer processing unit 302 may include in the answer it generates the teaching material that is the source (reference source) of the answer, its name, the page where the related text exists, the related text, etc.

The answer processing unit 302 transmits the generated answer to the user device 10 via the communication device 34.

<Operation of the Education Support System>
Next, an example of the operation of the education support system 1 will be described with reference to FIGS.

Figure 9 is a sequence diagram showing an example of the operation of an education support system according to an embodiment of the present disclosure. The example shown in Figure 9 relates to a user information registration process.

In step S101, the user device 10 (input receiving unit 101) receives user information, which is input from the user of the user device 10.

In step S102, the user device 10 (input reception unit 101) transmits the received user information to the user information management device 20.

In step S103, the user information management device 20 (user information registration unit 202) registers the user information by storing the received user information in the user information storage unit 211.

Figure 10 is a flowchart showing an example of the operation of a user device according to one embodiment of the present disclosure.

In step S201, the user device 10 (input receiving unit 101) receives login information input from the user of the user device 10.

In step S202, the user device 10 (input reception unit 101) transmits the received login information to the user information management device 20. Here, the user information management device 20 executes an authentication process using the login information, and assumes that the authentication is successful in the authentication process.

In step S203, the user device 10 (user information acquisition unit 102) acquires user information from the user information management device 20.

In step S204, the user device 10 (input receiving unit 101) receives input (question) from the user of the user device 10.

In step S205, the user device 10 (related text acquisition unit 103) executes a search based on the input from the user and attempts to acquire related text. Details of step S205 will be described later with reference to FIG. 11.

In step S206, the user device 10 (final input output unit 104) generates a final input by adding the above-mentioned auxiliary text to the input from the user based on the results of the search in step S205, and transmits the generated final input to the answer generation device 30.

In step S207, the user device 10 (answer acquisition unit 105) acquires the answer from the answer generation device 30.

In step S208, the user device 10 (display control unit 106) displays the answer on the display unit.

After this, steps S204 to S208 can be executed repeatedly.

FIG. 11 is a flowchart illustrating an example of the operation of a user device according to an embodiment of the present disclosure. This example shows details of step S205 shown in FIG. 10.

In step S251, the user device 10 (related text acquisition unit 103) generates one or more search queries based on input from the user.

In step S252, the user device 10 (related text acquisition unit 103) determines, based on the user information, whether the user has the authority to access the publications (teaching materials) stored in the publisher DB device 40.

On the other hand, if the user has access rights to the publication (step S252; YES), in step S253, the user device 10 (related text acquisition unit 103) queries the publisher DB device 40 using the generated search query.

In step S254, the user device 10 (related text acquisition unit 103) determines whether a positive search result has been obtained (whether text related to the search query (input from the user) exists in the teaching materials).

If a positive search result is obtained (step S254; YES), in step S255, the user device 10 (related text acquisition unit 103) acquires text related to the user's input from the publication. The flow then proceeds to step S206.

If no positive search results are obtained (step S254; NO), or if the user does not have permission to access the publication (step S252; NO), in step S256, the user device 10 (related text acquisition unit 103) queries the web device 50 using the search query. As a result of querying the web device 50, positive search results may or may not be obtained, but here it is assumed that positive search results are obtained.

In step S257, the user device 10 (related text acquisition unit 103) acquires text related to the user's input from the Internet. The flow then proceeds to step S206.

Figure 12 is a flowchart showing an example of the operation of a user information management device according to one embodiment of the present disclosure.

In step S301, the user information management device 20 (input acquisition unit 201) accepts login information, which is input from the user of the user device 10, from the user device 10.

In step S302, the user information management device 20 (authentication process execution unit 203) executes authentication processing based on the received login information and the login information stored in the user information storage unit 211. Here, it is assumed that authentication is successful in the authentication processing.

In step S303, the user information management device 20 (user information output unit 204) transmits the user information to the user device 10.

Figure 13 is a flowchart showing an example of the operation of an answer generation device according to one embodiment of the present disclosure.

In step S401, the answer generation device 30 (input acquisition unit 301) acquires the (final) input from the user device 10.

In step S402, the answer generation device 30 (answer processing unit 302) generates an answer to the acquired (final) input by inputting the acquired input into the LLM, which is a trained model.

In step S403, the answer generation device 30 (answer processing unit 302) transmits the generated answer to the user device 10.

FIG. 14 illustrates an example display screen including inputs and answers displayed on a user device according to one embodiment of the present disclosure.

As shown in FIG. 14, an example display screen 1400 displayed on the display unit of the user device 10 includes an input field 1401, an answer generation instruction button 1402, a response sentence field 1403, and a related teaching material display area 1404.

When the user inputs a question (user input sentence) into the input field 1401, the user device 10 accepts the question (step S204 in FIG. 10). Next, when the user presses the answer generation instruction button 1402, the user device 10 generates a search query based on the question and queries the publisher DB device 40 using the search query (step S205 in FIG. 10, steps S251 and S252 in FIG. 11). In the example shown in FIG. 14, it is assumed that text related to the question from the user is found in teaching material A and teaching material B (publications). Then, the user device 10 acquires text related to the question (the highlighted portion and the speech bubble portion in the related teaching material display area 1404 in FIG. 14) from teaching material A and teaching material B (step S254 in FIG. 11). Then, the user device 10 generates a final input that adds the related text to the question and an auxiliary sentence requesting an answer based on the related text to the question, and transmits the final input to the answer generation device 30 (step S206 in FIG. 10). When the answer generation device 30 receives (acquires) the final input, generates an answer, and transmits it to the user device 10 (steps S401 to S403 in FIG. 13), the user device 10 acquires the answer, displays the response sentence that is part of the answer in the response sentence field 1403, and displays predetermined information about teaching materials A and B (teaching material name, page, related text, etc.) in the related teaching material display area 1404 (steps S207 and S208 in FIG. 10).

Generally, generation AI has a limit to the length of text that can be input, making few shots using the contents of teaching material data published by a publisher unrealistic. Furthermore, fine-tuning generation AI using the contents of such teaching material data imposes a large burden and cost. On the other hand, according to this embodiment, for input of content not included in the learning data, a more accurate answer based on the contents of external teaching material data can be appropriately provided without few shot generation or fine tuning, contributing to an increase in the variety and quality of question answer response content that can be generated.

<Modification>
Modifications of the above embodiment will now be described.

[Modification 1]
Some of the functions of the user device 10 may be included in the user information management device 20 or the answer generation device 30. In this case, the user device 10 may transmit an input such as a question sentence from a user to the user information management device 20 or the answer generation device 30, and the user information management device 20 or the answer generation device 30 that receives the input may execute processing equivalent to the processing executed by the user information acquisition unit 102, the related text acquisition unit 103, the final input/output unit 104, and/or the answer acquisition unit 105.

[Modification 2]
The function related to answer generation of the answer generating device 30 may be included in the user device 10 or the user information management device 20. That is, the LLM, which is a trained model, may be included in the user device 10, or the user information management device 20, the answer generating device 30, etc. may be arranged on a network to which the user device 10 can connect. Therefore, the user device 10 or the user information management device 20 may acquire a user input sentence from a user, acquire text related to the user input sentence from a publication published by a publisher, generate a question sentence by adding the text and an auxiliary sentence to the user input sentence that requests an answer based on the text to the user input sentence, and input the question sentence into the trained model to generate and output an answer.

[Modification 3]
In the above, it has been explained that information about the teaching material in which the related text is written (such as the name of the teaching material, page, etc.) is displayed on the user device 10, but additionally or alternatively, link information about the teaching material (information indicating the location where the teaching material is located in the publisher DB device 40) may be included in the answer and displayed on the user device 10.

[Other Modifications]
The names of the above-mentioned devices (user device 10, user information management device 20, etc.) and the functional units constituting the devices (input receiving unit 101, user information acquisition unit 102) are merely examples and may be changed as appropriate. Furthermore, two or more of the functional units constituting the devices may be combined and integrated, or one functional unit may be divided into functional units. Furthermore, the transmission function and/or reception function possessed by each functional unit may be summarized as a notification unit, a notified unit, a communication unit, a transmission/reception unit, a transmission unit, a reception unit, etc.

The order of steps in the flowcharts shown above is merely an example and may be rearranged as appropriate, and some steps may be performed in parallel.

As described above, according to one aspect of the present disclosure, it is possible to provide appropriate answers to inputs whose content may not be included in the learning data.

Although the embodiments have been described above with reference to the drawings, the present disclosure is not limited to such examples. It is clear that a person skilled in the art can conceive of various modified or amended examples within the scope of the claims. It is understood that such modified or amended examples also fall within the technical scope of the present disclosure. Furthermore, each component or each item in the embodiments (including modified or amended examples) may be combined in any manner without departing from the spirit of the present disclosure.

REFERENCE SIGNS LIST 1 Education support system 10 User device 101 Input reception unit 102 User information acquisition unit 103 Related text acquisition unit 104 Final input output unit 105 Answer acquisition unit 106 Display control unit 20 User information management device 201 Input acquisition unit 202 User information registration unit 203 Authentication process execution unit 204 User information output unit 211 User information storage unit 30 Answer generation device 301 Input acquisition unit 302 Answer processing unit

Claims (8)

an input acquisition unit that acquires a user input sentence from a user;
a related text acquisition unit that acquires a first text related to the user input sentence from a publication published by a publisher;
an auxiliary input generating unit that generates a first question sentence by adding, to the user input sentence, the first text and an auxiliary sentence requesting an answer to the user input sentence based on the first text;
an answer output unit that inputs the first question sentence into a trained model to generate and output an answer;
An information processing device comprising: the first text cannot be obtained from the publication or the user does not have access to the publication;
the related text acquisition unit acquires a second text related to the user input sentence from the Internet;
the auxiliary input generating unit generates a second question sentence by adding, to the user input sentence, the second text and auxiliary sentence requesting an answer to the user input sentence based on the second text;
The answer output unit inputs the second question sentence into the trained model to generate and output an answer.
The information processing device according to claim 1 . When the first text cannot be obtained from the publication, or when the user does not have access to the publication and the second text cannot be obtained from the Internet, the answer output unit inputs the user input sentence into the trained model to generate and output an answer.
The information processing device according to claim 2 . If the first text is equal to or greater than a predetermined number of characters or bytes,
the auxiliary input generating unit generates a third question sentence by adding a summary sentence of the first text and an auxiliary sentence requesting an answer to the user input sentence based on the summary sentence to the user input sentence,
The answer output unit inputs the third question sentence into the trained model to generate and output an answer.
The information processing device according to claim 1 . The summary sentence is generated by inputting the first text and a sentence requesting the summary sentence into the trained model.
The information processing device according to claim 4. The related text acquisition unit acquires the first text from the publication according to the age or grade of the user.
The information processing device according to claim 1 . the auxiliary input generating unit generates the first question sentence by further adding an auxiliary sentence requesting an answer according to the age or school year of the user.
The information processing device according to claim 1 . Obtaining a user input sentence from a user;
obtaining text associated with the user input sentence from a publication published by a publisher;
generating a question sentence by adding the text and an auxiliary sentence to the user input sentence, the auxiliary sentence requesting an answer based on the text to the user input sentence;
inputting the question sentence into a trained model to generate and output an answer;
A program for causing an information processing device to execute a process including the steps of:

Images