Information processing method

Abstract

Techniques for selecting queries are improved. A method executed by an information processing device (10) includes: registering a corresponding domain keyword and a target phrase for each candidate of a search object; obtaining a query containing a domain keyword corresponding to a retrieval object determined according to the input question; executing a plurality of search processes including a step of inputting a query into a search engine, a step of specifying and counting each abstract including a target phrase corresponding to a search object on a screen on which a search result is displayed, and a step of adding a keyword included in the specified one or more abstracts to the query; acquiring an answer to the question on the basis of the result of the search in which the count value of the summary including the target phrase is the maximum in the search processing executed multiple times; and outputting the acquired answer.

Description

Technical Field

[0001] This disclosure relates to information processing methods. Background Technology

[0002] Currently, technologies related to dialogue systems that provide answers to user questions are known. For example, Patent Document 1 discloses a technology that generates a domain-specific chatbot using a large-scale language model based on a document provided in that domain.

[0003] Prior art literature

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2023-076413 Summary of the Invention

[0006] The problem that the invention aims to solve

[0007] The aforementioned literature discloses techniques for generating chatbots by constructing large-scale language models using machine learning methods such as autoregressive models. However, there is room for improvement regarding the selection of queries input into search engines in order to output appropriate responses in dialogue systems.

[0008] The purpose of this disclosure, made in view of the above, is to improve the technique for selecting queries for outputting appropriate answers.

[0009] Methods for solving problems

[0010] One embodiment of this disclosure relates to a method executed by an information processing apparatus, comprising:

[0011] For each candidate search target, register the corresponding domain keywords and target phrases;

[0012] Retrieve a query containing the domain keywords corresponding to the search object determined based on the input question;

[0013] Perform multiple search processes, the search processes including: inputting the query into a search engine; identifying and counting summaries containing the target phrase corresponding to the search object on a screen displaying search results; and appending keywords contained in one or more of the identified summaries to the query;

[0014] In the multiple executions of the retrieval process, the answer to the question is obtained based on the retrieval result with the largest count value of the summary containing the target phrase; and

[0015] Output the obtained answer.

[0016] The effects of the invention

[0017] According to one embodiment of this disclosure, the technique for selecting queries for outputting appropriate answers is improved. Attached Figure Description

[0018] Figure 1 This is a block diagram illustrating the general configuration of a system according to one embodiment of this disclosure.

[0019] Figure 2 It is a flowchart illustrating the operation of an information processing device. Detailed Implementation

[0020] The embodiments of this disclosure will now be described.

[0021] (Summary of the implementation method)

[0022] Reference Figure 1 The following is a summary description of System 1, which is included in the embodiments of this disclosure. System 1 includes an information processing device 10, a domain database (domain DB) 20, and a terminal device 30. System 1 is communicatively connected to an external server 40, for example, via a network 50 including the Internet and mobile communication networks. System 1 is, for example, a dialog system that outputs answers to questions such as "Please tell me the appraised value of the used car" asked by car dealers.

[0023] Information processing device 10 is, for example, a computer such as a server. Domain database 20 is a database that stores information related to the input and output of the search engine for each search object. Domain database 20 may be located on a computer such as a server deployed in a cloud environment or a local environment, or it may be located in information processing device 10. Terminal device 30 is, for example, a general-purpose device or a special-purpose device such as a smartphone, mobile phone, wearable device or tablet, navigation device installed in a vehicle, or PC (personal computer), but is not limited to these.

[0024] in addition, Figure 1 The external server 40 shown is a server for the organization providing the LLM (Large-Scale Language Model) 41. The LLM 41 is a language model built through machine learning on large-scale data. The LLM 41 outputs an answer to the question entered by the user. The external server 40 also has a RAG (Retrieval-Augmented Generation) 42. The RAG 42 passes data that the LLM has not learned from, including data from various organizations or data retrieved in real time, to the LLM to assist in the LLM's answer generation. In addition, the RAG 42 can also be incorporated into the information processing apparatus 10.

[0025] First, an overview of this embodiment will be given, with details to follow. In the method executed by the information processing device, for each candidate of the search object, corresponding domain keywords and target phrases are registered, and a query containing domain keywords corresponding to the search object determined based on the input question is obtained. Furthermore, this method performs multiple search processes, each including: inputting a query into a search engine; determining and counting summaries containing target phrases corresponding to the search object on a screen displaying search results; and appending keywords contained in one or more determined summaries to the query. Moreover, in the multiple executions of the search processes, this method obtains an answer to the question based on the search result with the highest count value for summaries containing the target phrase, and outputs the obtained answer.

[0026] In this way, according to this embodiment, since the query is automatically modified and expanded by increasing the number of summaries containing the target phrase on the screen displaying the search results from the search engine, for example, compared to a method that pre-registers multiple appropriate queries after manual screening, the user's burden can be reduced and the accuracy of the answer to the question can be improved. Thus, the technology for selecting appropriate queries is improved by providing highly accurate answers and reducing labor costs.

[0027] Next, the components of System 1 will be described in detail.

[0028] (Composition of information processing device)

[0029] like Figure 1 As shown, the information processing device 10 includes a communication unit 11, a storage unit 12, and a control unit 13.

[0030] The communication unit 11 includes one or more communication interfaces that are respectively connected to the domain DB 20 and the network 50. These communication interfaces may conform to mobile communication standards, wired LAN (Local Area Network) standards, or wireless LAN standards, but are not limited to these; they may conform to any communication standard. In this embodiment, the information processing device 10 communicates with the domain DB 20, the terminal device 30, and the external server 40 via the communication unit 11 and the network 50.

[0031] Storage unit 12 includes one or more memory units. Each memory unit included in storage unit 12 can function as a main storage device, an auxiliary storage device, or a buffer memory, for example. Storage unit 12 stores any information used for the operation of information processing device 10. Storage unit 12 may also store system programs and application programs, for example. In this embodiment, storage unit 12 stores web browser and any search engine application programs, as well as queries entered into the search engine, etc. The information stored in storage unit 12 can be updated, for example, based on information obtained from network 50 via communication unit 11.

[0032] The control unit 13 includes one or more processors, one or more programmable circuits, one or more special-purpose circuits, or combinations thereof. The processor may be a general-purpose processor such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), or a special-purpose processor designed for specific processing, but is not limited thereto. The programmable circuit may be, for example, a FPGA (Field-Programmable Gate Array), but is not limited thereto. The special-purpose circuit may be, for example, an ASIC (Application Specific Integrated Circuit), but is not limited thereto. The control unit 13 controls the overall operation of the information processing device 10.

[0033] (The composition of a domain database)

[0034] Domain DB 20 is a database that stores information related to the input and output of the search engine for each search object. Domain DB 20 stores, for example, search objects, domain keywords, and target phrases. Details regarding the data structure of Domain DB 20 will be described later.

[0035] (Composition of terminal device 30)

[0036] like Figure 1 As shown, the terminal device 30 includes a communication unit 31, a storage unit 32, a control unit 33, an output unit 34, and an input unit 35. The structure of the communication unit 31, the storage unit 32, and the control unit 33 is basically the same as that of the communication unit 11, the storage unit 12, and the control unit 13 of the information processing device 10, so the description is simplified.

[0037] The communication unit 31 includes one or more communication interfaces that are connected to the network 50.

[0038] The storage unit 32 includes one or more memory units. In this embodiment, the storage unit 32 stores applications such as the dialog system provided by the operator using the information processing device 10 and web browser applications.

[0039] The control unit 33 includes one or more processors, one or more programmable circuits, one or more dedicated circuits, or combinations thereof. The control unit 33 is capable of executing application programs stored in the storage unit 32.

[0040] The output unit 34 includes one or more output devices that output information. These output devices may be, for example, a display that outputs image information or a speaker that outputs sound information, but are not limited to these.

[0041] The input unit 35 includes one or more input devices that are operated by the user. These input devices may be, for example, a physical keyboard, a capacitive keyboard, a capacitive panel, a touchscreen integrated with the display, or a microphone that accepts voice input, but are not limited to these.

[0042] (The operation flow of the information processing device)

[0043] Reference Figure 2 The operation of the information processing apparatus 10 according to this embodiment will be explained.

[0044] Step S100: The control unit 13 registers the corresponding domain keywords and target phrases for each candidate of the search object. Specifically, the operator inputs the search object and the domain keywords and target phrases for each search object through the input unit 35 of the terminal device 30, which is then sent to the information processing device 10 via the communication unit 31 through the control unit 33. The information processing device 10's control unit 13 stores the data in the domain DB 20. Here, the data structure of the domain DB 20 is explained with reference to Table 1. The search object is a part or main word (e.g., "appraisal amount") in the question content determined based on the question input by the operator (e.g., "Please tell me the appraisal amount of a used car"). The domain keywords are one or more words (e.g., "used", "appraisal", and "amount") that are used as input to the search engine and are stored in association with the search object. The target phrase is one or more words (e.g., words representing appraisal amount such as "yen", "¥", or "amount range such as "¥*" or "¥*") that are the information desired after searching for the search object.

[0045] [Table 1]

[0046]

[0047] Step S101: The control unit 13 obtains a query containing domain keywords corresponding to the search object determined based on the input question. Specifically, the control unit 13 receives a question (e.g., "Please tell me the appraised value of the used car") input by the operator via the input unit 35 of the terminal device 30 from the terminal device 30 via the communication unit 11, and determines the search object (e.g., "appraised value") based on the received question. The control unit 13 obtains a query from the domain DB 20 containing domain keywords (e.g., "used car appraised value") corresponding to the search object (e.g., "appraised value").

[0048] Step S102: As part of the retrieval process, the control unit 13 inputs the query into the search engine. Specifically, the control unit 13 inputs the query obtained in step S101 (e.g., "second-hand appraisal amount") into the search engine. Here, the search engine can be any search engine such as Google (registered trademark).

[0049] Step S103: As part of the search process, the control unit 13 identifies and counts each summary (hereinafter referred to as a superior summary) containing the target phrase corresponding to the search object on the screen displaying the search results (hereinafter also referred to as the results screen). Specifically, the control unit 13 obtains the results screen through step S102. The control unit 13 obtains the target phrase (e.g., "Japanese Yen ¥") corresponding to the search object (e.g., "appraisal amount") from the domain DB 20. The control unit 13 determines the superior summary (e.g., summary A "A clear overview of used car market prices! Understand the used car market by combining price, mileage, and year.", summary B "The purchase appraisal price at the time of YYYY year MM month is ¥1,000,000~¥1,500,000.") by determining whether each summary contains the target phrase. (For example, summary A does not contain any of "Japanese Yen ¥", therefore it is not a superior summary. Summary B contains "¥", therefore it is a superior summary.) The control unit 13 also counts the number of superior summaries (e.g., summary B). Here, a summary refers to the text information displayed along with the webpage's link or title on the screen showing the results of a search performed after entering words or articles into a search engine, summarizing the content of the webpage. In this embodiment, the results screen may be the first page of the screen displaying the search results. Furthermore, the number of summaries displayed on the results screen can be arbitrarily set.

[0050] Step S104: As part of the retrieval process, the control unit 13 appends keywords (hereinafter also referred to as superior keywords) contained in one or more determined summaries to the query. Specifically, the control unit 13 extracts superior keywords (e.g., "time point of YYYY year MM month") from the superior summaries (e.g., summary B). The control unit 13 appends the superior keywords to the query. As a result, for example, the query becomes "second-hand appraisal amount YYYY year MM month time point". The phrase "time point of YYYY year MM month" can be a specific numerical value (e.g., October 2024) or a wildcard used for fuzzy search (e.g., *year*month). As an implementation of the appending, superior keywords can be one or both of the following: keywords that appear frequently in the summaries containing the target phrase and keywords that appear only in the summaries containing the target phrase. Specifically, the control unit 13 counts the keywords in the superior summaries and decides which keywords to append to the query based on the order of the highest count values ​​(e.g., sequentially or randomly from the highest-ranking keywords). As a further implementation, the control unit 13 may also determine whether the superior keyword is included in each summary that does not include the target phrase, and if it is determined that the superior keyword is not included, the superior keyword is added to the query.

[0051] Step S105: The control unit 13 performs multiple search processes. Specifically, the control unit 13 determines whether the steps S102 to S104, which are part of the search process, have been performed N times. If the control unit 13 determines that N search processes have not been performed, it returns to step S102 (step S105: No) and repeats the search process. If the control unit 13 determines that N search processes have been performed, it proceeds to step S106 (step S105: Yes). Here, N is any integer value greater than or equal to 1, and may also have a predetermined upper limit. As an additional implementation, the control unit 13 may also end the execution of the search process at the time point when the occurrence ratio of the summary containing the target phrase is greater than or equal to a threshold. For example, the control unit 13 may also count the summaries on the results screen, and end the search process when the number of summaries is 10 and the number of excellent summaries in step S103 is 9.

[0052] Step S106: In the multiple executions of the retrieval process, the control unit 13 obtains an answer to the question based on the retrieval result with the highest count value of the summary containing the target phrase (hereinafter also referred to as the best result). Specifically, the control unit 13 determines the best result based on the number of excellent summaries obtained in step S103, based on the results of repeated retrieval processes, and obtains the best result screen. This result screen can be stored in and retrieved from the storage unit 12 in association with the query in step S103, or it can be obtained by re-entering the query used in the best result into the search engine. The control unit 13 sends the best result screen along with the question entered by the operator to the external server 40 in any data form. For example, the control unit 13 can also generate an instruction such as "Please tell me the information provided regarding the valuation amount for the used car," add the information from the best result screen to the instruction, and send it to the external server 40. The LLM 41 of the external server 40 refers to the best received result screen as the RAG 42 and generates an answer to the question (e.g., "The appraised value of the used car is ¥1,000,000 to ¥1,500,000"). The LLM 41 sends the generated answer to the information processing device 10. The control unit 13 of the information processing device 10 receives the answer from the external server 40.

[0053] Step S107: The control unit 13 outputs the acquired response. Specifically, the control unit 13 sends the response received from the external server 40 to the terminal device 30 via the communication unit 11. The control unit 33 of the terminal device 30 outputs the response to the operator via the output unit 34.

[0054] As an additional implementation, the retrieval process can also delete a portion of the domain keywords and added keywords (hereinafter referred to as extended keywords) from the query. Specifically, the control unit 13 deletes a portion of the extended keywords (e.g., "second-hand valuation amount YYYY year MM month time point") from the extended keywords in the query, generating a query containing new extended keywords ("valuation amount YYYY year MM month time point"). The addition to the query in step S104 and the deletion from the query in this embodiment can be performed simultaneously or in any order. Furthermore, the control unit 13 can freely generate combinations of extended keywords by combining the addition and deletion of the query. The control unit 13 can, for example, use the historical records of previous extended keywords, determine the combination of extended keywords based on keyword sorting, or randomly. In addition, for example, a genetic algorithm can be used to determine the combination of multiple extended keywords. Multiple child genes ("second-hand valuation amount" and "market price in ten thousand yen", etc.) can also be generated from multiple parent genes ("second-hand amount in ten thousand yen", "valuation market price", "second-hand market price in ten thousand yen" and "second-hand market price", etc.). As a result, more appropriate query filtering can be performed.

[0055] As an additional implementation, the control unit 13 updates the domain keywords based on the query with the highest count value of the summary containing the target phrase. Specifically, the control unit 13 may also overwrite the domain keywords in the domain DB 20 with the extended keywords in the query entered into the search engine at the best time. For example, if "second-hand valuation amount YYYY year MM month time point" is an extended keyword in the query used at the best time, this extended keyword is overwritten and updated in the domain DB 20 with the domain keywords associated with the search object ("valuation amount"). Thus, appropriate queries can be obtained from the domain DB 20 in real time.

[0056] As an additional implementation, the query can also be configured to include vehicle model names. In this case, the domain DB 20 can be structured according to each vehicle model name. As a further additional implementation, the retrieval processing from step S102 to step S105 can be performed in parallel according to each vehicle model name, and the query with the highest average proportion of extended keywords containing excellent summaries is selected as the best query, and retrieval processing across each vehicle model name is adopted. Thus, appropriate queries can be efficiently selected.

[0057] As an additional implementation, this disclosure (especially step S100) can also be performed by any user. That is, any user of any dialogue system built using System 1 can implement this disclosure. This method can be used more widely.

[0058] While this disclosure has been described with reference to the accompanying drawings and embodiments, it should be noted that those skilled in the art can make various modifications and alterations based on this disclosure. Therefore, it should be understood that these modifications and alterations are all included within the scope of this disclosure. For example, the functions included in each component or step can be reconfigured in a logically consistent manner, multiple components or steps can be combined into one, or individual components or steps can be separated.

[0059] For example, in the above-described embodiments, it is also possible to distribute the configuration and operation of the information processing device 10 among multiple computers capable of communicating with each other. Alternatively, it is possible to include some or all of the components of the information processing device 10 in the terminal device 30. The number of terminal devices 30 included in System 1 can be arbitrarily determined.

[0060] Alternatively, it is possible to implement an embodiment in which a general-purpose computer functions as the information processing apparatus 10 described in the above embodiments. Specifically, a program describing the processing content for implementing the functions of the information processing apparatus 10 described in the above embodiments is stored in the memory of a general-purpose computer, and the program is executed by reading it through a processor. Thus, this disclosure can also be implemented as a processor-executable program or as a non-transitory computer-readable medium storing the program.

[0061] Explanation of reference numerals in the attached figures

[0062] 1 System

[0063] 10. Information processing device

[0064] 20 Domain Databases (DBs)

[0065] 30 Terminal devices

[0066] 40 External Servers

[0067] 50 Networks.

Claims

1. A method, which is performed by an information processing apparatus. The method includes: For each candidate search target, register the corresponding domain keywords and target phrases; Retrieve a query containing the domain keywords corresponding to the search object determined based on the input question; Perform multiple retrieval processes, the retrieval processes including: inputting the query into a search engine; and determining and counting each summary containing the target phrase corresponding to the retrieval object on the screen displaying the retrieval results. And the step of appending the keywords contained in one or more of the identified summaries to the query; In the multiple executions of the retrieval process, the answer to the question is obtained based on the retrieval result of the one with the largest count value of the summary containing the target phrase; as well as Output the obtained answer.

2. The method according to claim 1, further comprising: The domain keywords are updated based on the query that has the highest count value for the summary containing the target phrase.

3. The method according to claim 1, wherein, The keywords added in the retrieval process are one or both of the following: keywords that appear frequently in the summary containing the target phrase, and keywords that appear only in the summary containing the target phrase.

4. The method according to claim 1, wherein, The retrieval process further includes the step of deleting the domain keywords from the query and a portion of any additional keywords.

5. The method according to claim 1, further comprising: The retrieval process ends at the point in time when the occurrence rate of the summary containing the target phrase is above a threshold.

Images