Search support device, search support method, and recording medium

Abstract

This search support device is provided with an acquisition unit, a search query generation unit, a search query output unit, a search result acquisition unit, an evaluation unit, and a model update unit. The acquisition unit acquires a user query entered by a user. The search query generation unit generates a search query for a search engine to perform a search, using a generation model on the basis of the user query. The search query output unit outputs the search query to the search engine. The search result acquisition unit acquires a search result for the search query from the search engine. The evaluation unit evaluates the search result on the basis of criteria for evaluating the relevance of the search result to the user query. The model update unit updates the generation model on the basis of the result of the evaluation. The search support device can generate a search query using, for example, the generation model that has been optimized through update based on the result of the evaluation of the search result. Thus, the search support device can be used to obtain search results suitable for decision-making.

Description

Search support device, search support method, and recording medium

[0001] The present disclosure relates to a search support device and the like.

[0002] When searching for information using a search engine, a user of the search engine, for example, inputs a query indicating the information they wish to search for into the search engine. The user then obtains search results based on the query from the search engine. In addition, in order to properly search for the desired information, it may be necessary to generate a query that is appropriate for the search engine's specifications and the search target.

[0003] The information processing device of Patent Document 1 accepts input of a question in natural language and a correct answer that is an appropriate output for the question. The information processing device of Patent Document 1 then updates parameters of a converter that converts the natural language question into a query used for searching so that the correct answer matches an answer obtained by searching a search database using the query.

[0004] International Publication No. 2018 / 179355

[0005] With the technology described in Patent Document 1, it may be difficult to use a search engine to search for information that matches the user's desires.

[0006] In order to solve the above-mentioned problems, the present disclosure aims to provide a search device or the like that allows a user to easily search for desired information.

[0007] In order to solve the above problems, the search support device disclosed herein comprises an acquisition means for acquiring a user query, which is a query entered by a user; a search query generation means for generating a search query, which is a query for executing a search in a search engine, using a generative model based on the acquired user query; a search query output means for outputting the search query to the search engine; a search result acquisition means for acquiring search results for the search query from the search engine; an evaluation means for evaluating the search results based on criteria for evaluating the suitability of the search results to the user query; and a model update means for updating the generative model based on the results of the evaluation.

[0008] The search support method disclosed herein acquires a user query, which is a query entered by a user, and based on the acquired user query, uses a generative model to generate a search query, which is a query for executing a search in a search engine, outputs the search query to the search engine, acquires search results for the search query from the search engine, evaluates the search results based on criteria for evaluating the suitability of the search results to the user query, and updates the generative model based on the results of the evaluation.

[0009] The recording medium of the present disclosure non-temporarily records a search support program that causes a computer to execute the following processes: a process of acquiring a user query, which is a query entered by a user; a process of generating a search query, which is a query for executing a search in a search engine using a generative model based on the acquired user query; a process of outputting the search query to the search engine; a process of acquiring search results for the search query from the search engine; and a process of evaluating the search results based on criteria for evaluating the suitability of the search results for the user query and updating the generative model based on the results of the evaluation.

[0010] According to the present disclosure, users can easily search for desired information.

[0011] FIG. 1 is a diagram illustrating an example of the configuration of an information search system according to the present disclosure. FIG. 2 is a diagram illustrating an example of the configuration of a search support device according to the present disclosure. FIG. 3 is a diagram illustrating an example of a display screen according to the present disclosure. FIG. 4 is a diagram illustrating an example of a display screen according to the present disclosure. FIG. 5 is a diagram illustrating an example of an operation flow of a search support device according to the present disclosure. FIG. 6 is a diagram illustrating an example of the hardware configuration of a search support device according to the present disclosure.

[0012] An embodiment of the present disclosure will be described in detail with reference to the drawings. FIG. 1 is a diagram illustrating an example of the configuration of an information search system. The information search system includes a search support device 10, a terminal device 20, and a search device 30. The search support device 10 is connected to the terminal device 20 via a network, for example. The search support device 10 is also connected to the search device 30 via the network. There may be a plurality of terminal devices 20 and a plurality of search devices 30. The number of terminal devices 20 and the number of search devices 30 may be set as appropriate.

[0013] An information retrieval system is a system that searches for information, for example, via a search engine. The search target may be, for example, information that is publicly available on a website or the like on a network. The search target may also be information stored in a file or database stored in a storage device. The search target is not limited to the above.

[0014] In an information retrieval system, for example, a search query suitable for input to a search engine is generated from a user query. Then, for example, a search is performed in the search engine based on the search query. The user query is a query indicating a search target that is input by a user to perform a search. The user query is, for example, a keyword or sentence indicating the information the user wishes to search for. Furthermore, the user is, for example, a person who inputs the user query to perform a search.

[0015] A search query is, for example, a query in a search engine that is likely to find the information a user is looking for. A search query is, for example, a query that is suited to the search environment. For example, a query suited to a search may differ depending on the search environment. The search environment is, for example, the specifications of the search engine, the configuration of the storage device to be searched, and the arrangement of data within the storage device. Therefore, for example, by performing a search using a search query that is suited to the search environment, the likelihood of finding the information a user is looking for can be increased.

[0016] The search query is generated from the user query using, for example, a generative model. The generative model is, for example, a learning model generated by machine learning. The generative model is also updated based on the results of evaluating search results obtained by searching using the search query. Then, for example, the processes of updating the generative model based on the evaluation results, generating a search query using the updated generative model, outputting the search query to a search engine, and obtaining search results based on the search query are repeated until the conditions for outputting search results are met. This makes it possible to generate a search query that is suitable for the search in the environment in which the user is performing the search.

[0017] Here, a specific example of the configuration of the search support device 10 will be described. Fig. 2 is a diagram showing an example of the configuration of the search support device 10. The search support device 10 basically includes an acquisition unit 11, a search query generation unit 12, a search query output unit 13, a search result acquisition unit 14, an evaluation unit 15, and a model update unit 17. The search support device 10 also includes, for example, a determination unit 16, an output unit 18, and a storage unit 19.

[0018] The acquisition unit 11 acquires a user query. The user query is a query related to information that a user wishes to search. The acquisition unit 11 acquires the user query from, for example, the terminal device 20. The user query is input to, for example, the terminal device 20 by a user's operation. The acquisition unit 11 may acquire, as the user query, an input to the interactive artificial intelligence input by a user's operation. When the user query is an input from a user to a chatbot, the acquisition unit 11 acquires the user query from, for example, a management device for the chatbot.

[0019] For example, if a user wants to know why dinosaurs became extinct, the user outputs the sentence "Tell me why dinosaurs became extinct" as a user query to the terminal device 20. In this case, the acquisition unit 11 acquires, for example, from the terminal device 20, text indicating the sentence "Tell me why dinosaurs became extinct" as the user query.

[0020] The user query may be input to the terminal device 20, for example, using the user's voice. When input using the user's voice, for example, the terminal device 20 converts the user query input using voice into text. Then, the acquisition unit 11, for example, acquires the text of the user query converted by the terminal device 20 from the terminal device 20. The acquisition unit 11 may also acquire the user query as voice data. When acquiring the user query as voice data, for example, the user query in the form of voice data is converted into text by a conversion unit not shown. The format of the user query when input is not limited to the above.

[0021] The search query generation unit 12 generates a search query using a generative model based on the acquired user query. The search query is a query for executing a search in a search engine. The search query generation unit 12 uses the generative model to generate, from the user query, a search query that is likely to obtain search results that match the user query. The generative model is, for example, a learning model that takes the user query as input and outputs a search query. The search engine is, for example, a function provided in the search device 30. Furthermore, when the generative model is updated, the search query generation unit 12 generates a search query using the updated generative model based on the user query, for example.

[0022] The search query generation unit 12 generates a search query sentence from a user query sentence, for example, by using a generative model. The search query does not have to be in the form of a sentence. The search query may be in the form of a conditional expression, for example. The search query may also be in the form of an artificial language. The artificial language is, for example, program source code. When the search query is in the form of an artificial language, the search query generation unit 12 generates an artificial language indicating the search query from the user query sentence, for example, by using a generative model. The artificial language is not limited to the above. The form of the search query is not limited to the above.

[0023] The generative model is, for example, a language model. The language model is, for example, one that learns the relationships between words in a sentence. The language model is, for example, a model that generates related strings related to a target string from the target string. By using a language model that has been trained with sentences and sentences of various contexts, it is possible to generate related strings with content related to the target string. For example, a case will be described in which a language model is used in question answering. The language model receives an input question such as "What kind of country is Japan?" as the target string. The language model generates a string such as "Japan is an island country in the Northern Hemisphere..." as an answer to the question. The language model is generated, for example, by learning to output at least one sentence including the input string.

[0024] For example, a learning model that can output a sentence including an input character string by predicting a character string that is highly likely to follow the input character string can be used as the language model. For example, GPT-2 (Generative Pre-Training-2) or GPT-3 can be used as the language model. These language models are also called large-scale language models. Furthermore, T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), or ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately) may be used as the language model. The language model is not limited to the above. Furthermore, the generative model is not limited to the language model.

[0025] The generative model is created, for example, by machine learning. The generative model is created, for example, by deep learning using a neural network. The machine learning algorithm used to create the generative model is not limited to the above. The model update unit 17 updates the generative model by re-learning based on the results of evaluation of search results. When the generative model is updated, the search query generation unit 12 generates a search query using the updated generative model based, for example, on a user query.

[0026] The generative model may also be a rule-based model. In the case of a rule-based model, the generative model generates a search query from a user query using, for example, a dictionary in which weights are assigned to individual keywords in a table format. The generative model generates the search query by, for example, reading words corresponding to keywords representing the user query from the table based on the assigned weights.

[0027] The generative model may be selected from among multiple generative models. The generative model is generated, for example, according to factors that may affect search results. The factors that may affect search results include, for example, one or more of the search target field, search engine specifications, storage device configuration, data layout within the storage device, and the type of stored data. The factors that may affect search results are not limited to the above. Furthermore, the generative model may be created for each user or each user's attributes. For example, when users belong to the same department and perform the same work, the search engine and search scope used for the search are often the same. Therefore, for example, using a generative model created according to the user's attributes can improve search accuracy.

[0028] When one of the multiple generative models is selected and used, for example, the acquisition unit 11 acquires a selection result of a search query to be used to generate a search query from the terminal device 20. The selection result of the generative model is input to the terminal device 20 by, for example, a user operation.

[0029] The search query generation unit 12 generates a search query, for example, using a generative model selected by a user. The selection result of the generative model is acquired by the acquisition unit 11, for example, from the terminal device 20. The selection result of the generative model is input, for example, to the terminal device 20 by an operation by the user. The search query generation unit 12 may also generate a search query, for example, using a generative model set for each user. For example, the search query generation unit 12 identifies the user by referring to the login information of the terminal device 20. Then, the search query generation unit 12 selects a generative model to be used for generating the search query, for example, based on a table in which data indicating the relationship between the user and the generative model is recorded. The search query generation unit 12 may also select a generative model to be used for generating the search query, based on a table in which data indicating the relationship between the user's affiliation and the generative model is recorded.

[0030] The search query output unit 13 outputs a search query to a search engine. When a search query is generated using the updated generative model, the search query output unit 13 outputs the search query generated using the updated generative model to, for example, the search engine. The search query output unit 13 may output the search query to the search engine via a terminal device 20 or another information processing device. The other information processing device is, for example, a management device for an interactive artificial intelligence. Examples of the other information processing device are not limited to those described above.

[0031] The search result acquisition unit 14 acquires search results based on the search query from the search engine. The search results are, for example, the results of a search executed by the search engine using the search query as input.

[0032] When a search query generated using the updated generative model is output, the search result acquisition unit 14 acquires, for example, search results based on the search query generated using the updated generative model.

[0033] The evaluation unit 15 evaluates the search results based on criteria for evaluating the suitability of the search results to the user query. The suitability of the search results to the user query is, for example, the degree to which the information desired by the user is found. The evaluation unit 15 evaluates the search results by, for example, calculating an index value indicating the result of the evaluation of the search results based on the criteria for evaluating the suitability of the search results to the user query.

[0034] The evaluation unit 15 evaluates the search results using, for example, a criterion based on the similarity between the user query and the search results as a criterion for evaluating the suitability of the search results for the user query. The similarity is the Euclidean distance between the embedding vector of the user query and the embedding vector of the search results. The distance between the embedding vector of the user query and the embedding vector of the search results, which indicates the similarity, is not limited to the above. For example, the evaluation unit 15 evaluates that the higher the similarity between the user query and the search results, the higher the degree of suitability of the search results for the user query.

[0035] The evaluation unit 15 may evaluate the search results using the output from the language model when the search results are input to the language model. For example, in a question-answering system using a language model, when the search results are used as input to the system, the operation log of the question-answering system can be used for evaluation. In this case, for example, the criteria for evaluating the suitability of the search results to the user query are set so that the fewer the number of question answers recorded in the operation log, the higher the index indicating the evaluation result. For example, the fewer the number of question answers, the higher the possibility that an appropriate answer has been found. This is because the suitability of the search query to the user query is considered to be high. For example, Chat GPT can be used as the language model. The language model is not limited to Chat GPT.

[0036] The evaluation unit 15 may evaluate the search results using the results of a dialogue between the chatbot and the user using the output from the language model. When the results of the dialogue between the chatbot and the user are used, the number of dialogues required to search for information can be used for the evaluation. In this case, for example, the criteria for evaluating the suitability of the search results for the user query are set so that the fewer the number of dialogues required to search for information, the higher the index indicating the evaluation result.

[0037] In a dialogue between a chatbot and a user, when a search is performed, the chatbot presents search results to the user and repeatedly accepts new input from the user. In this case, for example, at the end of the dialogue, the evaluation unit 15 evaluates the search results for a certain turn using the number of dialogue turns that occurred after the turn. A turn refers, for example, to a single question between the user and the chatbot and a response to the question. The evaluation unit 15 evaluates the search results based on, for example, the value of a monotonically decreasing function. The monotonically decreasing function can be, for example, the inverse of the number of turns. In this case, the suitability of the search results to the user query is evaluated such that the fewer the number of turns after the turn to be evaluated, the higher the evaluation of the search results. If the number of turns after the turn to be evaluated is large, the search results are likely to differ from what the user desired. In other words, if the number of turns after the turn to be evaluated is large, the evaluation of the search results will be low. On the other hand, if the number of turns after the turn to be evaluated is small, the search results are likely to match the information the user desired, and therefore the evaluation of the search results will be high.

[0038] The evaluation unit 15 may evaluate the search results using the result of the user's evaluation of the search results as a criterion for evaluating the suitability of the search results for the user query. The result of the user's evaluation of the search results is, for example, acquired by the acquisition unit 11 from the terminal device 20. The result of the user's evaluation of the search results is, for example, input to the terminal device 20 by the user's operation. The result of the user's evaluation of the search results is acquired as an evaluation result indicated by, for example, two values, "Good" or "Bad." "Good" is selected, for example, when the search results are suitable for the content of information desired by the user. Furthermore, "Bad" is selected, for example, when the search results are not suitable for the content of information desired by the user.

[0039] The evaluation unit 15 may evaluate the search results using the output of the language model as a criterion for evaluating the suitability of the search results for the user query. The evaluation unit 15, for example, inputs the results of the evaluation performed by the evaluation unit 15 and the search results into the language model. The evaluation unit 15, for example, inputs the similarity between the sentences of the search results and the user query and the search results into the language model. Then, the evaluation unit 15 evaluates the search results based on, for example, the output from the language model. The language model, for example, references the results of the evaluation performed by the evaluation unit 15 and outputs statements regarding the suitability of the search results for the user query. The evaluation unit 15, for example, extracts information regarding suitability from the sentences output by the language model and evaluates the search results.

[0040] The determination unit 16 determines, for example, whether a condition for outputting search results is satisfied. The condition for outputting search results is, for example, a condition for determining the timing of presenting the search results to a user. When the condition for outputting search results is satisfied, the output unit 18 outputs the search results to, for example, the terminal device 20. When the condition for outputting search results is not satisfied, the search support device 10 repeats the following process again: updating the generative model in the model update unit 17; generating a search query using the updated generative model in the search query generation unit 12; outputting the search query to the search engine in the search query output unit 13; obtaining search results from the search engine in the search result acquisition unit 14; and evaluating the search results in the evaluation unit 15. In other words, when the condition for outputting search results is not satisfied, a search is performed using the search query generated by the updated generative model. Then, the search results are evaluated and it is determined again whether the condition for outputting search results is satisfied. Furthermore, one iteration of the process of updating the generative model, generating a search query using the updated generative model, outputting the search query to the search engine, obtaining search results from the search engine, and evaluating the search results is also referred to as a cycle.

[0041] The condition for outputting search results may be, for example, that at least one of the evaluation result of the search results and the number of times the repeated processing is performed satisfies the criteria for outputting search results. The criteria for outputting search results may be set so that the time until the search results are presented and the accuracy of the search results are appropriate for the user. The condition for outputting search results may be set based on only either the evaluation result of the search results or the number of times the repeated processing is performed. The condition for outputting search results may be that both the evaluation result of the search results and the number of times the repeated processing is performed satisfy the criteria for outputting search results. The criterion for outputting search results may be that a score based on the evaluation result of the search results for the user query and the number of search results satisfies the criteria for outputting search results.

[0042] The time until the search results are presented and the accuracy of the search results are appropriate for the user, for example, if the user can obtain useful search results within an acceptable time. For example, increasing the number of repetitions further optimizes the generative model, increasing the likelihood of obtaining search results appropriate for the user. On the other hand, increasing the number of repetitions may increase the time required to present search results to the user. Therefore, the output of search results is set to achieve a balance between the time until the search results are presented and the accuracy of the search results appropriate for the user, for example, depending on the purpose of the search and the search environment. For example, the output of search results is set when the number of repetitions is equal to or greater than a certain number. Furthermore, for example, a condition for outputting search results is when the evaluation results are equal to or greater than a certain number. Furthermore, for example, a condition for outputting search results is when the number of repetitions is equal to or greater than a certain number and the evaluation results of the search results are equal to or greater than a certain number. The criteria for the number of repetitions and the criteria for the evaluation results are set based on, for example, a balance between the time until the search results are presented and the accuracy of the search results. The condition for outputting search results is set, for example, by an administrator of the search support device 10.

[0043] The conditions for outputting search results may be input by the user. The conditions for outputting search results are acquired by the acquisition unit 11, for example, from the terminal device 20. For example, the user inputs criteria for outputting search results, which are indicated by the evaluation result of the search results or the number of repetitions, into the terminal device 20. The conditions for outputting the search results input by the user are acquired by the acquisition unit 11, for example, from the terminal device 20.

[0044] The model update unit 17 updates the generative model based on the evaluation results. The model update unit 17 updates the parameters of the generative model by, for example, performing machine learning using the evaluation results of the user query, the search query, and the search result as learning data. That is, the model update unit 17 updates the generative model by, for example, re-learning the generative model using the evaluation results of the user query, the search query, and the search result as learning data to update the parameters of the generative model. The model update unit 17 updates the parameters of the generative model by, for example, learning the relationship between the user query, the search query, and the evaluation results of the search result.

[0045] The model update unit 17 performs, for example, a model update process by relearning among the repeated processes of updating a generative model, generating a search query, searching, and evaluating. That is, the model update unit 17 performs a model update process by relearning among the repeated processes that are performed until the determination unit 16 determines that it is time to output search results.

[0046] The model update unit 17 stores, for example, the generative model last updated in the repeated processing in the storage unit 19. The stored generative model is used to generate a search query, for example, when a new user query is input.

[0047] The output unit 18 outputs, for example, the search results. The output unit 18 outputs the search results when, for example, the determination result of the determination unit 16 satisfies the condition for outputting the search results. The condition for outputting the search results is, for example, the condition described in the above description of the model update unit 17. The output unit 18 may also output the result of evaluation of the search results together with the search results.

[0048] The output unit 18 outputs the search results to, for example, the terminal device 20. The output unit 18 also outputs the search results to, for example, an information processing device that is a management device for an interactive artificial intelligence. The output unit 18 outputs the search results to, for example, an information processing device that is a management device for a chatbot.

[0049] When a user evaluates search results, the output unit 18 outputs, for example, a screen on which the user inputs the results of the evaluation of the search results. The output unit 18 outputs, for example, to the terminal device 20, a screen on which the user inputs the results of the evaluation of the search results. Furthermore, when a generative model to be used for generating a search query is selected by the user, the output unit 18 may output a screen on which the user selects the generative model to be used for generating the search query from a plurality of generative models. The output unit 18 outputs, for example, a screen on which the user selects the generative model to be used for generating the search query, to the terminal device 20.

[0050] The storage unit 19 stores, for example, data related to searches performed based on user queries. The storage unit 19 stores, for example, user queries and search queries in association with each other. The storage unit 19 also stores, for example, search results. The storage unit 19 stores, for example, user queries, search queries, and search results in association with each other. The storage unit 19 may also store search results in association with the search queries. The storage unit 19 stores, for example, results of evaluation of search results. The storage unit 19 stores, for example, user queries, search queries, and results of evaluation of search results in association with each other. The storage unit 19 also stores, for example, generative models. The generative models may be stored in a storage means external to the search support device 10.

[0051] Fig. 3 is a diagram showing an example of a display screen of search results. In the example of the display screen of Fig. 3, the sentence "Tell me why dinosaurs became extinct" is entered in the input field for the user query. In addition, in the example of the display screen of Fig. 3, a "Search" button is displayed to the right of the input field for the user query. In addition, in the example of the display screen of Fig. 3, the search results and the evaluation results for each search result are displayed.

[0052] In the example display screen of FIG. 3 , when a user inputs a user query into the user query input field on the screen and presses the “Search” button, the terminal device 20 outputs the user query and information for executing a search to, for example, the acquisition unit 11 of the search support device 10. After acquiring the user query and information for executing a search, the search support device 10 repeats the following processes until a condition for outputting search results is met: updating the generative model in the model update unit 17, generating a search query using the updated generative model in the search query generation unit 12, outputting the search query to the search engine in the search query output unit 13, acquiring search results from the search engine in the search result acquisition unit 14, and evaluating the search results in the evaluation unit 15. Then, when the condition for outputting search results is met, the output unit 18 of the search support device 10 outputs the search results to, for example, the terminal device 20. Then, the terminal device 20 displays the screen shown in the example display screen of FIG. 3 on, for example, a display device (not shown). In the example display screen of Fig. 3, a sentence beginning with "Why did dinosaurs become extinct?" and a sentence beginning with "Tell me! The mystery of dinosaurs" are displayed as search results. Also, in the example display screen of Fig. 3, the similarity to the user query is displayed as a search result evaluation result for each of the sentence beginning with "Why did dinosaurs become extinct?" and the sentence beginning with "Tell me! The mystery of dinosaurs."

[0053] Fig. 4 is a diagram showing an example of a display screen on which the number of repetitions of processing can be input when a search is performed. Similar to the display screen of Fig. 3, the example of the display screen of Fig. 4 displays an input field for a user query, a "Search" button for executing a search, and search results. The example of the display screen of Fig. 4 also displays an input field for the number of repetitions of processing. Also, similar to the display screen of Fig. 3, the example of the display screen of Fig. 4 displays search results and evaluation results for each search result.

[0054] In the example of the display screen of FIG. 4 , assume that a user inputs a user query into the user query input field on the screen, inputs a repetition count into the repetition count input field, and then presses the “Search” button. In this case, the terminal device 20 outputs, for example, the user query, the repetition count, and information for executing a search to the acquisition unit 11 of the search support device 10. After acquiring the user query, the repetition count, and information for executing a search, the search support device 10 repeatedly generates a search query, executes a search based on the search query, evaluates the search results, and updates the generative model until a condition for outputting search results is met. In the example of the display screen of FIG. 4 , because the repetition count is set to three as a condition for outputting search results, the search support device 10 repeats the steps of generating a search query, executing a search based on the search query, evaluating the search results, and updating the generative model three times. After the repetition process is performed three times, the determination unit 16 determines, for example, that the condition for outputting search results is met and therefore it is time to output the search results. Furthermore, if the result of the evaluation of the search results meets the condition for outputting search results, the determination unit 16 may determine that the condition for outputting search results is met when the number of repetitions is less than three. When it is determined that the conditions for outputting the search results are met, the output unit 18 of the search support device 10 outputs the search results to, for example, the terminal device 20. Then, the terminal device 20 displays, for example, the screen shown in the example display screen of Fig. 4 on a display device (not shown).

[0055] FIG. 5 is a diagram showing an example of a display screen on which a user inputs evaluation results for search results. Similar to the display screen of FIG. 3 , the example of the display screen of FIG. 5 displays an input field for a user query, a "Search" button for executing a search, and search results. The example of the display screen of FIG. 5 also displays radio buttons for selecting "Good" or "Bad" for each search result. The example of the display screen of FIG. 5 also displays a "Send Evaluation Results" button. The "Send Evaluation Results" button in the example of the display screen of FIG. 5 is a button that outputs the results of the user's evaluation to the acquisition unit 11 of the search support device 10 when the user presses "Send Evaluation Results" after selecting the radio button by the user's operation.

[0056] The output unit 18 outputs a screen shown in the example display screen of FIG. 5 to the terminal device 20, for example. Then, the terminal device 20 displays the screen shown in the example display screen of FIG. 5 on a display device (not shown), for example. The user, for example, selects "Good" or "Bad" for each search result on the screen. Then, the screen shown in the example display screen of FIG. 5 is displayed. The user, for example, presses a "Send evaluation result" button on the screen. When the "Send evaluation result" button is pressed, the terminal device 20 outputs the evaluation results of the search results to, for example, the acquisition unit 11 of the search support device 10. When the evaluation results of the search results are acquired, the determination unit 16 determines, for example, whether the conditions for outputting the search results are met. If the conditions for outputting search results are not met, the search support device 10 repeats the following processes again until the conditions for outputting search results are met: updating the generative model in the model update unit 17; generating a search query using the updated generative model in the search query generation unit 12; outputting the search query to the search engine in the search query output unit 13; acquiring search results from the search engine in the search result acquisition unit 14; and evaluating the search results in the evaluation unit 15. In this case, the model update unit 17 updates the generative model by re-learning the generative model using the user query and search results as input data and the results of the user's evaluation of the search results as training data. If the conditions for outputting search results are met, the output unit 18 of the search support device 10 outputs the search results to, for example, the terminal device 20. Then, the terminal device 20 displays, for example, the screen shown in the example display screen of FIG. 5 on a display device (not shown).

[0057] The terminal device 20 is, for example, an information processing device used by a user to search for information. The terminal device 20 accepts, for example, an input of a user query, which is a query related to information that the user wishes to search for. The user query is, for example, input to the terminal device 20 by an operation of the user. The terminal device 20 outputs the accepted input user query to, for example, the acquisition unit 11 of the search support device 10.

[0058] The terminal device 20 acquires search results for the user query, for example, from the output unit 18 of the search support device 10. The search results for the user query are the results of a search executed by a search engine included in the search device 30 based on a search query generated from the user query.

[0059] When the search results are evaluated by a user, the terminal device 20 accepts the evaluation result of the search results input by, for example, the user's operation. The terminal device 20 outputs the input evaluation result to, for example, the acquisition unit 11 of the search support device 10.

[0060] When a user selects a generative model, the terminal device 20 receives the generative model selection result input by the user's operation, and outputs the generative model selection result to the acquisition unit 11 of the search support device 10, for example.

[0061] The search device 30 executes a search based on, for example, a search query. The search device 30 executes a search based on, for example, the search query using a search engine.

[0062] The search engine may search for information from a database based on a search query. The database may be stored in a storage means included in the search device 30. The database may be stored in a storage device connected to the search device 30 via a network. The search engine may search for information based on the search query from information published on a network.

[0063] The search device 30 acquires a search query from, for example, the search query output unit 13 of the search support device 10. The search device 30 outputs search results to, for example, the search result acquisition unit 14 of the search support device 10. The search device 30 may acquire a search query from the terminal device 20. The search device 30 may output search results to the terminal device 20. The search device 30 may also be a device that acquires a query and outputs search results using interactive artificial intelligence. The search device 30 may also be a device that acquires a query and outputs search results using a chatbot.

[0064] A description will be given of a process of performing a search based on a search query generated from a user query in the search support device 10. Fig. 6 is a diagram showing an example of an operational flow of the process of performing a search based on a search query generated from a user query.

[0065] The acquisition unit 11 acquires a user query, which is a query input by a user (step S11). The acquisition unit 11 acquires the user query from the terminal device 20, for example.

[0066] When the user query is acquired, the search query generation unit 12 generates a search query, which is a query for executing a search in a search engine, based on the acquired user query and using a generation model (step S12).

[0067] When the search query is generated, the search query output unit 13 outputs the search query to the search engine (step S13). The search query output unit 13 outputs the search query to the search engine included in the search device 30, for example.

[0068] The search device 30, for example, acquires a search query from the search query output unit 13. Then, the search device 30, for example, uses the search query as input to a search engine and executes a search for information corresponding to the search query. After executing the search, the search device 30 outputs search results based on the search query to, for example, the search result acquisition unit 14 of the search support device 10.

[0069] The search result acquisition unit 14 of the search support device 10 acquires search results based on the search query from the search engine (step S14). The search result acquisition unit 14 acquires search results based on the search query from the search engine included in the search device 30, for example.

[0070] Once the search results are obtained, the evaluation unit 15 evaluates the search results based on criteria for evaluating the suitability of the search results for the user query (step S15).

[0071] When the search results are evaluated, the determination unit 16 determines, for example, whether the conditions for outputting the search results are met. If the conditions for outputting the search results are not met (No in step S16), the model update unit 17 updates the generative model based on the results of the evaluation of the search results (step S17). The model update unit 17 updates the generative model by re-learning the generative model using the user query, the search results as input data, and the results of the evaluation of the search results as training data. After updating the generative model, the process returns to step S12, and the search query generation unit 12 generates a search query using the updated generative model based on the user query.

[0072] In step S16, if the condition for outputting the search results is met (Yes in step S16), the output unit 18 outputs the search results, for example, to the terminal device 20 (step S18).

[0073] When the search results are output, the model update unit 17 updates the generative model based on the evaluation results of the search results (step S19). The model update unit 17 updates the generative model by re-learning the generative model using the user query, the search results, and the evaluation results of the search results as input data and training data. After updating the generative model, the model update unit 17 stores the updated generative model in, for example, the storage unit 19 (step S20). The stored generative model is used, for example, when the acquisition unit 11 starts a search process based on a newly acquired user query.

[0074] A specific example of a process in which the search support device 10 performs a search based on a search query generated from a user query will be described. An example will be described in which a user searches for information about dinosaurs. For example, the user inputs a sentence such as "Tell me why dinosaurs became extinct" as a user query into the terminal device 20. When the user query is input, the terminal device 20 outputs the input user query to the search support device 10, for example.

[0075] The acquisition unit 11 of the search support device 10 acquires the sentence "Tell me why dinosaurs became extinct" from the terminal device 20 as a user query. When the user query is acquired, the search query generation unit 12 generates three sentences, search query A, search query B, and search query C, from the sentence "Tell me why dinosaurs became extinct." Here, it is assumed that search query A is, for example, "dinosaur, extinction, meteorite." Search query B is, for example, "dinosaur, extinction, volcano." Furthermore, it is assumed that search query C is, for example, "dinosaur, extinction, reason." When the three search queries are generated, the search query output unit 13 outputs the three search queries to a search engine included in the search device 30.

[0076] The search device 30 acquires, for example, three search queries from the search support device 10. When the three search queries are acquired, the search engine of the search device 30 executes a search using, for example, each of the search queries as input. The search engine then outputs multiple search results for each of the search queries. The search result acquisition unit 14 of the search support device 10 acquires the search results from, for example, the search engine. Here, it is assumed that the search results for search query A are multiple search results including search results A1 and A2. It is assumed that search result A1 is a sentence that begins with a sentence such as "A meteorite destroyed the dinosaur's biological system." It is assumed that the search results for search query B are multiple search results including search results B1 and B2. It is assumed that search result B1 is a sentence that begins with the sentence "Mount Fuji was an active volcano." It is also assumed that the search results for search query C are multiple search results including search results C1 and C2. It is also assumed that search result C1 is a sentence that begins with the sentence "Tell me! The mystery of dinosaur extinction."

[0077] When the search result acquisition unit 14 acquires the search results, the evaluation unit 15 calculates, for example, a score indicating the similarity of each search result with the user query. For example, assume that search result A1 has a score of 0.4 and search result A2 has a score of 0.2. Also, assume that search result B1 has a score of 0.2 and search result A2 has a score of 0.1. Also, assume that search result C1 has a score of 0.8 and search result C2 has a score of 0.5. The score indicating the similarity is, for example, an index expressed as a numerical value between 0 and 1. Also, the score indicating the similarity is, for example, an index whose numerical value increases as the similarity increases.

[0078] When the search results are evaluated, the determination unit 16 determines, for example, whether the condition for outputting the search results is met. For example, assume that the condition for outputting the search results is set to two repetitions. In this case, because the number of repetitions is one, the determination unit 16 determines that the condition for outputting the search results is not met. If it is determined that the condition for outputting the search results is not met, the model update unit 17 re-trains the generative model based on the evaluation results. The model update unit 17 updates the generative model by performing machine learning using the user query and the search results as input data and the results of the evaluation of the search results as training data. Because the score of the search result C1 is high at 0.8, the updated generative model is more likely to output, for example, the search result C1 as a search query.

[0079] When the generative model is updated, the search query generation unit 12 generates a search query using the updated generative model based on, for example, a user query. For example, the search query generation unit 12 generates three sentences, search query C, search query D, and search query E, from the sentence "Tell me why dinosaurs became extinct." Here, it is assumed that search query C is, for example, "dinosaur, extinction, reason." It is assumed that search query C is, for example, "dinosaur, extinction, why." It is also assumed that search query E is, for example, "dinosaur, extinction theory." When the three search queries are generated, the search query output unit 13 outputs the three search queries to a search engine included in the search device 30.

[0080] The search device 30 acquires, for example, three search queries from the search support device 10. Upon acquiring the three search queries, the search engine of the search device 30 executes a search using, for example, each of the search queries as input. The search engine then outputs multiple search results for each of the search queries. Here, the search results for search query C are assumed to be, for example, multiple search results including search results C1 and C2. The search result C1 is assumed to be, for example, a sentence beginning with the sentence, "Tell me! The mystery of the extinction of dinosaurs." The search results for search query D are assumed to be, for example, multiple search results including search results D1 and D2. The search result D1 is assumed to be, for example, a sentence beginning with the sentence, "Why did dinosaurs become extinct?" Furthermore, the search results for search query E are assumed to be, for example, multiple search results including search results E1 and E2. Furthermore, the search result E1 is assumed to be a sentence beginning with the sentence, "There are various theories about the extinction of dinosaurs, and no definitive theory exists."

[0081] When the search engine outputs the search results, the search device 30 outputs the search results to, for example, the search result acquisition unit 14 of the search support device 10. Then, the search result acquisition unit 14 of the search support device 10 acquires the search results from, for example, the search device 30.

[0082] When the search result acquisition unit 14 acquires the search results, the evaluation unit 15 calculates, for example, a score for each search result indicating the degree of similarity with the user query. For example, assume that search result C1 has a score of 0.8 and search result C2 has a score of 0.5. Also, assume that search result D1 has a score of 0.9 and search result D2 has a score of 0.8. Also, assume that search result E1 has a score of 0.6 and search result E2 has a score of 0.2.

[0083] When the search results are evaluated, the determination unit 16 determines, for example, whether the conditions for outputting the search results are met. In this case, since the number of repetitions is two, the determination unit 16 determines that the conditions for outputting the search results are met. If it is determined that the conditions for outputting the search results are met, the output unit 18 outputs the search results to, for example, the terminal device 20. The output unit 18 outputs, for example, the three search results with the highest scores to the terminal device 20. The output unit 18 outputs, for example, three search results, search results D1, C1, and D2 to the terminal device 20. The number of search results to be output may be other than three.

[0084] The terminal device 20 acquires search results from, for example, the output unit 18 of the search support device 10. The terminal device 20 acquires, for example, three search results D1, C1, and D2. After acquiring the search results, the terminal device 20 displays the search results in descending order of score on a display device (not shown), for example.

[0085] When the search results are output, the model update unit 17 of the search support device 10 re-learns the generative model based on the results of the evaluation of the search results. The model update unit 17 updates the generative model by performing machine learning using the user query and the search results as input data and the results of the evaluation of the search results as training data. After updating the generative model, the model update unit 17 stores the updated generative model in, for example, the storage unit 19. When a new user query is acquired, a search query is generated using, for example, the generative model stored in the storage unit 19.

[0086] As another example, a case where a user searches for a procedure within a company will be described. Assume that the user operates, for example, the terminal device 20 to access a search device 30 that performs a document search within the company. For example, the user inputs a user query such as "Please tell me about the procedure for a domestic business trip" into an input field for the information to be searched on the display screen. When the user query is input, the terminal device 20 outputs the input user query to, for example, the search device 30.

[0087] The search device 30 acquires a user query from, for example, the terminal device 20. Upon acquiring the user query, the search device 30 outputs the acquired user query to, for example, the acquisition unit 11 of the search support device 10.

[0088] The acquisition unit 11 of the search support device 10 acquires the sentence "Please tell me the procedures for domestic business trips" as a user query from the search device 30. When the user query is acquired, the search query generation unit 12 generates, for example, the sentence "Domestic business trip procedures" from the sentence "Please tell me the procedures for domestic business trips" as a search query A1. When the search query is generated, the search query output unit 13 outputs the search query to a search engine included in the search device 30.

[0089] The search device 30 acquires a search query from, for example, the search support device 10. Upon acquiring the search query, the search engine of the search device 30 executes a search using, for example, the search query as input. The search engine then outputs search results for each search query. Here, it is assumed that the search result for the search query A1 is only search result A1 "Domestic business trip procedures are summarized here [URL]".

[0090] When the search engine outputs the search results, the search device 30 outputs the search results to, for example, the search result acquisition unit 14 of the search support device 10. Then, the search result acquisition unit 14 of the search support device 10 acquires the search results from, for example, the search device 30.

[0091] When the search result acquisition unit 14 acquires the search results, the evaluation unit 15 calculates, for example, a score indicating the similarity of the search results with the user query. Then, the evaluation unit 15 calculates a score based on the similarity and an overall score based on the number of search results. For example, suppose that the score of search result A1 based on the similarity to the user query is 0.6. Also, suppose that the overall score based on the number of search results being 1 is 0.2. Also, suppose that the condition for outputting the search results is set to an overall score of 0.5 or more.

[0092] When the search results are evaluated, the determination unit 16 determines, for example, whether the conditions for outputting the search results are met. In this case, since the overall score does not meet the conditions for outputting the search results, the determination unit 16 determines that the conditions for outputting the search results are not met. If it is determined that the conditions for outputting the search results are not met, the model update unit 17 re-learns the generative model. The model update unit 17 updates the generative model by, for example, performing machine learning using the user query and search result A1 as input data and the results of the evaluation of the search results as training data. The generative model updated by re-learning is more likely to output a search query other than "domestic business trip procedures" in response to the user query "Please tell me the procedures for domestic business trips."

[0093] When the generative model is updated, the search query generation unit 12 generates a search query using the updated generative model based on, for example, a user query. The search query generation unit 12 generates search query B "How to process a domestic business trip" from, for example, a user query sentence such as "Please tell me how to process a domestic business trip." When the search query is generated, the search query output unit 13 outputs the search query to, for example, the search device 30.

[0094] The search device 30 acquires a search query from, for example, the search support device 10. Upon acquiring the search query, the search engine of the search device 30 executes a search using, for example, the search query as input. The search engine then outputs search results for the search query. Here, the search results for search query B are, for example, search result B1 "Domestic and overseas business trip procedure manual" and search result B2 "Domestic business trip procedures are complicated, so please apply early."

[0095] When the search engine outputs the search results, the search device 30 outputs the search results to, for example, the search result acquisition unit 14 of the search support device 10. Then, the search result acquisition unit 14 of the search support device 10 acquires the search results from, for example, the search device 30.

[0096] When the search result acquisition unit 14 acquires the search results, the evaluation unit 15 calculates, for example, a score indicating the similarity of the search results with the user query. Then, the evaluation unit 15 calculates an overall score based on the score based on the similarity and the number of search results. For example, assume that the score of search result B1 based on the similarity to the user query is 0.8. Also, assume that the score of search result B2 based on the similarity to the user query is 0.3. Furthermore, the evaluation unit 15 calculates an overall score based on, for example, the first search result and the second search result and the total number of search results from the first and second searches. The calculated overall score is, for example, 0.6.

[0097] When the search results are evaluated, the determination unit 16 determines, for example, whether the conditions for outputting the search results are met. In this case, since the total score is 0.6, the determination unit 16 determines that the conditions for outputting the search results are met. If it is determined that the conditions for outputting the search results are met, the output unit 18 outputs the search results to, for example, the search device 30. The output unit 18 outputs, for example, the search result with the highest score based on similarity to the search device 30. The output unit 18 outputs, for example, the search result B1 with the highest score based on similarity. The output unit 18 may output multiple search results to the search device 30, for example, in order of the scores based on similarity.

[0098] The search device 30 acquires the search results from, for example, the output unit 18 of the search support device 10. Upon acquiring the search results, the search device 30 outputs the search results to, for example, the terminal device 20. Then, the terminal device 20 acquires the search results from, for example, the search device 30. Upon acquiring the search results, the terminal device 20 displays the search results on, for example, a display device (not shown).

[0099] When the search results are output, the model update unit 17 of the search support device 10 re-learns the generative model based on the results of the evaluation of the search results. The model update unit 17 updates the generative model by performing machine learning using the user query and the search results as input data and the results of the evaluation of the search results as training data. After updating the generative model, the model update unit 17 stores the updated generative model in, for example, the storage unit 19. When a new user query is acquired, a search query is generated using, for example, the generative model stored in the storage unit 19.

[0100] The search support device 10 may be used to search for information in the medical and healthcare field. The medical and healthcare field is evolving daily, and information in this field is constantly being updated. Generating an appropriate query can be important to obtain necessary information from specialized databases.

[0101] For example, by combining the search support device 10 with a search device 30, which is a question-answering device using a language model, an information search system for medical use that searches for specialized papers is constructed. A user inputs a user query, for example, "latest results on diabetes treatment methods." The search support device 10 generates a search query from the user query, for example. The search support device 10 then outputs the search query to the search device 30, for example.

[0102] The search device 30 searches for, for example, papers on diabetes treatment methods based on the acquired search query. The question-answering system generates, for example, interactive answers to the user query. The answers from the user are used as evaluation results and are used to retrain the generative model. Furthermore, by performing a search using the search query generated by the retraining, it is possible to perform a search using a generative model optimized for searching for papers on diabetes treatment methods.

[0103] Additionally, a "redo" button may be provided on the answer display screen. The user can repeat the cycle of inputting a user query and obtaining an answer. This allows the user to make adjustments to get closer to the information they are looking for. Such an information retrieval system can assist users in making accurate decisions in the medical and healthcare field by aiding in information searches in the field.

[0104] The search support device 10 uses a generative model to generate a search query, which is a query for executing a search in a search engine, based on a user query, which is a query related to information the user wishes to search. The search support device 10 also evaluates search results based on the search query based on criteria for evaluating the suitability of the search results for the user query. The search support device 10 then updates the generative model based on the evaluation results. By updating the generative model in this way, the generative model can automatically generate a search query, based on the user query, that can more appropriately search for the information the user wishes to search in a search engine. Therefore, by using the search support device 10, the user can easily search for the information they desire.

[0105] The search support device 10 repeatedly generates a search query using a generative model updated based on the results of the evaluation of the search results, outputs the search query to the search engine, obtains the search results, and evaluates the search results, for example, until a condition for outputting the search results is met. Then, the search support device 10 outputs the search results, for example, when it determines that the condition for outputting the search results is met. By performing the repeated processing, the search support device 10 can, for example, generate a search query that is suitable for the user query and the search engine. Generating a search query that is suitable for the user query and the search engine makes it possible, for example, to search for information that is more suited to the user's desired information. In other words, generating a search query that is suitable for the user query and the search engine can improve the accuracy of the search. Therefore, using the search support device 10 can improve the accuracy of the search for information desired by the user.

[0106] The processes in the search support device 10 may be distributed and executed among multiple information processing devices connected via a network. For example, the processes related to generating a search query, the processes related to evaluating search results, and the processes related to updating a generative model may be performed in different information processing devices. For example, the processes in the search query generation unit 12, the search query output unit 13, the search result acquisition unit 14, and the evaluation unit 15 may be performed in different information processing devices, and the processes in the model update unit 17 may be performed in different information processing devices. It may be set as appropriate which information processing device performs each process in the search support device 10.

[0107] Each process in the search support device 10 can be realized by executing a computer program on a computer. Fig. 7 shows an example of the configuration of a computer 100 that executes a computer program that performs each process in the search support device 10. The computer 100 includes a processor 101, a memory 102, a storage device 103, an input / output I / F (Interface) 104, and a communication I / F 105.

[0108] The processor 101 reads and executes computer programs for performing each process from the storage device 103. The processor 101 may be configured as a combination of multiple processors. The processor 101 may be, for example, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a micro-processing unit (MPU), a floating-point unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, or a microcontroller. The processor 101 may be a combination of multiple types of processors. The memory 102 is configured with a dynamic random access memory (DRAM) or the like, and temporarily stores computer programs executed by the processor 101 and data being processed. The storage device 103 stores the computer programs executed by the processor 101. The storage device 103 is configured with, for example, a non-volatile semiconductor storage device. The storage device 103 may also be another storage device, such as a hard disk drive. The input / output I / F 104 is an interface that receives input from an operator and outputs display data, etc. The communication I / F 105 is an interface that transmits and receives data between the terminal device 20, the search device 30, and other information processing devices. The terminal device 20 and the search device 30 may have the same configuration as the computer 100.

[0109] The computer program used to execute each process can also be stored and distributed on a computer-readable recording medium that non-temporarily stores data. Examples of recording media that can be used include magnetic tapes for recording data and magnetic disks such as hard disks. Optical disks such as CD-ROMs (Compact Disc Read Only Memory) can also be used as recording media. Non-volatile semiconductor storage devices can also be used as recording media.

[0110] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0111] [Supplementary Note 1] A search support device comprising: an acquisition means for acquiring a user query that is a query input by a user; a search query generation means for generating a search query that is a query for executing a search in a search engine using a generative model based on the acquired user query; a search query output means for outputting the search query to the search engine; a search result acquisition means for acquiring search results for the search query from the search engine; an evaluation means for evaluating the search results based on criteria for evaluating the suitability of the search results for the user query; and a model update means for updating the generative model based on the results of the evaluation.

[0112] [Supplementary Note 2] The search support device according to Supplementary Note 1, further comprising a determination means for determining whether a condition for outputting the search results is met, and if the condition for outputting the search results is not met, updating the generative model by the model update means, generating the search query using the updated generative model by the search query generation means, outputting the search query by the search query output means, obtaining the search results by the search result acquisition means, and evaluating the search results by the evaluation means are repeated.

[0113] [Supplementary Note 3] The search support device according to Supplementary Note 2, further comprising an output unit that outputs the search results when a condition for outputting the search results is satisfied as a result of the determination.

[0114] [Supplementary Note 4] The search support device according to Supplementary Note 2 or 3, wherein the model update means stores the last updated generative model in the repeated processing as the generative model whose update has been completed.

[0115] [Supplementary Note 5] The search support device according to any one of Supplementary Notes 2 to 4, wherein a condition for outputting the search results is that at least one of the result of the evaluation and the number of times the repeated processing has been performed satisfies a criterion for outputting the search results.

[0116] [Supplementary Note 6] The search support device according to any one of Supplementary Notes 2 to 4, wherein a condition for outputting the search results is that a score based on an evaluation result of the search results for the user query and the number of search results satisfies a criterion for outputting the search results.

[0117] [Supplementary Note 7] The search support device according to any one of Supplementary Notes 1 to 6, wherein the evaluation means evaluates the search results using criteria based on the similarity of the search results to the user query as criteria for evaluating the suitability of the search results to the user query.

[0118] [Supplementary Note 8] The search support device according to Supplementary Note 7, wherein the similarity is a distance between an embedding vector of the user query and an embedding vector of the search result.

[0119] [Supplementary Note 9] The search support device according to any one of Supplementary Notes 1 to 6, wherein the evaluation means evaluates the search results using an output from a language model when the search results are input to the language model.

[0120] [Supplementary Note 10] The search support device according to Supplementary Note 9, wherein the evaluation means evaluates the search results using a result of a dialogue with the user using an output from the language model.

[0121] [Supplementary Note 11] The search support device according to any one of Supplementary Notes 1 to 6, wherein the evaluation means evaluates the search results based on an evaluation of the search results input by the user.

[0122] [Supplementary Note 12] The search support device according to any one of Supplementary Notes 1 to 11, wherein the model update means updates parameters of the generative model by performing machine learning using the user query, the search query, and the evaluation results as training data.

[0123] [Supplementary Note 13] A search support method comprising: acquiring a user query that is a query input by a user; generating a search query that is a query for executing a search in a search engine using a generative model based on the acquired user query; outputting the search query to the search engine; acquiring search results for the search query from the search engine; evaluating the search results based on criteria for evaluating the suitability of the search results for the user query; and updating the generative model based on the results of the evaluation.

[0124] [Supplementary Note 14] A search support method according to Supplementary Note 13, comprising determining whether a condition for outputting the search results is met, and if the result of the determination indicates that the condition for outputting the search results is not met, updating the generative model, generating the search query using the updated generative model, outputting the search query, obtaining the search results, and evaluating the search results as a repeated process.

[0125] [Supplementary Note 15] The search support method according to Supplementary Note 14, wherein the search results are output when a condition for outputting the search results is satisfied as a result of the determination.

[0126] [Supplementary Note 16] The search support method according to Supplementary Note 14 or 15, wherein in the repeated processing, the last updated generative model is saved as the generative model whose update has been completed.

[0127] [Supplementary Note 17] The search support method according to any one of Supplementary Notes 14 to 16, wherein the condition for outputting the search results is that at least one of the result of the evaluation and the number of times the repeated processing has been performed satisfies a criterion for outputting the search results.

[0128] [Supplementary Note 18] The search support method described in any one of Supplementary Notes 14 to 16, wherein the condition for outputting the search results is that a score based on the result of evaluation of the search results for the user query and the number of search results satisfies a criterion for outputting the search results.

[0129] [Supplementary Note 19] The search support method according to any one of Supplementary Notes 13 to 18, wherein the search results are evaluated using a criterion based on the similarity of the search results to the user query as a criterion for evaluating the suitability of the search results to the user query.

[0130] [Supplementary Note 20] The search support method according to Supplementary Note 19, wherein the similarity is a distance between an embedding vector of the user query and an embedding vector of the search result.

[0131] [Supplementary Note 21] The search support method according to any one of Supplementary Notes 13 to 18, further comprising: evaluating the search results using an output from a language model when the search results are input to the language model.

[0132] [Supplementary Note 22] The search support method according to Supplementary Note 21, further comprising: evaluating the search results using a result of a dialogue with the user using an output from the language model.

[0133] [Supplementary Note 23] The search support method according to any one of Supplementary Notes 13 to 18, wherein the search results are evaluated based on an evaluation of the search results input by the user.

[0134] [Supplementary Note 24] The search support method according to any one of Supplementary Notes 13 to 23, wherein parameters of the generative model are updated by performing machine learning using the user query, the search query, and the evaluation results as training data.

[0135] [Supplementary Note 25] A recording medium that non-temporarily records a search support program that causes a computer to execute the following processes: a process of acquiring a user query, which is a query input by a user; a process of generating a search query, which is a query for executing a search in a search engine, using a generative model based on the acquired user query; a process of outputting the search query to the search engine; a process of acquiring search results for the search query from the search engine; a process of evaluating the search results based on criteria for evaluating the suitability of the search results for the user query; and a process of updating the generative model based on the results of the evaluation.

[0136] [Supplementary Note 26] A recording medium according to Supplementary Note 25, which non-temporarily records a search support program that causes a computer to execute the following processes: a process of determining whether the conditions for outputting the search results are met; and, if the results of the determination indicate that the conditions for outputting the search results are not met, a process of updating the generative model, generating the search query using the updated generative model, outputting the search query, obtaining the search results, and evaluating the search results again as a repetitive process.

[0137] [Supplementary Note 27] The recording medium according to Supplementary Note 26, which non-temporarily records a search support program, wherein the search results are output when a condition for outputting the search results is satisfied as a result of the determination.

[0138] [Supplementary Note 28] A recording medium according to Supplementary Note 27, which non-temporarily records a search support program that causes a computer to execute a process of saving the last updated generative model as the generative model whose update has been completed in the repeated process.

[0139] [Supplementary Note 29] The recording medium according to any one of Supplementary Notes 26 to 28, which non-temporarily records a search support program, wherein the condition for outputting the search results is that at least one of the result of the evaluation and the number of times the repeated processing has been performed satisfies a criterion for outputting the search results.

[0140] [Appendix 30] A recording medium described in any of Appendices 26 to 28, which non-temporarily records a search support program, wherein the condition for outputting the search results is that a score based on the result of evaluation of the search results for the user query and the number of search results satisfies a criterion for outputting the search results.

[0141] [Appendix 31] A recording medium described in any of Appendices 25 to 30, which non-temporarily records a search support program that causes a computer to execute a process of evaluating the search results using criteria based on the similarity of the search results to the user query as criteria for evaluating the suitability of the search results to the user query.

[0142] [Supplementary Note 32] The recording medium according to Supplementary Note 31, which non-temporarily records a search support program, wherein the similarity is a distance between an embedding vector of the user query and an embedding vector of the search result.

[0143] [Appendix 33] A recording medium according to any one of Appendices 25 to 30, which non-temporarily records a search support program that causes a computer to execute a process of evaluating the search results using an output from the language model when the search results are input into the language model.

[0144] [Supplementary Note 34] The recording medium according to Supplementary Note 33, which non-temporarily records a search support program that causes a computer to execute a process of evaluating the search results using a result of a dialogue with the user using an output from the language model.

[0145] [Supplementary Note 35] A recording medium according to any one of Supplementary Notes 25 to 30, which non-temporarily records a search support program that causes a computer to execute a process of evaluating the search results based on the evaluation of the search results input by the user.

[0146] [Supplementary Note 36] A recording medium according to any one of Supplementary Notes 25 to 35, which non-temporarily records a search support program that causes a computer to execute a process of updating parameters of the generative model by performing machine learning using the user query, the search query, and the evaluation results as learning data.

[0147] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0148] REFERENCE SIGNS LIST 10 Search support device 11 Acquisition unit 12 Search query generation unit 13 Search query output unit 14 Search result acquisition unit 15 Evaluation unit 17 Model update unit 18 Output unit 19 Storage unit 20 Terminal device 30 Search device 100 Computer 101 Processor 102 Memory 103 Storage device 104 Input / output I / F 105 Communication I / F

Claims

1. A search support device comprising: an acquisition means for acquiring a user query input by a user; a search query generation means for generating a search query for executing a search in a search engine using a generative model based on the acquired user query; a search query output means for outputting the search query to the search engine; a search result acquisition means for acquiring search results for the search query from the search engine; an evaluation means for evaluating the search results based on criteria for evaluating the suitability of the search results for the user query; and a model update means for updating the generative model based on the results of the evaluation.

2. The search support device of claim 1, further comprising a determination means for determining whether a condition for outputting the search results is satisfied, and if the condition for outputting the search results is not satisfied, updating the generative model in the model update means, generating the search query using the updated generative model in the search query generation means, outputting the search query in the search query output means, acquiring the search results in the search result acquisition means, and evaluating the search results in the evaluation means are repeated as a repetitive process.

3. The search support device according to claim 2, further comprising an output means for outputting the search results when the result of the determination satisfies a condition for outputting the search results.

4. The search support device according to claim 2 or 3, wherein the model update means stores the last updated generative model in the repeated processing as the generative model whose update has been completed.

5. A search support device according to any one of claims 2 to 4, wherein the condition for outputting the search results is that at least one of the result of the evaluation and the number of times the repeated processing has been performed satisfies a criterion for outputting the search results.

6. A search support device as described in any one of claims 2 to 4, wherein the condition for outputting the search results is that a score based on the result of evaluation of the search results for the user query and the number of search results satisfies a criterion for outputting the search results.

7. A search support device according to any one of claims 1 to 6, wherein the evaluation means evaluates the search results using a criterion based on the similarity of the search results to the user query as a criterion for evaluating the suitability of the search results to the user query.

8. The search support device according to claim 7, wherein the similarity is a distance between an embedding vector of the user query and an embedding vector of the search result.

9. The search support device according to claim 1, wherein said evaluation means evaluates said search results by using an output from a language model when said search results are input to said language model.

10. The search support device according to claim 9, wherein the evaluation means evaluates the search results using a result of a dialogue with the user using an output from the language model.

11. A search support device according to any one of claims 1 to 6, wherein the evaluation means evaluates the search results based on an evaluation result of the search results input by the user.

12. A search support device as described in any one of claims 1 to 11, wherein the model update means updates parameters of the generative model by performing machine learning using the user query, the search query, and the evaluation results as learning data.

13. A search support method comprising: acquiring a user query input by a user; generating a search query, which is a query for executing a search in a search engine, using a generative model based on the acquired user query; outputting the search query to the search engine; acquiring search results for the search query from the search engine; evaluating the search results based on criteria for evaluating the suitability of the search results to the user query; and updating the generative model based on the results of the evaluation.

14. A search support method as described in claim 13, further comprising: determining whether a condition for outputting the search results is met; and if the condition for outputting the search results is not met, updating the generative model, generating the search query using the updated generative model, outputting the search query, obtaining the search results, and evaluating the search results again as an iterative process.

15. The search support method according to claim 14, further comprising the step of: outputting the search results when a condition for outputting the search results is satisfied as a result of the determination.

16. The search support method according to claim 14 or 15, wherein in the repeated processing, the last updated generative model is saved as the generative model whose updating has been completed.

17. A search support method according to any one of claims 14 to 16, wherein the condition for outputting the search results is that at least one of the result of the evaluation and the number of times the repeated processing has been performed satisfies a criterion for outputting the search results.

18. A search support method according to any one of claims 14 to 16, wherein the condition for outputting the search results is that a score based on the result of evaluation of the search results for the user query and the number of search results satisfies a criterion for outputting the search results.

19. A search support method according to any one of claims 13 to 18, further comprising: evaluating the search results using a criterion based on the similarity of the search results to the user query as a criterion for evaluating the suitability of the search results to the user query.

20. A recording medium that non-temporarily records a search support program that causes a computer to execute the following processes: a process of acquiring a user query, which is a query input by a user; a process of generating a search query, which is a query for executing a search in a search engine, using a generative model based on the acquired user query; a process of outputting the search query to the search engine; a process of acquiring search results for the search query from the search engine; a process of evaluating the search results based on criteria for evaluating the suitability of the search results to the user query; and a process of updating the generative model based on the results of the evaluation.

Images