Search control system, search control method and program product

Abstract

The prompt generation unit (46) of the retrieval control system of the present invention generates a prompt to the language model based on the chat messages received from the terminal (14). The prompt output unit (48) outputs the prompt to the language model. The display control unit (56) outputs the chat message based on the language model's response to the prompt to the terminal (14). The first retrieval intermediary unit (50a) generates a retrieval query based on the chat message exchange records between itself and the terminal (14). The first retrieval intermediary unit (50a) outputs the retrieval query to the first retrieval system (12a). The display control unit (56) outputs the retrieval results received by the first retrieval system (12a) based on the retrieval query being output to the first retrieval system (12a) to the terminal (14).

Description

Technical Field

[0001] This invention relates to a retrieval control system, a retrieval control method, and an information memory medium. Prior Technology

[0002] Patent document 1 describes a technology that supports users in searching for products on shopping websites on the Internet.

[0003] Furthermore, in recent years, the technology of utilizing large-scale language models (LLM) has attracted attention. As one example of such a technology, Patent Document 2 describes a technique for using a large-scale language model when generating automatic assistant responses.

[0004] [Previous Technical Documents] [Patent Literature] [Patent Document 1] Japanese Patent Application Publication No. 2011-133926 [Patent Document 2] Japanese Patent Publication No. 2024-521053 Summary of the Invention

[0005] [The problem that the invention aims to solve] By repeatedly outputting prompts to language models such as LLM or receiving responses from those language models, users can obtain information as if they were having a natural conversation with someone.

[0006] However, in order to perform certain searches based on information obtained through chat using language models, users need to input search queries into the search system, which is more cumbersome.

[0007] The present invention was made in view of the above-mentioned problems, and one of its objectives is to provide a retrieval control system, retrieval control method and information memory medium in which users can seamlessly obtain the retrieval results of the retrieval system in a way that allows them to chat naturally with others.

[0008] [Technical means to solve the problem] The retrieval control system of the present invention includes: a prompt generation mechanism that generates a prompt to a language model based on chat messages received by a terminal; a prompt output mechanism that outputs the prompt to the language model; a chat message output mechanism that outputs chat messages based on the language model's response to the prompt to the terminal; a retrieval query generation mechanism that generates a retrieval query based on chat message exchange records with the terminal; a retrieval query output mechanism that outputs the retrieval query to the retrieval system; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the retrieval query being output to the retrieval system to the terminal.

[0009] The retrieval control method of the present invention includes the following steps: generating a prompt to a language model based on a chat message received by the terminal; outputting the prompt to the language model; outputting a chat message based on the response of the language model to the prompt to the terminal; generating a retrieval query based on the chat message exchange records between the terminal and the terminal; outputting the retrieval query to a retrieval system; and outputting the retrieval results received by the retrieval system based on the retrieval query to the terminal.

[0010] The information memory medium of this invention stores a program for causing a computer to perform the following steps: generating a prompt to a language model based on a chat message received from the terminal; outputting the prompt to the language model; outputting a chat message based on the language model's response to the prompt to the terminal; generating a search query based on the chat message exchange records between the terminal and the terminal; outputting the search query to a search system; and outputting the search results received by the search system based on the search query to the terminal. Simple Explanation of the Diagram

[0011] Figure 1 is a diagram showing an example of the configuration of a chat system according to one embodiment of the present invention. Figure 2 shows an example of a chat screen. Figure 3 shows an example of a chat screen. Figure 4 is a functional block diagram showing an example of the functions of a retrieval control system according to one embodiment of the present invention. Figure 5A is a flowchart showing an example of the processing flow performed by the retrieval control system of one embodiment of the present invention. Figure 5B is a flowchart showing an example of the processing flow performed by the retrieval control system of one embodiment of the present invention. Figure 6 is a functional block diagram showing one example of the function of the retrieval control system of a variation of one embodiment of the present invention. Implementation

[0012] Hereinafter, one embodiment of the present invention will be described in detail based on the drawings.

[0013] Figure 1 is a diagram showing an example of the configuration of the chat system 1 of this embodiment. As shown in Figure 1, the chat system 1 of this embodiment includes a search control system 10, a search system 12 (a first search system 12a and a second search system 12b), and a terminal 14. The search control system 10, the first search system 12a, the second search system 12b, and the terminal 14 are connected to a computer network 16 such as the Internet.

[0014] The retrieval control system 10 may be composed of a computer, such as a server computer. The retrieval control system 10 may include a processor 10a (e.g., CPU or GPU), a memory unit 10b (e.g., memory or hard disk drive), a communication unit 10c (e.g., network interface card), etc. The retrieval control system 10 may include multiple computers or a single computer.

[0015] The retrieval system 12 (first retrieval system 12a and second retrieval system 12b) may, for example, be composed of a server computer including a CPU, GPU, memory, communication interface, etc. Furthermore, the retrieval system 12 may include multiple computers or a single computer.

[0016] Terminal 14 may be composed of a computer (e.g., personal computer, smartphone, tablet computer, etc.) that includes a CPU, memory, communication interface, input device and display.

[0017] In this embodiment, for example, chat messages are exchanged between the retrieval control system 10 and the terminal 14. During this exchange, the retrieval system 12 is appropriately executed. In the retrieval system 12, for example, a retrieval is performed for proposals (suggestions) to the user (more specifically, for example, a retrieval to generate options for the user to purchase goods or services).

[0018] In this embodiment, for example, the search objects of the first search system 12a are different from those of the second search system 12b. Furthermore, the search system 12 can perform searches that specify goods or services as search objects. For example, it is feasible for the search performed by the first search system 12a to specify information on goods sold online, while the search performed by the second search system 12b may specify information on travel package itineraries.

[0019] Figure 2 shows an example of a chat screen 20 displayed on the monitor of the terminal 14 in this embodiment. The chat screen 20 may be a screen that can be scrolled in a predetermined direction (e.g., vertically).

[0020] In the chat screen 20 shown in Figure 2, user message images 22 (22a, 22b, and 22c) are configured to represent chat messages input by the user of the operating terminal 14.

[0021] Furthermore, in the chat screen 20 shown in Figure 2, a response message image 24 (24a, 24b) is configured to represent the chat message sent from the retrieval control system 10 to the terminal 14 based on the chat message received from the terminal 14. Also, the response message image 24 may include option images 26 representing the options available to the user. In the example of Figure 2, response message image 24a includes option images 26a, 26b, 26c, and 26d. Furthermore, response message image 24b includes option images 26e, 26f, 26g, 26h, and 26i.

[0022] Furthermore, in the chat screen 20 shown in Figure 2, an input form 28 is configured for the user to input chat messages.

[0023] When a user enters a string in input form 28 and performs a specified operation (e.g., pressing the Enter key or tapping the send button), a chat message containing the string entered in input form 28 is sent from terminal 14 to retrieval control system 10.

[0024] Furthermore, when a user performs an operation to select option image 26 (e.g., a tap operation on option image 26), a chat message containing the string configured in the selected option image 26 is sent from terminal 14 to retrieval control system 10.

[0025] Furthermore, a new chat screen 20, which includes a user message image 22 containing chat messages sent from terminal 14 to the retrieval control system 10, is displayed on the monitor of terminal 14.

[0026] In the chat screen 20 shown in Figure 2, by selecting the option image 26i, the chat message "Search for me" is displayed, indicating that it is sent from the terminal 14 to the search control system 10.

[0027] Thus, in this embodiment, for example, as shown in FIG3, the search results executed by the search system 12 are sent from the search control system 10 to the terminal 14, and a chat screen 20 displaying the search result image 30 representing the search result is shown on the display of the terminal 14. As shown in FIG3, the search results may display goods or services. For example, as shown in FIG3, the search results may include an image of the goods or services, the name of the goods or services, and the price of the goods or services.

[0028] Furthermore, as shown in Figure 3, the display of terminal 14 can show a chat screen 20 that arranges chat messages and search results in a time sequence. For example, as shown in Figure 3, a chat screen 20 can be displayed that arranges chat messages and search results in a time sequence from oldest to newest, from top to bottom.

[0029] Furthermore, in this embodiment, for example, by having the user perform a prescribed operation on the search results shown in the search result image 30 (e.g., a tap operation), the screen displayed on the terminal 14 changes to a purchase page for purchasing the goods or services shown in the search results. The user can then purchase the goods or services through this purchase page.

[0030] In this embodiment, users can seamlessly obtain search results from the search system 12 as if they were having a natural conversation with someone.

[0031] The functions of the retrieval control system 10 in this embodiment and the processing performed by the retrieval control system 10 will be further explained below.

[0032] Figure 4 is a functional block diagram showing one example of the functions installed in the retrieval control system 10 of this embodiment. Furthermore, in the retrieval control system 10 of this embodiment, it is not necessary to install all the functions shown in Figure 4, and functions other than those shown in Figure 4 can also be installed.

[0033] As shown in Figure 4, the retrieval control system 10 of this embodiment functionally includes, for example, a recording and memory unit 40, a chat message receiving unit 42, an intent determination unit 44, a prompt generation unit 46, a prompt output unit 48, a retrieval intermediary unit 50 (a first retrieval intermediary unit 50a and a second retrieval intermediary unit 50b), a general dialogue unit 52, a chat message generation unit 54, and a display control unit 56. The recording and memory unit 40 is mainly installed based on the memory unit 10b. The chat message receiving unit 42 is mainly installed based on the communication unit 10c. The intent determination unit 44, the prompt generation unit 46, the prompt output unit 48, the general dialogue unit 52, and the chat message generation unit 54 are mainly installed based on the processor 10a. The first retrieval intermediary unit 50a, the second retrieval intermediary unit 50b, and the display control unit 56 are mainly installed based on the processor 10a and the communication unit 10c.

[0034] The above functions can be installed by the search and control system 10 executing a program containing instructions corresponding to the above functions, which is installed on the search and control system 10, which is a computer. Furthermore, this program can be supplied to the search and control system 10 via computer-readable information storage media such as CDs, magnetic disks, magnetic tapes, or the Internet.

[0035] In this embodiment, the recording memory unit 40 stores, for example, records of chat messages exchanged between the retrieval control system 10 and the terminal 14. For instance, the recording memory unit 40 can store records of chat messages received by the retrieval control system 10 from the terminal 14, or chat messages output from the retrieval control system 10 to the terminal 14. Here, for example, the recording memory unit 40 can also store chat messages associated with a timestamp indicating the date and time of receiving or outputting the chat message, and a flag indicating whether the chat message was received by the retrieval control system 10 or output by the retrieval control system 10.

[0036] In this embodiment, the chat message receiving unit 42 receives chat messages from the terminal 14, for example. Furthermore, in this embodiment, the chat message receiving unit 42 stores the chat messages in the recording memory unit 40, for example.

[0037] In this embodiment, the intent determination unit 44 uses existing intent determination technologies such as natural language processing (NLP) models, large-scale language models (LLM) or knowledge graphs to determine the user's intent based on the chat messages received by the terminal 14.

[0038] For example, the intent determination unit 44 may include a pre-trained language model (e.g., LLM) for intent determination. Hereinafter, this language model will be referred to as the intent determination LLM. Furthermore, the intent determination unit 44 may output, for example, a template such as "Please refer to the records at the following location to determine your intent" and a prompt containing a string representing the location of the recording memory unit 40 (e.g., a URL) to the intent determination LLM. Then, the intent determination unit 44 can determine the user's intent based on the response output from the intent determination LLM according to the prompt. In this way, the intent determination unit 44 can determine the user's intent based on the records stored in the recording memory unit 40.

[0039] Here, for example, the intent determination unit 44 may determine the user's intent based on at least one of the records of chat messages received by the terminal 14 in the recording memory unit 40 or the records of chat messages output to the terminal 14.

[0040] Alternatively, the user's intent can be determined by weighting the chat messages based on the date and time of receipt or output.

[0041] Furthermore, the intent determination unit 44 can determine the user's intent based on the latest chat message received from the terminal 14. For example, the intent determination unit 44 can output a template containing "Please determine your intent based on the following message" and a prompt of the latest chat message received from the terminal 14 to the intent determination LLM. Then, the intent determination unit 44 can determine the user's intent based on the response output from the intent determination LLM according to the prompt.

[0042] In this embodiment, the prompt generation unit 46 generates a prompt for a language model (e.g., LLM) based on the chat message received by the terminal 14. Here, for example, the prompt generation unit 46 may generate a prompt that includes the chat message received by the chat message receiving unit 42.

[0043] In this embodiment, the prompt output unit 48 outputs the prompt generated by the prompt generation unit 46 to the language model.

[0044] Here, each of the first retrieval mediation unit 50a, the second retrieval mediation unit 50b, and the general dialogue unit 52 may contain a pre-trained language model (e.g., an LLM). Hereinafter, the language model contained in the first retrieval mediation unit 50a will be referred to as the first LLM, the language model contained in the second retrieval mediation unit 50b will be referred to as the second LLM, and the language model contained in the general dialogue unit 52 will be referred to as the third LLM.

[0045] Furthermore, the prompt output unit 48 can output the prompt generated by the prompt generation unit 46 to any one of the 1st LLM, the 2nd LLM, or the 3rd LLM.

[0046] Here, the intent determination unit 44 can determine the language model (hereinafter referred to as the target model) of the prompt output object from a plurality of language models based on the records stored in the recording memory unit 40. For example, the intent determination unit 44 can determine the user's intent based on the records stored in the recording memory unit 40, and determine the target model from a plurality of language models based on the determined intent. Here, for example, the intent determination unit 44 can store data that establishes an association between the intent and the language model. Then, the intent determination unit 44 can determine the language model in the data that is associated with the intent determined by the intent determination unit 44 as the target model. Then, the prompt output unit 48 can output a prompt to the determined target model.

[0047] Specifically, for example, when the user's intention is determined to be the purchase of goods (e.g., buying a gift for a partner), the intention determination unit 44 may determine the first LLM as the target model. Similarly, for example, when the user's intention is determined to be the travel (e.g., a trip with a partner), the intention determination unit 44 may determine the second LLM as the target model. Furthermore, when the user's intention is not any of the above, the intention determination unit 44 may determine the third LLM as the target model. Then, a prompt may be output to the target model determined in this way.

[0048] Furthermore, in this embodiment, the intent determination unit 44 can repeatedly execute the process of determining the target model from among multiple language models based on the records stored in the recording memory unit 40. For example, the intent determination unit 44 can perform the process of determining the target model whenever the terminal 14 receives a chat message. Then, the prompt output unit 48 can output a prompt based on the chat message received by the terminal 14 to the target model at the time the chat message was received.

[0049] The retrieval intermediary unit 50 (in this embodiment, for example, the first retrieval intermediary unit 50a and the second retrieval intermediary unit 50b) generates, for example, a response to the prompts received. In this embodiment, it is assumed that knowledge related to the retrieval object of the first retrieval system 12a (e.g., goods sold in online shopping) is pre-registered (learned) in the language model contained in the first retrieval intermediary unit 50a, i.e., the first LLM. Similarly, in this embodiment, it is assumed that knowledge related to the retrieval object of the second retrieval system 12b (e.g., travel package itineraries) is pre-registered (learned) in the language model contained in the second retrieval intermediary unit 50b, i.e., the second LLM. Furthermore, in this embodiment, it is assumed that both the first LLM and the second LLM can be referenced and stored in the recording memory unit 40.

[0050] The retrieval intermediary unit 50 can determine whether it has obtained sufficient information (context) to perform the retrieval of the retrieval system 12 based on the records stored in the recording memory unit 40.

[0051] For example, when the target model is the first LLM, the prompt generation unit 46 can generate a prompt containing a template that says "Please refer to the record at the following location to determine whether a product search can be performed" and a string representing the location of the record memory unit 40 (e.g., URL). Then, when the target model is the first LLM, the prompt output unit 48 can output the prompt to the first LLM. Then, when the first search intermediary unit 50a receives an affirmative response (e.g., "Yes") from the first LLM to the prompt output, it can determine that sufficient information (context) has been obtained to perform the search of the first search system 12a.

[0052] Furthermore, when the target model is the second LLM, the prompt generation unit 46 can generate a prompt containing a template that reads "Please refer to the records at the following locations to determine whether a search for a travel package itinerary is possible," and a string representing the location of the record memory unit 40 (e.g., a URL). The prompt output unit 48 can output this prompt to the second LLM. Then, when the second retrieval intermediary unit 50b receives an affirmative response (e.g., a "yes") from the second LLM to the prompt output, it can determine that sufficient information (context) has been obtained to perform the search of the second retrieval system 12b.

[0053] Further, for example, the search intermediary unit 50 can determine whether at least one preset attribute (for example, gender, age group, address, interests, number of family members, etc.) can be extracted from the records stored in the record storage unit 40. Then, when it is determined that extraction is possible, the search intermediary unit 50 determines that sufficient information (context) for performing the search by the search system 12 has been obtained. Furthermore, the attributes determined by the first search intermediary unit 50a and the attributes determined by the second search intermediary unit 50b may be the same or different.

[0054] Then, as described later, when it is determined that sufficient information (context) for performing the search by the search system 12 has been obtained, a prescribed chat message indicating that search is executable (for example, a chat message including the option image 26i with the character string "Search for me") can be sent to the terminal 14.

[0055] Furthermore, in the present embodiment, the search intermediary unit 50 can generate a search query based on the interaction record of chat messages between the search control system 10 and the terminal 14, for example. The search query may be a keyword used for searching goods or services. Here, for example, the search intermediary unit 50 can generate a search query when a predetermined search execution condition is satisfied. The search execution condition may be, for example, a condition that the chat message received by the chat message receiving unit 42 contains a predetermined character string (for example, "Search for me"). Furthermore, the predetermined character string is not necessarily the character string "Search for me", and for example, it may also be the Japanese character string "を検索して (Search for me)". In addition, the search execution condition may also be a condition that it is determined that sufficient information (context) for performing the search by the search system 12 has been obtained.

[0056] Here, it is feasible that, for example, when the chat message received by the chat message receiving unit 42 satisfies a predetermined search execution condition, the prompt generating unit 46 generates a search query generation prompt that instructs generation of a search query based on the records stored in the record storage unit 40, and the prompt output unit 48 outputs the search query generation prompt to the language model. Then, the search intermediary unit 50 can generate the search query based on a reply output from the language model in response to the search query generation prompt being output to the language model.

[0057] For example, when the target model is a first LLM, the search query generation prompt generated by the prompt generating unit 46 can be output to the first LLM. Then, the first search intermediary unit 50a can generate the search query based on a reply output from the first LLM in response to the search query generation prompt being output to the first LLM.

[0058] For example, it is possible for the prompt generation unit 46 to generate a prompt containing a template that says "Please refer to the record at the following location and generate a suitable Japanese keyword phrase for the search in one sentence," and a string indicating the location of the record memory unit 40 (e.g., URL), and the prompt output unit 48 to output the prompt to the first LLM. Alternatively, it is possible for the prompt generation unit 46 to generate a prompt containing the content of the record to be referenced and an instruction to generate a search query, and the prompt output unit 48 to output the prompt to the first LLM. Then, the first search intermediary unit 50a can obtain the response from the first LLM to the prompt output as a search query.

[0059] For example, if the target model is the second LLM, the search query generated by the prompt generation unit 46 can be output to the second LLM. Then, the second search intermediary unit 50b can generate a search query based on the response output from the second LLM after the search query prompt is output to the second LLM.

[0060] For example, it is feasible for the prompt generation unit 46 to generate a prompt containing a template that says "Please refer to the record at the following location and generate a Japanese keyword phrase suitable for searching a travel package itinerary in one sentence," and a string indicating the location of the record memory unit 40 (e.g., URL), and the prompt output unit 48 to output the prompt to the second LLM. Alternatively, it is feasible for the prompt generation unit 46 to generate a prompt containing the content of the record to be referenced and an instruction to generate a search query, and the prompt output unit 48 to output the prompt to the second LLM. Then, the second search intermediary unit 50b can obtain the response from the second LLM to the prompt output as a search query.

[0061] Furthermore, the search intermediary department 50 can generate search queries based on chat messages received by the terminal 14 (e.g., the latest chat messages received by the terminal 14).

[0062] Hereinafter, the search query generated by the search intermediary department 50 will be referred to as the generated search query.

[0063] Then, in this embodiment, the retrieval intermediary unit 50, for example, outputs the generated retrieval query to the retrieval system 12 (e.g., the retrieval system 12 associated with the retrieval intermediary unit 50). Then, in this embodiment, the retrieval intermediary unit 50, for example, accepts the retrieval results output from the retrieval system 12 based on the generated retrieval query being output to the retrieval system 12.

[0064] For example, the first search intermediary unit 50a outputs the generated search query to the first search system 12a. Then, in this embodiment, the first search intermediary unit 50a, for example, receives the search results output from the first search system 12a based on the generated search query being output to the first search system 12a.

[0065] Furthermore, in this embodiment, the second search intermediary unit 50b, for example, outputs the generated search query to the second search system 12b. Then, in this embodiment, the second search intermediary unit 50b, for example, receives the search results output from the second search system 12b based on the generated search query being output to the second search system 12b.

[0066] In this embodiment, when performing a search using the retrieval system 12, it is feasible for the prompt generation unit 46 to generate a prompt containing the search results of the retrieval system 12, and the prompt output unit 48 to output the prompt to the language model. Furthermore, the prompt may include chat messages received by the chat message receiving unit 42. Also, the prompt may include reference instructions stored in the recording memory unit 40. Then, the retrieval intermediary unit 50 can obtain the language model's response to the prompt output.

[0067] For example, if the target model is the first LLM and a search is performed using the first search system 12a, it is feasible for the suggestion generation unit 46 to generate a suggestion containing the search results of the first search system 12a, and the suggestion output unit 48 to output the suggestion to the first LLM. Then, the first search intermediary unit 50a can obtain the response from the first LLM to the suggestion output.

[0068] For example, if the target model is the second LLM and a search is performed using the second search system 12b, it is feasible for the prompt generation unit 46 to generate a prompt containing the search results of the second search system 12b, and the prompt output unit 48 to output the prompt to the second LLM. Then, the second search intermediary unit 50b can obtain the response from the second LLM to the prompt output.

[0069] On the other hand, if the retrieval system 12 is not performed, it is feasible for the prompt generation unit 46 to generate a prompt containing the chat message received by the chat message receiving unit 42, and the prompt output unit 48 to output the prompt to the language model. Furthermore, the prompt may include reference instructions stored in the recording memory unit 40. Then, the retrieval mediation unit 50 can obtain the response from the language model to the prompt output.

[0070] For example, if the target model is the first LLM and the search of the first search system 12a is not performed, it is possible for the prompt generation unit 46 to generate a prompt containing the chat message received by the chat message receiving unit 42, and the prompt output unit 48 to output the prompt to the first LLM. Then, the first search intermediary unit 50a can obtain the response from the first LLM to the prompt output.

[0071] Furthermore, if the target model is the second LLM and the search of the second search system 12b is not performed, it is feasible for the prompt generation unit 46 to generate a prompt containing the chat message received by the chat message receiving unit 42, and the prompt output unit 48 to output the prompt to the second LLM. Then, the second search intermediary unit 50b can obtain the response from the second LLM to the prompt output.

[0072] In this embodiment, the general dialogue unit 52 generates, for example, a response to a prompt for acceptance. In this embodiment, it is assumed that the third LLM can refer to the record stored in the recording memory unit 40.

[0073] For example, if the target model is the 3rd LLM, it is feasible for the prompt generation unit 46 to generate a prompt containing the chat message received by the chat message receiving unit 42, and the prompt output unit 48 to output the prompt to the 3rd LLM. Furthermore, the prompt may contain a reference instruction stored in the recording memory unit 40. Then, the general dialogue unit 52 can obtain the 3rd LLM's response to the prompt output.

[0074] In this embodiment, the chat message generation unit 54 generates chat messages that are output to the terminal 14, for example.

[0075] For example, when the target model is the first LLM, the chat message generation unit 54 can generate a chat message that includes the response generated by the first LLM. Here, when the first retrieval intermediary unit 50a receives the retrieval results output by the first retrieval system 12a, the chat message generation unit 54 can generate a chat message that includes the retrieval results received by the first retrieval intermediary unit 50a from the first retrieval system 12a and the response generated by the first LLM.

[0076] For example, when the target model is the second LLM, the chat message generation unit 54 can generate a chat message that includes the response generated by the second LLM. Here, when the second retrieval intermediary unit 50b receives the retrieval results output from the second retrieval system 12b, the chat message generation unit 54 can generate a chat message that includes the retrieval results received by the second retrieval intermediary unit 50b from the second retrieval system 12b and the response generated by the second LLM.

[0077] For example, if the target model is the 3rd LLM, the chat message generation unit 54 can generate a chat message that includes the reply generated by the 3rd LLM.

[0078] Furthermore, in this embodiment, the chat message generation unit 54 can also generate a chat message indicating that an executable search is possible based on the fulfillment of given executable search conditions. For example, the chat message generation unit 54 can generate a chat message containing an option image 26i with the string "Search for me" configured, based on the fulfillment of given executable search conditions. Here, executable search conditions can be, for example, conditions that determine whether sufficient information (context) is obtained to perform a search.

[0079] Furthermore, a prompt containing an indication of the generated options to be presented to the user can be output to the 1st LLM, 2nd LLM, or 3rd LLM. Then, the chat message generation unit 54 can generate a chat message containing option images 26 that correspond to one or more options output from the LLMs based on the prompt.

[0080] In this embodiment, the display control unit 56, for example, displays the chat screen 20 on the display of the terminal 14.

[0081] For example, the display control unit 56 can generate a new chat screen 20 containing the chat message when the chat message receiving unit 42 receives the chat message from the terminal 14, and send the chat screen 20 to the terminal 14. Then, the terminal 14 can display the chat screen 20 on the display screen of the terminal 14 upon receiving the chat screen 20.

[0082] Furthermore, the display control unit 56 can generate a chat message based on the chat message generation unit 54, generate a new chat screen 20 containing the chat message, and send the chat screen 20 to the terminal 14. Here, as described above, the chat message may include the search results received by the self-search system 12 (for example, search result image 30). Moreover, the chat message may include a portion of the search results received by the self-search system 12 (for example, displaying a portion of the search results received by the self-search system 12, as shown in the search result image 30). Then, the terminal 14 can display the chat screen 20 on its display screen upon receiving the chat screen 20.

[0083] Thus, the display control unit 56 of this embodiment functions as a chat message output unit that outputs chat messages based on responses to prompts from the language model to the terminal 14. Furthermore, the display control unit 56 of this embodiment functions as a search result output unit that outputs search results received by the search system 12 based on generated search queries to the terminal 14.

[0084] In this embodiment, it is feasible for the retrieval intermediary unit 50 to generate a retrieval query output to the retrieval system 12 based on the acceptance of a retrieval request corresponding to the output of a chat message indicating that a retrieval can be performed (for example, the acceptance of a "Search for me" chat message corresponding to the output of an option image 26i containing the string "Search for me").

[0085] For example, based on the determination that sufficient information (context) is obtained for the first search intermediary unit 50a to perform a search, the chat message generation unit 54 can generate a chat message containing an option image 26i configured with the string "Search for me". Then, the display control unit 56 can send the chat screen 20 containing the option image 26i to the terminal 14. Then, based on the acceptance of the "Search for me" chat message corresponding to the sending of the chat screen 20, the first search intermediary unit 50a can generate a search query and output the generated search query to the first search system 12a.

[0086] Further, for example, when it is determined that sufficient information (context) for performing a search by the second search mediation unit 50b is obtained, the chat message generating unit 54 may generate a chat message including an option image 26i having the character string "Search for me" arranged therein. Then, the display control unit 56 may transmit the chat screen 20 including the option image 26i to the terminal 14. Then, in response to receiving the chat message of "Search for me" corresponding to the transmission of the chat screen 20, the second search mediation unit 50b may generate a generated search query, and output the generated generated search query to the second search system 12b.

[0087] Furthermore, when a search request corresponding to the output of a chat message specifying that search can be performed is received, the search mediation unit 50 may not generate a generated search query, and output the generated search query to the search system 12. Even when no chat message specifying that search can be performed is output, in response to receiving a search request, the search mediation unit 50 may generate a generated search query, and output the generated generated search query to the search system 12. For example, when the chat message receiving unit 42 receives a message including a predetermined character string (for example, the character string "Search for me" expressed in Japanese as "を検索して"), the search mediation unit 50 may generate a generated search query, and output the generated generated search query to the search system 12.

[0088] As described above, in the present embodiment, the plurality of language models (the first LLM, the second LLM, and the third LLM in the above example) are each either a language model associated with the search system 12 or a language model not associated with the search system 12. In the above example, the first LLM and the second LLM are language models associated with the search system 12, and the third LLM is a language model not associated with the search system 12.

[0089] Further, as described above, in the present embodiment, each of the plurality of language models is associated with a search system 12 having a different search object. In the above example, the first LLM is associated with the first search system 12a, and the second LLM is associated with the second search system 12b.

[0090] Furthermore, in the present embodiment, for example, when the above search execution condition is satisfied in a case where a target model is a language model associated with the search system 12, a generated search query is output to the search system 12 associated with the target model.

[0091] For example, based on the above-mentioned search execution conditions being met when the target model is the first LLM, the first search intermediary unit 50a outputs a generated search query to the first search system 12a. Then, the display control unit 56 outputs the search results of the first search system 12a for the search object, which were processed by the first search system 12a and output to the terminal 14, based on the generated search query being output to the first search system 12a.

[0092] Furthermore, based on the above-mentioned search execution conditions being met when the target model is the second LLM, the second search intermediary unit 50b outputs a generated search query to the second search system 12b. Then, the display control unit 56 outputs the search results of the second search system 12b for the search object, which were processed by the second search system 12b and output to the terminal 14, based on the generated search query being output to the second search system 12b.

[0093] Furthermore, as described above, it is feasible to refer to and store multiple language models (in the example above, the 1st LLM, the 2nd LLM, and the 3rd LLM) in the recording memory unit 40. Moreover, the records stored in the recording memory unit 40 can be records of chat messages exchanged through these multiple language models.

[0094] In this embodiment, the prompt generation unit 46 can store the prompts generated by the prompt generation unit 46 in the recording memory unit 40. Furthermore, the display control unit 56 can store the chat screen 20 generated by the display control unit 56 in the recording memory unit 40.

[0095] Furthermore, in this embodiment, the first search intermediary unit 50a can store the search query generated by the first search intermediary unit 50a in the recording memory unit 40. Also, the second search intermediary unit 50b can store the search query generated by the second search intermediary unit 50b in the recording memory unit 40.

[0096] Furthermore, in this embodiment, the records stored in the recording memory unit 40 may include the search results performed by the search system 12. For example, the first search intermediary unit 50a may store the search results received by the first search system 12a in the recording memory unit 40. Similarly, the second search intermediary unit 50b may store the search results received by the second search system 12b in the recording memory unit 40.

[0097] Moreover, the various data stored in the memory recording unit 40 can be referenced in the various processes described above.

[0098] Here, an example of the process performed by the retrieval control system 10 of this embodiment is described with reference to the flowcharts illustrated in Figures 5A and 5B.

[0099] When the chat message receiving unit 42 receives a chat message from the terminal 14 (S101), the display control unit 56 generates a chat screen 20 containing the chat message and sends the chat screen 20 to the terminal 14 (S102). Upon receiving the chat screen 20, the terminal 14 updates the chat screen 20 displayed on the screen to the received chat screen 20.

[0100] Then, the intent determination unit 44 determines the user's intent (S103). Then, based on the intent determined by the process shown in S103, the intent determination unit 44 determines the target model (S104).

[0101] Then, the prompt generation unit 46 determines whether a search query needs to be generated based on the chat message received through the processing shown in S101 (S105).

[0102] When it is determined that no search query needs to be generated (S105: No), the prompt generation unit 46 generates a prompt containing the chat message received by the process shown in S101 (S106), and the prompt output unit 48 outputs the prompt generated by the process shown in S106 to the target model (S107).

[0103] Then, the chat message generation unit 54 generates a chat message based on the output of the target model to the prompt output to the target through the processing shown in S107 (S108).

[0104] Then, the display control unit 56 generates a chat screen 20 containing the chat messages generated by the process shown in S108, and sends the chat screen 20 to the terminal 14 (S109), ending the process shown in this example. Upon receiving the chat screen 20, the terminal 14 updates the chat screen 20 displayed on the screen to the received chat screen 20.

[0105] When it is determined by the process shown in S105 that a search query needs to be generated (S105: Yes), the prompt generation unit 46 generates a search query generation prompt (S110), and the prompt output unit 48 outputs the search query generation prompt generated by the process shown in S110 to the target model (S111).

[0106] Then, the retrieval intermediary unit 50, which includes the target model, obtains the output from the target model that provides a prompt for the retrieval query output to the target model through the processing shown in S111, and thus generates a retrieval query (S112).

[0107] Then, the search intermediary unit 50 outputs the generated search query obtained by the process shown in S112 to the search system 12 associated with the search intermediary unit 50 (S113), and the search intermediary unit 50 accepts the search results output from the search system 12 based on the generated search query (S114).

[0108] Then, the prompt generation unit 46 generates a prompt containing the search results received through the process shown in S114 (S115), and the prompt output unit 48 outputs the prompt generated through the process shown in S115 to the target model (S116).

[0109] Then, the chat message generation unit 54 generates a chat message (S117) based on the search results received by the processing shown in S114 and the output of the target model to the prompts output to the target model by the processing shown in S116.

[0110] Then, the display control unit 56 generates a chat screen 20 containing the chat messages generated by the process shown in S117, and sends the chat screen 20 to the terminal 14 (S118), ending the process shown in this example. Upon receiving the chat screen 20, the terminal 14 updates the chat screen 20 displayed on the screen to the received chat screen 20.

[0111] The processing shown in S101 to S118 is performed in the chat message receiving unit 42 whenever a chat message is received.

[0112] According to this embodiment, as described above, users can seamlessly obtain the search results of the search system 12 through the chat screen 20 in a way that feels like chatting naturally with a person.

[0113] In this embodiment, multiple language models (in the example above, the 1st LLM, the 2nd LLM, and the 3rd LLM) can all be pre-set to generate responses in a common communication style.

[0114] For example, prompts such as "Communication Style: • Use emojis as needed to maintain a positive and cheerful tone. • Show excitement for a positive user experience. • Use the first person to create a more welcoming impression (e.g., "I recommend..."). • Offer non-authoritative advice or suggestions as a friendly support." can be pre-outputted for the 1st, 2nd, and 3rd LLMs.

[0115] By operating in this way, even though chatting through multiple language models can be conducted, it can avoid causing discomfort to the user.

[0116] Furthermore, in this embodiment, as described above, the language models used in generating chat messages do not need to be three (in the example above, the 1st LLM, the 2nd LLM, and the 3rd LLM), nor do they need to be multiple.

[0117] For example, as shown in Figure 6, it is feasible to omit the first retrieval mediation unit 50a, the second retrieval mediation unit 50b, and the general dialogue unit 52 from the retrieval control system 10, and instead include a comprehensive retrieval mediation unit 60. The comprehensive retrieval mediation unit 60 may include a pre-trained language model (e.g., LLM). Hereinafter, the language model included in the comprehensive retrieval mediation unit 60 will be referred to as the comprehensive LLM.

[0118] Furthermore, the prompt generation unit 46 can pre-memorize multiple templates. Moreover, the intent determination unit 44 can repeatedly execute the process of determining the target template from among the multiple templates based on the records stored in the recording unit 40. For example, the intent determination unit 44 can execute the process of determining the target template whenever a chat message is received from the terminal 14.

[0119] Here, for example, the intent determination unit 44 may memorize data that associates intent with templates. Then, the intent determination unit 44 may select the template in the data that is associated with the intent determined by the intent determination unit 44 as the target template.

[0120] Then, the prompt generation unit 46 can generate a prompt based on the chat message received by the terminal 14 and the target template at the time of receiving the chat message. For example, the prompt generation unit 46 can generate a prompt that includes the chat message received by the chat message receiving unit 42 and the target template. Then, the prompt output unit 48 can output the prompt to the integrated LLM.

[0121] Furthermore, the plurality of templates can be either templates associated with the retrieval system 12 or templates not associated with the retrieval system 12. Also, each of the plurality of templates can be associated with a retrieval system 12 that has different search targets.

[0122] For example, the plurality of templates can be a first template associated with the first retrieval system 12a, a second template associated with the second retrieval system 12b, and a third template not associated with any retrieval system 12. Here, the first template may contain information related to the first retrieval system 12a (e.g., the URL of the first retrieval system 12a). Similarly, the second template may contain information related to the second retrieval system 12b (e.g., the URL of the second retrieval system 12b).

[0123] Furthermore, if the above-mentioned search execution conditions are met when the target template is a template associated with the search system 12, a search query can be generated and output to the search system 12 associated with the target template. Then, the search results of the search system 12 for the search object, which are processed by the search system 12 based on the generated search query, can be output to the terminal 14.

[0124] For example, if the above-mentioned search execution conditions are met when the target template is the first template, the comprehensive search intermediary unit 60 can output a generated search query to the first search system 12a. Then, the display control unit 56 can output the search results of the first search system 12a for the search object, which were processed by the first search system 12a after the generated search query was output to the first search system 12a, to the terminal 14.

[0125] Furthermore, based on the above-mentioned search execution conditions being met when the target template is the second LLM, the integrated search intermediary unit 60 can output a generated search query to the second search system 12b. Then, the display control unit 56 can output the search results of the second search system 12b for the search object, which were processed by the second search system 12b after the generated search query was output to the second search system 12b, to the terminal 14.

[0126] Furthermore, the present invention is not limited to the embodiments described above.

[0127] For example, the language model in this embodiment may not be a large-scale language model (LLM). For example, the language model in this embodiment may be an n-gram model, a recurrent neural network (RNN), a Seq2Seq model, a word vector model, etc.

[0128] Furthermore, the search targets of the search system 12 in this embodiment are not limited to goods or services.

[0129] Furthermore, the retrieval system 12 included in the chat system 1 of this embodiment does not need to be two.

[0130] Furthermore, the retrieval intermediary unit 50 and the retrieval system 12 are not necessarily linked one-to-one. For example, the first retrieval intermediary unit 50a may be linked to multiple retrieval systems 12. Similarly, the second retrieval intermediary unit 50b may be linked to multiple retrieval systems 12.

[0131] Furthermore, at least a portion of the pre-trained language model described above may not be included in the retrieval control system 10, but may be included in an external system that can communicate with the retrieval control system 10.

[0132] Furthermore, the specific character strings or values ​​mentioned above and the specific character strings or values ​​in the diagrams are examples, and are not limited to such character strings or values.

[0133] For example, the retrieval control system can be configured as follows. [1] A retrieval control system includes: a prompt generation mechanism that generates prompts for a language model based on chat messages received by a terminal; The prompt output mechanism outputs the aforementioned prompt to the aforementioned language model; The chat message output mechanism outputs chat messages to the aforementioned terminal based on the responses to the aforementioned prompts from the aforementioned language model; The retrieval query generating organization generates retrieval queries based on chat message exchange records with the aforementioned terminals; The retrieval query output mechanism outputs the aforementioned retrieval query to the retrieval system; and The search result outputting mechanism outputs the search results received by the aforementioned search system to the aforementioned terminal based on the aforementioned search query. [2] As in the retrieval control system of [1], the aforementioned retrieval query generating mechanism generates an instruction based on the aforementioned record to generate a retrieval query prompt. The aforementioned search query is generated by the search query generating organization based on the response output from the language model after the prompts generated according to the aforementioned search query are output to the language model. [3] The retrieval control system, such as [1] or [2], further includes a decision-making mechanism that, based on the aforementioned records, repeatedly performs the process of determining the target model from among several language models. The aforementioned prompt output mechanism will output a prompt based on the chat message received by the aforementioned terminal to the target model mentioned above when the chat message was received. The aforementioned records are records of chat messages exchanged through the aforementioned plurality of the aforementioned language models. [4] The retrieval control system, as described in [3], wherein the aforementioned decision-making mechanism performs the processing of deciding the aforementioned target model whenever it receives a chat message from the aforementioned terminal. [5] Such as the retrieval control system in [3] or [4], wherein the aforementioned plurality of the aforementioned language models are either language models that are associated with the retrieval system or language models that are not associated with the retrieval system. The aforementioned retrieval query output mechanism outputs the aforementioned retrieval query to the retrieval system associated with the target model, based on the given conditions that are met when the target model is a language model associated with the retrieval engine. [6] For example, in any of [3] to [5], a retrieval control system is established, wherein each of the aforementioned multiple language models is associated with a retrieval system whose retrieval objects are different from each other. The aforementioned retrieval query output mechanism outputs the aforementioned retrieval query to the retrieval system that is associated with the aforementioned target model, based on the given conditions. The aforementioned search result outputting agency will output the search results of the aforementioned search system to the aforementioned terminal based on the aforementioned search query being output to the aforementioned search system. [7] For example, the retrieval control system of any of [3] to [6], wherein the aforementioned multiple language models can refer to the aforementioned records. [8] Such as the retrieval control system of any of [3] to [7], wherein the aforementioned multiple language models are all pre-set to generate answers in a common communication style. [9] The retrieval control system, such as [1] or [2], further includes a decision mechanism that, based on the aforementioned records, repeatedly performs the process of determining a target template from among several templates, and The aforementioned prompt output mechanism will output the prompt based on the chat message received from the aforementioned terminal and the prompt of the aforementioned target template when the chat message is received to the aforementioned language model.

[10] The retrieval control system, as described in [9], wherein the aforementioned decision-making body performs the process of deciding the aforementioned target template whenever it receives a chat message from the aforementioned terminal.

[11] Such as the retrieval control system in [9] or

[10] , wherein the aforementioned plurality of the aforementioned templates are either templates that are associated with the retrieval system or templates that are not associated with the retrieval system. The aforementioned search query output mechanism outputs the aforementioned search query to the search system associated with the target template, based on the given conditions being met when the target template is a template associated with the search system.

[12] Such as the retrieval control system in any of [9] to

[11] , wherein each of the aforementioned plurality of templates is associated with a retrieval system whose retrieval objects are different from each other. The aforementioned search query output mechanism outputs the aforementioned search query to the search system associated with the aforementioned target template based on the given conditions. The aforementioned search result outputting agency will output the search results of the aforementioned search system to the aforementioned terminal based on the aforementioned search query being output to the aforementioned search system.

[13] The retrieval control system of any one of [1] to

[12] , wherein the aforementioned chat message output mechanism outputs a chat message indicating that a retrieval can be performed based on the fulfillment of given conditions. The aforementioned search query output agency accepts the search request corresponding to the output of the chat message specified above and outputs the aforementioned search query to the aforementioned search system.

[14] As in the retrieval control system of

[13] , the aforementioned retrieval query output mechanism accepts the chat message corresponding to the output of the aforementioned specified chat message according to the expression request, and outputs the aforementioned retrieval query to the aforementioned retrieval system.

[15] Such as the retrieval control system of any of [1] to

[14] , wherein the retrieval results of the aforementioned retrieval are included in the aforementioned record.

[16] Such as the retrieval control system of any one of [1] to

[15] , wherein the aforementioned retrieval system performs a retrieval with goods or services as the retrieval object.

[17] Such as the retrieval control system of any of [1] to

[16] , wherein the aforementioned retrieval query is a keyword used in the retrieval of goods or services.

[18] Such as the search control system of any of [1] to

[17] , wherein the aforementioned search results represent goods or services.

[19] The retrieval control system of any one of [1] to

[18] further includes the aforementioned terminal. The aforementioned terminal includes a display mechanism that displays a screen showing chat messages and the aforementioned search results arranged in a time sequence.

[0134] 1: Chat System 10: Retrieval and Control System 10a: Processor 10b: Memory Department 10c: Communications Department 12a: First Retrieval System 12b: Second retrieval system 14: Terminal 16: Computer Network 20: Chat screen 22, 22a, 22b, 22c: User message images 24, 24a, 24b: Response message images 26,26a,26b,26c,26d,26e,26f,26g,26h,26i: Option Images 28: Input Form 30: Image of search results 40: Recording Memory Department 42: Chat Message Handling Department 44: Intent Determination Unit 46: Prompt Generation Department 48: Prompt Output Section 50a: First Search Intermediary Department 50b: Second Search Agency 52: General Relationship Department 54: Chat Message Generation Department 56: Display Control Unit 60: Comprehensive Search Agency Department

Claims

1. A retrieval control system comprising: a decision mechanism that repeatedly performs processing to determine a target model from a plurality of language models based on logs of chat messages exchanged with a terminal, wherein the plurality of language models are either language models associated with a retrieval system or language models not associated with a retrieval system; a prompt output mechanism that outputs a prompt generated from a chat message received by the terminal to the target model when the chat message has been received; a chat message output mechanism that outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output mechanism that, based on given conditions satisfied when the target model is a language model associated with a retrieval engine, outputs a retrieval query generated from the logs to a retrieval system associated with the target model; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

2. A retrieval control system, comprising: a decision mechanism that repeatedly performs processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal, wherein the plurality of language models are associated with retrieval systems that are different from the retrieval object; a prompt output mechanism that outputs a prompt generated based on a chat message received by the terminal to the aforementioned target model when the chat message has been received; a chat message output mechanism that outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output mechanism that outputs a retrieval query generated based on the aforementioned records to the retrieval system associated with the aforementioned target model, according to given conditions; and a retrieval result output mechanism that outputs the retrieval results of the retrieval system for the retrieval object received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

3. A retrieval control system comprising: a decision mechanism that repeatedly performs processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal; a prompt output mechanism that outputs a prompt generated based on a chat message received by the terminal to the target model that has received the chat message; a chat message output mechanism that outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output mechanism that outputs a retrieval query generated based on the aforementioned records to the retrieval system; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; wherein all of the plurality of language models can refer to the aforementioned records.

4. A retrieval control system comprising: a decision mechanism that repeatedly performs processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal; a prompt output mechanism that outputs a prompt generated based on a chat message received by the terminal to the target model that has received the chat message; a chat message output mechanism that outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output mechanism that outputs a retrieval query generated based on the aforementioned records to the retrieval system; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; wherein all of the plurality of language models are pre-set to generate responses in a common communication style.

5. The retrieval control system of any one of the requests 1 to 4, wherein the aforementioned decision-making body performs the process of deciding the aforementioned target model whenever it receives a chat message from the aforementioned terminal.

6. The retrieval control system of any one of claims 1 to 4 further includes: a retrieval query generating mechanism that generates a retrieval query generating prompt based on the aforementioned record, outputs the aforementioned retrieval query generating prompt to the aforementioned target model, and generates the aforementioned retrieval query based on the answer output from the target model.

7. A retrieval control system as described in any of requests 1 to 4, wherein the aforementioned records contain the aforementioned retrieval results.

8. A search control system as described in any of the requests 1 to 4, wherein the aforementioned search system performs a search for goods or services.

9. A retrieval control system as requested in any of items 1 to 4, wherein the aforementioned retrieval query is: keywords used for retrieving goods or services.

10. A search control system as described in any of the requests 1 to 4, wherein the aforementioned search results represent goods or services.

11. The retrieval control system of any one of claims 1 to 4 further includes the aforementioned terminal, the aforementioned terminal including: a display mechanism that displays a screen that configures chat messages and the aforementioned retrieval results in a time sequence.

12. A retrieval control system comprising: a decision mechanism that repeatedly performs a process of determining a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are: either templates associated with a retrieval system or templates not associated with a retrieval system; a prompt output mechanism that outputs a prompt generated based on a chat message received from the terminal and the target template at the time the chat message was received to a language model; a chat message output mechanism that outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output mechanism that, based on given conditions satisfied when the target template is a template associated with a retrieval system, outputs a retrieval query generated based on the aforementioned records to the retrieval system associated with the target template; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

13. A retrieval control system comprising: a decision mechanism that repeatedly performs processing to determine a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are associated with retrieval systems that are different from the retrieval object; a prompt output mechanism that outputs a prompt generated based on a chat message received from the terminal and the target template at the time the chat message was received to a language model; a chat message output mechanism that outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output mechanism that outputs a retrieval query generated based on the aforementioned records to a retrieval system associated with the target template, according to given conditions; and a retrieval result output mechanism that outputs the retrieval results of the retrieval system for the retrieval object received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

14. The retrieval control system of request item 12 or 13, wherein the aforementioned decision-making body performs the process of deciding the aforementioned target template whenever it receives a chat message from the aforementioned terminal.

15. The retrieval control system as requested in item 12 or 13, wherein the aforementioned records contain the aforementioned retrieval results.

16. The search control system as requested in item 12 or 13, wherein the aforementioned search system performs a search for goods or services.

17. The retrieval control system as requested in item 12 or 13, wherein the aforementioned retrieval query is: keywords used for retrieving goods or services.

18. The search control system as requested in item 12 or 13, wherein the aforementioned search results represent goods or services.

19. The retrieval control system of request item 12 or 13 further includes the aforementioned terminal, the aforementioned terminal including: a display mechanism that displays a screen in which chat messages and the aforementioned retrieval results are configured in a time sequence.

20. A retrieval control system comprising: a prompt output mechanism that outputs a prompt generated based on a chat message received from a terminal to a language model; a chat message output mechanism that outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output mechanism that outputs a retrieval query generated based on records of chat messages exchanged with the terminal to a retrieval system; and a retrieval result output mechanism that outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; wherein the chat message output mechanism outputs a specified chat message indicating that a retrieval can be performed based on the fulfillment of given conditions, and the retrieval query output mechanism outputs the retrieval query to the retrieval system based on the acceptance of the retrieval request in response to the output of the specified chat message.

21. The retrieval control system of request item 20, wherein the aforementioned retrieval query output mechanism outputs the aforementioned retrieval query to the aforementioned retrieval system based on the acceptance of the chat message indicating a request for the output of the aforementioned specified chat message.

22. The retrieval control system as requested in claim 20, wherein the aforementioned records contain the aforementioned retrieval results.

23. The search control system as described in request item 20, wherein the aforementioned search system performs searches targeting goods or services.

24. The retrieval control system as requested in item 20, wherein the aforementioned retrieval query is: keywords used for retrieving goods or services.

25. The search control system of request item 20, wherein the aforementioned search results represent goods or services.

26. The retrieval control system of claim 20 further includes the aforementioned terminal, the aforementioned terminal including: a display mechanism that displays a screen in which chat messages and the aforementioned retrieval results are configured in a time sequence.

27. A retrieval control method comprising the following steps: a decision-making mechanism repeatedly performing a process to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal, wherein the plurality of language models are: either language models associated with a retrieval system or language models not associated with a retrieval system; a prompt output mechanism outputting a prompt generated from a chat message received by the terminal to the target model that received the chat message; a chat message output mechanism outputting a chat message based on the target model's response to the prompt to the terminal; a retrieval query output mechanism outputting a retrieval query generated from the records to a retrieval system associated with the target model, provided that given conditions are met when the target model is a language model associated with a retrieval engine; and a retrieval result output mechanism outputting retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

28. A retrieval control method comprising the following steps: a decision-making agency repeatedly performs processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal, wherein the plurality of language models are associated with retrieval systems that are different from the retrieval object; a prompt output agency outputs a prompt generated from a chat message received by the terminal to the aforementioned target model when the chat message has been received; a chat message output agency outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output agency outputs a retrieval query generated from the aforementioned records to a retrieval system associated with the aforementioned target model, based on given conditions; and a retrieval result output agency outputs the retrieval results for the retrieval object received by the retrieval system from the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

29. A retrieval control method comprising the following steps: a decision-making agency repeatedly performs processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal; a prompt output agency outputs a prompt generated from a chat message received by the terminal to the target model that received the chat message; a chat message output agency outputs a chat message based on the target model's response to the prompt to the terminal; a retrieval query output agency outputs a retrieval query generated based on the aforementioned records to a retrieval system; and a retrieval result output agency outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; and all of the aforementioned plurality of language models can refer to the aforementioned records.

30. A retrieval control method comprising the following steps: a decision-making agency repeatedly performing processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal; a prompt output agency outputting a prompt generated from a chat message received by the terminal to the target model that received the chat message; a chat message output agency outputting a chat message based on the target model's response to the prompt to the terminal; a retrieval query output agency outputting a retrieval query generated based on the aforementioned records to a retrieval system; and a retrieval result output agency outputting retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; wherein all of the plurality of language models are pre-set to generate responses in a common communication style.

31. A retrieval control method comprising the following steps: a decision-making agency repeatedly performs a process of determining a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are: either templates associated with a retrieval system or templates not associated with a retrieval system; a prompt output agency outputs a prompt generated from a chat message received from the terminal and the aforementioned target template at the time the chat message was received to a language model; a chat message output agency outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output agency outputs a retrieval query generated based on the aforementioned records to the retrieval system associated with the target template, provided that given conditions are met when the target template is a template associated with the retrieval system; and a retrieval result output agency outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

32. A retrieval control method comprising the following steps: a decision-making agency repeatedly performs processing to determine a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are associated with retrieval systems that are different from the retrieval object; a prompt output agency outputs a prompt generated based on the chat message received from the terminal and the aforementioned target template at the time the chat message was received to a language model; a chat message output agency outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output agency outputs a retrieval query generated based on the aforementioned records to the retrieval system associated with the aforementioned target template, according to given conditions; and a retrieval result output agency outputs the retrieval results for the retrieval object from the retrieval system received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal.

33. A retrieval control method comprising the following steps: a prompt output mechanism outputs a prompt generated based on a chat message received from a terminal to a language model; a chat message output mechanism outputs a chat message based on the language model's response to the prompt to the terminal; a retrieval query output mechanism outputs a retrieval query generated based on records of chat messages exchanged with the terminal to a retrieval system; and a retrieval result output mechanism outputs retrieval results received by the retrieval system based on the output of the retrieval query to the retrieval system to the terminal; wherein, in the step of outputting the chat message to the terminal, a chat message indicating that a retrieval can be performed is output according to given conditions; and in the step of outputting the retrieval query to the retrieval system, the retrieval query is output to the retrieval system based on the acceptance of a retrieval request for the output of the aforementioned chat message.

34. A program product that causes a computer to perform the following steps: Based on records of chat messages exchanged with a terminal, repeatedly performing processing to determine a target model from a plurality of language models, wherein the plurality of language models are: either language models associated with a retrieval system or language models not associated with a retrieval system; Outputting a prompt generated based on a chat message received from the terminal to the aforementioned target model when the chat message was received; Outputting a chat message based on the target model's response to the prompt to the terminal; Outputting a search query generated based on the aforementioned records to a retrieval system associated with the target model, provided that given conditions are met when the target model is a language model associated with a retrieval engine; and Outputting search results received by the retrieval system based on the output of the search query to the retrieval system to the terminal.

35. A program product that causes a computer to perform the following steps: Based on records of chat messages exchanged with a terminal, repeatedly performing processing to determine a target model from a plurality of language models, wherein the plurality of language models are: associated with retrieval systems that are different from the retrieval objects; outputting prompts generated based on chat messages received from the terminal to the aforementioned target model when the chat message was received; outputting chat messages based on responses from the aforementioned target model to the aforementioned terminal; outputting search queries generated based on the aforementioned records to the retrieval system associated with the aforementioned target model according to given conditions; and outputting the search results for the retrieval objects from the retrieval system received by the retrieval system based on the output of the aforementioned search query to the aforementioned retrieval system to the aforementioned terminal.

36. A program product that causes a computer to perform the following steps: repeatedly executing processing to determine a target model from a plurality of language models based on records of chat messages exchanged with a terminal; outputting a prompt generated based on a chat message received from the aforementioned terminal to the aforementioned target model when the chat message was received; outputting a chat message based on the target model's response to the aforementioned prompt to the aforementioned terminal; outputting a search query generated based on the aforementioned records to a search system; and outputting search results received by the aforementioned search system based on the output of the aforementioned search query to the aforementioned search system to the aforementioned terminal; wherein all of the aforementioned plurality of language models can refer to the aforementioned records.

37. A program product that causes a computer to perform the following steps: repeatedly executing processing of a target model from a plurality of language models based on records of chat messages exchanged with a terminal; outputting a prompt generated based on a chat message received from the aforementioned terminal to the aforementioned target model when the chat message was received; outputting a chat message based on the response of the aforementioned target model to the aforementioned prompt to the aforementioned terminal; outputting a search query generated based on the aforementioned records to a search system; and outputting search results received by the aforementioned search system based on the output of the aforementioned search query to the aforementioned search system to the aforementioned terminal; wherein all of the aforementioned plurality of language models are pre-set to generate responses in a common communication style.

38. A program product that causes a computer to perform the following steps: repeatedly executing a process to determine a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are: either templates associated with a retrieval system or templates not associated with a retrieval system; outputting a prompt generated based on a chat message received from the terminal and the aforementioned target template when the chat message was received to a language model; outputting a chat message based on the language model's response to the prompt to the terminal; outputting a search query generated based on the aforementioned records to a retrieval system associated with the target template, provided that given conditions are met when the target template is a template associated with a retrieval system; and outputting search results received by the retrieval system based on the output of the search query to the retrieval system to the terminal.

39. A program product that causes a computer to perform the following steps: repeatedly executing a process of determining a target template from a plurality of templates based on records of chat messages exchanged with a terminal, wherein the plurality of templates are associated with search systems that are different from the search target; outputting a prompt generated based on a chat message received from the terminal and the aforementioned target template at the time the chat message was received to a language model; outputting a chat message based on the language model's response to the prompt to the terminal; outputting a search query generated based on the aforementioned records to a search system associated with the aforementioned target template, according to given conditions; and outputting the search results of the search system for the search target, which were received by the search system based on the output of the search query to the search system, to the terminal.

40. A program product that causes a computer to perform the following steps: outputting a prompt generated based on a chat message received from a terminal to a language model; outputting a chat message based on the language model's response to the prompt to the terminal; outputting a search query generated based on records of chat messages exchanged with the terminal to a search system; and outputting search results received by the search system based on the output of the search query to the search system to the terminal; wherein in the step of outputting the chat message to the terminal, a chat message indicating that a search can be performed is output according to given conditions; and in the step of outputting the search query to the search system, the search query is output to the search system based on the acceptance of the search request for the output of the specified chat message.

Images