Information processing device, information processing method, and information processing program

Abstract

This information processing device comprises: a first acquisition unit that acquires a response sentence to a user request, the response sentence being generated by a first trained language model; a second acquisition unit that acquires a search result, which is a result of a search performed by inputting a first prompt including a search condition for searching for an official web page providing first information included in the response sentence to a second trained language model; and a verification unit that verifies the first information included in the response sentence by using second information included in the search result.

Description

Information processing apparatus, information processing method and information processing program

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and an information processing program.

[0002] Correction apparatuses that automatically detect input errors in facility names and addresses in map data and correct data errors are known.

[0003] Japanese Unexamined Patent Publication No. 2010-231560

[0004] By the way, when searching for facilities that meet user needs using a large language model (LLM), the address information of the facility presented by the LLM may be incorrect. In the related art, in such a case, it is conceivable to use the WEB to check whether the facility address or the like is correct. However, since the WEB is full of information of unknown authenticity, there are cases where confirmation is performed using incorrect information.

[0005] The present disclosure has been made in view of the above, and an object of the present disclosure is to provide an information processing apparatus, an information processing method, and an information processing program that can be verified with correct information.

[0006] In order to solve the above-described problems and achieve the object, an information processing apparatus according to the present disclosure includes: a first acquisition unit that acquires a response text to a user request generated by a first trained language model; a second acquisition unit that acquires a search result obtained by inputting a first prompt including a search condition for searching an official website providing first information included in the response text into a second trained language model; and a verification unit that verifies the first information included in the response text using second information included in the search result.

[0007] Furthermore, the information processing method relating to this disclosure is an information processing method executed by an information processing device, and is characterized by comprising: a first acquisition step of acquiring a response sentence to a user request generated by a first trained language model; a second acquisition step of inputting a first prompt including search conditions for searching an official homepage that provides first information contained in the response sentence to a second trained language model and acquiring the search results obtained; and a verification step of verifying the first information contained in the response sentence using the second information contained in the search results.

[0008] Furthermore, the information processing program relating to this disclosure is characterized by causing a computer to execute a first acquisition procedure for acquiring a response sentence to a user's request generated by a first trained language model; a second acquisition procedure for inputting a first prompt including search conditions for searching an official homepage that provides the first information contained in the response sentence into a second trained language model and acquiring the search results; and a verification procedure for verifying the first information contained in the response sentence using the second information contained in the search results.

[0009] Figure 1 is a diagram showing an example of the usage of the information processing device. Figure 2 is a diagram showing an example of the application of the information processing device. Figure 3 is a block diagram showing an example of the configuration of the information processing device according to Embodiment 1. Figure 4 is a diagram showing an example of the process of generating a response statement based on a request according to Embodiment 1. Figure 5 is a diagram showing an example of the process flow for searching and verifying based on the information contained in the response statement according to Embodiment 1. Figure 6 is a diagram showing an example of the search process for the official homepage using the search model according to Embodiment 1. Figure 7 is a diagram showing an example of the process of verifying the address of a facility using the determination model according to Embodiment 1. Figure 8 is a diagram showing an example of the process of updating and responding to a response statement according to Embodiment 1. Figure 9 is a flowchart showing an example of the information processing flow of the information processing device according to Embodiment 1. Figure 10 is a diagram showing an example of the search process for a platform using the search model according to Embodiment 2. Figure 11 is a flowchart showing an example of the information processing flow of the information processing device according to Embodiment 2. Figure 12 is a diagram showing an example of variation in address display according to Embodiment 3. Figure 13 is a diagram showing an example of an exception when determining whether addresses are the same or not according to Embodiment 3. Figure 14 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing devices of Embodiment 1, Embodiment 2, Embodiment 3, and Embodiment 4.

[0010] Embodiments of the information processing apparatus, information processing method, and information processing program relating to this disclosure will be described in detail below with reference to the drawings. However, this disclosure is not limited to the embodiments described below. Furthermore, each embodiment can be combined as appropriate within a non-consistent scope.

[0011] [Embodiment 1] Figure 1 is a diagram showing an example of the usage of the information processing device 1. As shown in Figure 1, the information processing device 1 is a computer or the like installed in a vehicle, and is a device that responds to requests from the vehicle user (hereinafter sometimes simply referred to as "user"), who is the driver or passenger. For example, the vehicle user starts up the information processing device 1 installed in the vehicle and makes a request to the information processing device 1, such as searching for a store. The information processing device 1, upon receiving this request, responds to the user's request.

[0012] The information processing device 1 may be a dedicated navigation device built into or mounted in a vehicle. For example, the information processing device 1 may be a single device having a navigation function, a display function, and an interactive function.

[0013] Furthermore, users can substitute their everyday portable devices (e.g., smartphones, tablet devices, notebook PCs, desktop PCs, PDAs, etc.) with the information processing device 1 by installing a predetermined application on them. For example, when a portable device is used as the information processing device 1, it is installed on the vehicle's dashboard or similar location while driving.

[0014] (Examples of Application) By the way, the information processing device 1 is not limited to in-vehicle devices. The information processing device 1 can be used not only in in-vehicle devices but also in various other devices that can interact with users. Here, we will illustrate some examples of applications for the information processing device 1.

[0015] Figure 2 shows an example of how the information processing device 1 is used. For example, as shown in Figure 2(1), the information processing device 1 can be used in a smart speaker. In this case, the user activates a smart speaker that has the same functions as the information processing device 1 and makes a request to the smart speaker, such as searching for a store. The smart speaker that receives this request responds to the user's request.

[0016] Furthermore, as shown in Figure 2(2), for example, the information processing device 1 can be used in a smartphone dialogue application or the like. In this case, the user launches a dialogue application or the like that includes functions equivalent to the information processing device 1 and makes a request to the smartphone, such as searching for a store. The dialogue application or the like that receives this request responds to the user's request. Note that the information processing device 1 is not limited to smartphones; it can also be applied to mobile phones, tablets, and the like.

[0017] Furthermore, as shown in Figure 2(3), for example, the information processing device 1 can be used with a personal computer. In this case, the user launches interactive software or the like that includes functions equivalent to the information processing device 1 and makes a request to the personal computer, such as searching for a store. The interactive software or the like that receives this request responds to the user's request. Note that the information processing device 1 is not limited to personal computers, but can be similarly applied to various computers such as game consoles with internet communication capabilities.

[0018] Incidentally, conventional information processing devices, when they receive a user request and use a trained language model to search for a facility that meets the user's needs, may sometimes have incorrect facility address information provided by the trained language model. Therefore, conventional information processing devices use the web to verify whether the facility address is correct or not, but because the web is full of information of uncertain veracity, situations can arise where the trained language model makes a judgment about whether the facility address information provided by the facility is correct or not based on incorrect information on the web.

[0019] Therefore, this disclosure describes an information processing device 1 that generates a prompt including search conditions to search for the official website from the facility name contained in the response sentence, inputs it into a trained language model, and verifies the search results and the response sentence.

[0020] Specifically, the information processing device 1 in this disclosure acquires a response statement corresponding to a user request generated by a trained response model 61. The information processing device 1 generates a prompt that includes search conditions for searching for the official homepage that provides the facility name, etc., included in the response statement, inputs it into the search model 62, and acquires the search results. The information processing device 1 then verifies the address and other information included in the response statement against the address and other information included in the search results. The information processing device 1 updates the response statement based on the verification results.

[0021] For example, a user requests the information processing device 1 to "find a nearby cafe." The information processing device 1 then inputs the request into the response model 61 and retrieves the response model 61's generated message: "The nearby cafes are as follows: 1. Facility name: Cafe W Address: K1-2-3, Kawagoe City, Saitama Prefecture." The information processing device 1 generates a prompt that includes the search criteria for the official website of the facility name "Cafe W" and inputs it into the search model 62. The search model 62 then outputs the search result: "Official website found. The address of Cafe W is K1-2-13, Kawagoe City, Saitama Prefecture." The information processing device 1 retrieves the outputted search result. Next, the information processing device 1 verifies the address of "Cafe W" included in the response message, "K1-2-3, Kawagoe City, Saitama Prefecture," and the address of "Cafe W" included in the search result, "K1-2-13, Kawagoe City, Saitama Prefecture," and determines that the addresses are different. The information processing device 1 determined that the addresses were different, so it updated the address included in the response message, "K1-2-3, Kawagoe City, Saitama Prefecture," to the search result address, "K1-2-13, Kawagoe City, Saitama Prefecture."

[0022] In this way, the information processing device 1 can verify the address and other information contained in the response message and the address and other information contained in the official website of the search results, and update the response message appropriately, thereby providing accurate information to the user.

[0023] (Configuration of Information Processing Device 1) Next, an example of the functional configuration of the information processing device 1 according to Embodiment 1 will be described. Figure 3 is a block diagram showing an example of the configuration of the information processing device 1 according to Embodiment 1.

[0024] As shown in Figure 3, the information processing device 1 consists of a communication unit 2, an input unit 3, an output unit 4, a control unit 5, a storage unit 6, and the like.

[0025] The communication unit 2, under the control of the control unit 5, connects to the server via the network and sends and receives various types of information. For example, if a trained language model or the like, contained in the storage unit 6 of the information processing device 1, is stored in an external device such as a server, the communication unit 2 sends and receives data via the network.

[0026] The input unit 3 includes a microphone or the like for capturing sound, and generates audio information based on the captured sound. The input unit 3 also includes a display or the like for receiving operation input, and generates input information based on the input characters or other information. For example, under the control of the control unit 5, the input unit 3 generates audio information when it detects a voice request from the user. The input unit 3 then outputs the generated audio information to the control unit 5.

[0027] The output unit 4 outputs information under the control of the control unit 5. For example, the output unit 4 includes an audio output device and outputs the information input from the control unit 5 as audio. The output unit 4 also includes a display device and, in addition to audio output, may display characters, symbols, etc., on an in-vehicle display of the information processing device 1 or on various displays connected to the information processing device 1.

[0028] The storage unit 6 stores various programs executed by the control unit 5 (information processing programs according to this embodiment), as well as data necessary for the control unit 5 to perform processing. As shown in Figure 3, the storage unit 6 includes a response model 61, a search model 62, a determination model 63, and the like.

[0029] The response model 61 is a language model trained to generate and output a response sentence in accordance with an input prompt. For example, when the response model 61 receives a prompt requesting the generation of a response sentence in response to a user's request such as a store search, it outputs a response sentence in accordance with the instructions of the prompt. The response model 61 is included in the first trained language model relating to this disclosure.

[0030] The search model 62 is a language model trained to search for information and output search results in accordance with input prompts. For example, when a prompt is input that includes the name of a facility, search criteria for the official website, and the information to be retrieved, the search model 62 searches the official website of the relevant facility from the internet or other sources and outputs the retrieved information. The search model 62 is included in the second trained language model relating to this disclosure.

[0031] The judgment model 63 is a language model trained to determine whether the input information is correct or incorrect according to the input prompt and output the result of the determination. For example, if a prompt is input that asks the judgment model 63 to determine whether the facility information contained in the response sentence is the same as the facility information contained in the search results, the judgment model 63 will determine this according to the instructions of the prompt and output the result of the determination.

[0032] The response model 61, the search model 62, and the decision model 63 may be the same. For example, Large Language Models (LLMs) can be used for the response model 61, the search model 62, and the decision model 63. Furthermore, the response model 61, the search model 62, and the decision model 63 may be a group of language models containing multiple language models. The configuration and type of the response model 61, the search model 62, and the decision model 63 are not particularly limited, as long as they are machine learning models trained to handle the processing described later.

[0033] The control unit 5 is implemented by a controller such as a CPU (Central Processing Unit) or MPU (Micro Processing Unit) executing various programs stored in the memory unit 6, thereby controlling the operation of the entire information processing device 1. Note that the control unit 5 is not limited to a CPU or MPU; it may also be composed of integrated circuits such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).

[0034] As shown in Figure 3, this control unit 5 includes a generation unit 51, a first acquisition unit 52, a second acquisition unit 53, a verification unit 54, and the like.

[0035] When the generation unit 51 receives a request from the user, such as a store search request, via the input unit 3, it requests the response model 61 to generate a response statement corresponding to the user's request.

[0036] Here, we will explain the generation of a response using Figure 4. Figure 4 is a diagram showing the generation of a response using the response model 61 after receiving a user request.

[0037] As shown in Figure 4(1), the generation unit 51 receives a user's request for store search or the like according to Embodiment 1 and requests the response model 61 to generate a response statement. In the example of Figure 4(1), the generation unit 51 receives "Find a nearby cafe" and generates the prompt "Please find a nearby cafe." and inputs it to the response model 61. Then, as shown in Figure 4(2), the response model 61 outputs the response statement A in Figure 4: "The nearby cafes are as follows: 1. Name: A Kawagoe Store Address: E1-1, Kawagoe City, Saitama Prefecture 2. Name: B Coffee Kawagoe Store Address: F25-1, Kawagoe City, Saitama Prefecture 3. Name: C Coffee Shop Address: G3-2-3, Kawagoe City, Saitama Prefecture

[0038] Returning to Figure 3, the first acquisition unit 52 acquires the response text output by the response model 61. For example, it acquires the response text A in Figure 4, "The nearby cafes are as follows..."

[0039] The second acquisition unit 53 generates a prompt that includes the information contained in the response statement of the response model 61 acquired by the first acquisition unit 52 and the conditions for searching the official website, inputs it to the search model 62, and acquires the search results output from the search model 62. The information contained in the response statement includes the facility name or facility address, etc., and is included in the first information of this disclosure. If the second acquisition unit 53 cannot acquire the official website, for example, because there is no official website, it outputs the response statement output by the response model 61 to the output unit 4 and responds to the user.

[0040] Here, using Figures 5 and 6, we will explain the process of generating a prompt to search for facility information contained in the response, inputting it into the search model 62, and obtaining the search results. Figure 5 is a diagram showing the flow of the process of searching and determining based on the facility name contained in the obtained response, and Figure 6 is a diagram specifically showing the process of searching based on the facility name contained in the obtained response.

[0041] As shown in (1) of FIG. 5, for each of the facility names "A Kawagoe Store", "B Coffee Kawagoe Store", and "C Coffee Shop" included in the obtained response sentence of the response model 61 shown in A of FIG. 5 (same as A in FIG. 4), the second acquisition unit 53 generates a prompt and inputs the prompt to the search model 62. This prompt includes the name of each facility, search conditions for searching an official website which will be described later, and information to be acquired from the official website. This prompt is included in the first prompt of the present disclosure. Note that the output result of the search model 62 is input to the determination model 63 and used to determine whether the information included in the response sentence is correct. Detailed processing using the determination model 63 will be described later.

[0042] Here, with reference to FIG. 6, the prompt input to the search model 62 and the output result of the search model 62 will be specifically described by taking "B Coffee Kawagoe Store" included in the response sentence of the response model 61 as an example. Note that the same processing is also executed for "A Kawagoe Store" and "C Coffee Shop" included in the response sentence.

[0043] As shown in FIG. 6, the second acquisition unit 53 generates a prompt to be input to the search model 62 in order to search the official website for "B Coffee Kawagoe Store" and acquire the address from the official website. For example, as an example of conditions for searching an official website, the second acquisition unit 53 generates a prompt (B in FIG. 6) that includes the following four conditions and instructs to search for a website on the WEB that satisfies at least two or more of the conditions: 1. The facility name or a part thereof is included in the URL or the title of the homepage; 2. Contact information for the facility is described; 3. Information has been published or updated from the official website within the most recent six months; 4. There is a link for online ordering.

[0044] Note that conditions for searching the official website may include "having a privacy policy and terms of use", "having a link to social media", "operation information is official", "clear brand, logo, and store information match", "reliable SSL certificate", "containing a copyright notice", and the like.

[0045] Furthermore, the second acquisition unit 53 includes, in a prompt, a specification of "address" as information to be output from the retrieved official website, and the instruction "If found, please output the address of the facility described on the official website" as information to be output.

[0046] As a result of these processes, the second acquisition unit 53 generates the prompt "Official website of B Coffee Kawagoe Store..." shown in B of FIG. 6, and inputs the prompt to the retrieval model 62.

[0047] Then, the retrieval model 62 to which the prompt is input outputs the retrieval result "The official website has been found..." shown in C of FIG. 6 in accordance with the prompt. As a result, the second acquisition unit 53 acquires the output retrieval result.

[0048] Furthermore, as another example of website retrieval conditions, a point system can be adopted. Specifically, the second acquisition unit 53 assigns points to each condition for retrieving an official website. Then, when the total points of applicable conditions exceed a predetermined threshold, the second acquisition unit 53 generates a prompt for determining the website as an official website and inputs the prompt to the retrieval model 62.

[0049] As a specific example, the second acquisition unit 53 sets conditions for retrieving an official website as follows. 1. The facility name or a part thereof is included in the URL or the title of the website: 30 points 2. Contact information for the facility is described: 20 points 3. Information has been disseminated or updated from the official website within the most recent six months: 20 points 4. There is a link for online ordering: 10 points

[0050] Then, the retrieval model 62 follows the above point rules for applicable conditions, determines that the website is an official website when the total points are equal to or greater than the predetermined threshold of 40 points, and outputs the result as a determination result.

[0051] Returning to Figure 3, the verification unit 54 verifies whether the information contained in the response statement of the response model 61 acquired by the first acquisition unit 52 is correct, based on the information contained in the search results acquired by the second acquisition unit 53. If the verification unit 54 determines that the information contained in the response statement is incorrect, the verification unit 54 corrects the information contained in the response statement to match the information from the search results.

[0052] Here, in addition to Figure 5 mentioned above, Figures 7 and 8 will be used to explain the process of verifying whether the information contained in the response sentence is correct based on the search result information and updating the information contained in the response sentence if necessary. Figure 7 is a diagram that specifically shows one of the verification methods of the verification unit 54, which compares the information in the search result with the information contained in the response sentence and determines whether they are the same or not. Figure 8 is a diagram that shows the process of updating the response sentence based on the determination result and responding to the user.

[0053] As shown in Figure 5(2), an example is described in which the search results of the search model 62 are input to the determination model 63, and the response sentence of the response model 61 is compared with the search results of the search model 62 for verification. Specifically, the verification unit 54 uses the search results acquired by the second acquisition unit 53 to generate a prompt that verifies the information contained in the response sentence and the information contained in the search results. The verification unit 54 then inputs this prompt to the determination model 63 and obtains the determination result. This prompt is included in the fourth prompt of this disclosure. Details of this process will be explained using Figures 7 and 8.

[0054] First, using Figure 7, we will specifically explain the processing of "B Coffee Kawagoe Store" included in the response statement of the response model 61. The verification unit 54 generates a prompt to compare the address information of "B Coffee Kawagoe Store" with the address included in the response statement and the address included in the official website. Then, the verification unit 54 inputs this prompt to the judgment model 63 and obtains the judgment result.

[0055] Specifically, as shown in Figure 7D, the verification unit 54 generates a prompt requesting verification of whether the address "F25-1, Kawagoe City, Saitama Prefecture" of "B Coffee Kawagoe Store" included in the response message is the same as the address "F1733-1, Kawagoe City, Saitama Prefecture" included in the search results on the official website, and inputs this prompt into the judgment model 63.

[0056] Next, as shown in Figure 7E, the judgment model 63 outputs the judgment result, "These two addresses are not the same place." The verification unit 54 obtains this judgment result, determines that the address included in the response is incorrect, and updates the response output by the response model 61.

[0057] Next, the update of the response statement of the response model 61 will be explained. The verification unit 54 updates the address of "B Coffee Kawagoe Store" included in the response statement generated by the response model 61, as shown in Figure 8A (same as Figure 4A), to the address included in the search results. Specifically, the verification unit 54 updates the response statement in Figure 8A, "2. Name: B Coffee Kawagoe Store Address: F25-1, Kawagoe City, Saitama Prefecture," to the search result "2. Name: B Coffee Kawagoe Store Address: F1733-1, Kawagoe City, Saitama Prefecture," which is underlined in Figure 8F.

[0058] The verification unit 54 then sends the updated response message "The nearest cafes are as follows..." shown in Figure 8G to the user via the output unit 4.

[0059] Note that "A Kawagoe Store" and "C Coffee Shop" included in Figure 8 are examples of cases where the judgment result of the judgment model 63 is "identical". In this case, the verification unit 54 does not update the addresses of "A Kawagoe Store" and "C Coffee Shop" included in the response statement A in Figure 8.

[0060] Furthermore, while the verification unit 54 shows an example using the determination model 63, it is also possible to convert text based on a rule base and determine if they are identical by comparing the text.

[0061] Furthermore, the information processing device 1 may process using a facility matching AI. The facility matching AI is a model that includes the processing of the search model 62 and the determination model 63 described above, and is a model that performs the entire process, from prompting instructions to responding to the user via the output unit 4, in one go.

[0062] For example, the prompt for the facility matching AI might say, "Include the generated response text and the search criteria from the official website. Search the official website for the facility address included in the response text. Compare the address in the search results with the address in the response text. If they differ, update the address in the response text to the address in the search results and respond to the user." The facility matching AI then executes the process according to the instructions in the prompt and responds to the user.

[0063] (Processing flow of the information processing device 1) Next, an example of the processing procedure by the information processing device 1 according to Embodiment 1 will be described using Figure 9. Figure 9 is a flowchart showing an example of the information processing flow of the information processing device 1 according to Embodiment 1. Note that each step in the flowchart shown in Figure 9 can be executed in a different order, and additional or omitted processes may be included.

[0064] First, the information processing device 1 receives a user request (S1). Next, the information processing device 1 generates a response statement using the response model 61 (S2). The information processing device 1 obtains the generated response statement (S3). The information processing device 1 searches for the official website providing the facility information (name, address, etc.) included in the response statement using the search model 62, and if it is able to obtain search results (S4; YES), it verifies whether the address etc. included in the response statement is correct based on the address etc. included in the search results (S5). If the verification results show that the address etc. included in the response statement is incorrect (S6; NO), the information processing device 1 corrects the address etc. included in the response statement to the address etc. included in the search results (S7). After that, the information processing device 1 responds to the user with the corrected response statement (S8), and the process ends.

[0065] The information processing device 1 searches the official website providing the facility information (name, address, etc.) included in the response statement using the search model 62. If it cannot obtain search results (S4; NO), or if the verification results show that the address etc. included in the response statement is correct (S6; YES), it responds to the user with the response statement as is (S8) and terminates the process.

[0066] (Effects) The above-described embodiment 1 provides the following effects. The information processing device 1 verifies the address and other information contained in the response message and the address and other information contained in the official website, and updates the response message based on the verification result. As a result, the information processing device 1 can provide the user with correct information, improving user convenience.

[0067] The information processing device 1 searches the official website each time a request such as a store search is made, and can update the response text based on the latest information. This allows the information processing device 1 to provide the user with the most up-to-date information.

[0068] [Embodiment 2] In Embodiment 1, an example was described in which the information processing device 1 verifies the facility information contained in the response statement based on information from the official website. However, there may be cases where the official website cannot be searched, such as when an official website has not been created. Therefore, in Embodiment 2, the processing of the information processing device 1 when the official website cannot be searched will be described.

[0069] Note that a detailed explanation of the configuration and processing of the information processing device 1 according to Embodiment 2 will be omitted, as it is similar to the information processing device 1 according to Embodiment 1. The following will describe the processing content that differs from that of Embodiment 1. In this Embodiment 2, the second acquisition unit 53, which differs from that of Embodiment 1, will be described in detail.

[0070] For example, if the second acquisition unit 53 is unable to find the official website containing the information obtained by the first acquisition unit 52, it generates a prompt that includes conditions for searching a platform where the user can post facility ratings and reviews, inputs it into the search model 62, and obtains the search results output from the search model 62. If the second acquisition unit 53 is unable to obtain information from the platform, it outputs the response message output by the response model 61 to the output unit 4 and responds to the user. This prompt is included in the second prompt of this disclosure.

[0071] Here, using Figure 10, we will explain the process of generating a prompt, inputting it into the search model 62, and obtaining search results. Figure 10 is a diagram that specifically shows the process when the official website cannot be found based on the facility name included in the obtained response text.

[0072] First, the second acquisition unit 53 generates a "prompt instructing a search for the official website, which includes conditions for searching the official website," as shown in Figure 10 (H). Then, as shown in Figure 10 (1), the second acquisition unit 53 inputs the generated prompt (Figure 10 H) into the search model 62.

[0073] Then, as shown in (2) of Figure 10, the search model 62 outputs the search results. The second acquisition unit 53 acquires the search result shown in I of Figure 10, "Could not find the official homepage of B Coffee Kawagoe store."

[0074] Here, the second acquisition unit 53 generates a prompt to retrieve information from the platform because it could not find the official website.

[0075] For example, the second acquisition unit 53 generates a prompt (J in Figure 10) that instructs the system to search for information on B Coffee Kawagoe store from a platform that meets at least two of the following three conditions, as an example of conditions for searching a platform: 1. More than 40 million users are registered. 2. More than 450,000 stores are listed. 3. More than 1 million user reviews have been posted.

[0076] Furthermore, the second acquisition unit 53 includes "address" as the information to be output from the searched platform, and the prompt "If found, please output the address of the facility listed on the platform" as the information to be output. Specifically, as shown in (3) of Figure 10, the second acquisition unit 53 inputs the generated prompt (J in Figure 10) into the search model 62.

[0077] Then, as shown in (4) of Figure 10, the search model 62 outputs the search results. The second acquisition unit 53 acquires the search result shown in K of Figure 10, "According to "Tabelog" (trademark registered), the address of B Coffee Kawagoe store is as follows: F1733-1, Kawagoe City, Saitama Prefecture." Note that "Tabelog" is an example of a platform that matches the conditions from the search model 62.

[0078] As another example, the second acquisition unit 53 may also generate a prompt that includes both homepage search conditions and platform search conditions. This prompt includes a statement that it will search for official homepages that meet the conditions, and if no such search is found, it will output information from the prompts that do meet the conditions. This prompt is included in the third prompt of this disclosure.

[0079] In this alternative example, the second acquisition unit 53 generates a prompt that reads, "Search the official website of B Coffee Kawagoe store based on the following conditions, finding one that meets at least two of the conditions. If found, output the address of the facility as listed on the official website. If not found, search from a platform that meets at least two of the following conditions. If found, output the address of the facility..." and inputs it into the search model 62.

[0080] (Processing flow of the information processing device 1) Next, an example of the processing procedure by the information processing device 1 according to Embodiment 2 will be described using Figure 11. Figure 11 is a flowchart showing an example of the information processing flow of the information processing device 1 according to Embodiment 2. Note that each step in the flowchart shown in Figure 11 can be executed in a different order, and additional or omitted processes may be included.

[0081] First, the information processing device 1 receives a user request (S1). Next, the information processing device 1 generates a response statement using the response model 61 (S2). The information processing device 1 obtains the generated response statement (S3). The information processing device 1 searches for the official website providing the facility information (name, address, etc.) included in the response statement using the search model 62, and if it is able to obtain search results (S4; YES), it verifies whether the address etc. included in the response statement is correct based on the address etc. included in the search results (S6).

[0082] Furthermore, if the information processing device 1 searches for the official website providing the facility information (name, address, etc.) included in the response using the search model 62 and cannot obtain search results (S4; NO), it searches for the platform providing the facility information (name, address, etc.) included in the response using the search model 62 (S5). If it can obtain search results by searching for the platform providing the facility information (name, address, etc.) included in the response using the search model 62 (S5; YES), it verifies the address etc. included in the response with the address etc. included in the search results (S6).

[0083] Next, if the information processing device 1 finds that the address or other information included in the response is incorrect as a result of the verification (S7; NO), it corrects the address or other information included in the response to the address or other information included in the search results (S8). After that, the information processing device 1 responds to the user with the corrected response (S9) and terminates the process.

[0084] On the other hand, if the information processing device 1 searches for the platform that provides facility information (name, address, etc.) included in the response statement using the search model 62 and cannot obtain a search result (S5; NO), or if the verification result is that the address etc. included in the response statement is correct (S7; YES), it responds to the user with the response statement as is (S9) and terminates the process.

[0085] According to the embodiment 2 described above, the information processing device 1 can acquire information from the platform even in cases where a small business or other entity does not have an official website. This improves the accuracy of the information provided to the user.

[0086] [Embodiment 3] The information processing device 1 of the present disclosure may generate a prompt after performing a text conversion on the information before it is input to the determination prompt, or it may include an instruction to consider the information as identical in the prompt. This can improve the accuracy of the determination.

[0087] Therefore, Embodiment 3 describes a process in which text conversion is performed in advance and an instruction to consider them identical is included in the prompt. Furthermore, the verification unit 54 of the information processing device 1 according to Embodiment 3, which differs from that of Embodiment 1, will be described in detail.

[0088] For example, the verification unit 54 performs text conversion based on rules, using the information contained in the response and the information contained in the search results obtained from the search model 62. Specifically, the verification unit 54 performs text conversion on the address based on the rules shown in Figure 12.

[0089] For example, in the case of "full-width and half-width characters" as shown in Figure 12 (1), the verification unit 54 performs a process to align the alphabet, numbers, etc. in the address to either full-width or half-width characters. Also, in the case of "presence or absence of spaces" as shown in 12 (2), the verification unit 54 performs a process to remove spaces in the address and align them.

[0090] In the case of "presence or absence of building designation" as shown in (3) of Figure 12, the verification unit 54 performs a process to delete and remove the building name from the address. For example, the verification unit 54 converts "AB Town 10-10A Building 105" to "AB Town 10-10-105".

[0091] In the case of the "building floor notation" shown in (4) of Figure 12, the verification unit 54 performs a process to align the floor notation of the address from "floor" to "F". For example, the verification unit 54 converts "AB town 10-10-2 floor" to "AB town 10-10-2F".

[0092] In the case of "presence or absence of address notation" as shown in (5) of Figure 12, the verification unit 54 performs a process to remove the address notation and align the addresses. For example, the verification unit 54 converts "2-5" and "2-5" to "2-5".

[0093] As described above, the verification unit 54 converts the address based on the rules shown above, generates a prompt based on the converted address, and inputs it into the judgment model 63.

[0094] As an alternative, the verification unit 54 can consider items to be identical during comparison by including an instruction to consider them the same in the prompt. For example, the verification unit 54 generates a prompt that includes a rule for addresses as shown in Figure 13.

[0095] Here, we will explain the following points from "1. Optional Notation" shown in Figure 13: "(1-1) The difference between the presence or absence of Oaza and Koaza is considered the same." and "(1-2) Those that cover the city / ward / town / village name with a postal code are considered the same."

[0096] First, let's explain the case where the prompt includes "(1-1) The difference between the presence or absence of Oaza and Koaza is considered the same," as shown in Figure 13. For example, the judgment model 63 compares "Pattern 1. Saitama Prefecture Kawagoe City F1733-1" output from the response model 61 with "Pattern 2. Saitama Prefecture Kawagoe City Oaza F1733-1" output from the search model 62 and identifies the difference that Pattern 1 does not have "Oaza" and Pattern 2 does. Here, the judgment model 63 considers the difference between the presence or absence of Oaza to be the same according to the prompt shown in (1-1) above, so it considers Pattern 1 and Pattern 2 to be the same.

[0097] Next, we will explain the case where the prompt includes "(1-2) Addresses that cover a city / ward / town / village name with a postal code are considered the same." For example, the determination model 63 compares "Pattern 1. 〒350-0822 F1733-1" output from the response model 61 with "Pattern 2. Saitama Prefecture Kawagoe City F1733-1" output from the search model 62 and determines that Pattern 1 is a combination of a postal code and street address, etc., and Pattern 2 is a combination of a city / ward / town / village name and street address, etc. Subsequently, the determination model 63 searches the postal code database for 〒350-0822 in Pattern 1 and identifies it as "Saitama Prefecture Kawagoe City". As a result, the determination model 63 determines that the address in Pattern 1 is identical to "Saitama Prefecture Kawagoe City F1733-1", and considers Pattern 1 and Pattern 2 to be the same.

[0098] As shown in Figure 13 under "2. Differences in Notation," there are items such as "(2-1) The difference between Chinese numerals and Arabic numerals is considered the same," "(2-2) The difference between "‐ (hyphen)" and "ノ" in street numbers is considered the same," "(2-3) Variations in address notation are considered the same," and "(2-4) The difference between English and Katakana is considered the same." From these, we will specifically explain examples of "(2-1) The difference between Chinese numerals and Arabic numerals is considered the same," and "(2-2) The difference between "‐ (hyphen)" and "ノ" in street numbers is considered the same."

[0099] Furthermore, "(2-3) Variations in address notation are considered the same." may refer to the following concepts included in Figure 12 above: (1) full-width and half-width characters, (2) presence or absence of spaces, (3) presence or absence of building notation, (4) presence or absence of building floor notation, and (5) presence or absence of street number notation.

[0100] Let's explain the case where the prompt includes "(2-1) Consider the difference between Chinese numerals and Arabic numerals to be the same," as shown in Figure 13. For example, the judgment model 63 determines that, based on "Pattern 1. Saitama Prefecture, Kawagoe City, F25-1" output from the response model 61 and "Pattern 2. Saitama Prefecture, Kawagoe City, F25-1" output from the search model 62, Pattern 1 uses Arabic numerals for the address, etc., and Pattern 2 uses Chinese numerals for the address, etc. Here, the judgment model 63 considers the difference between Arabic numerals and Chinese numerals to be the same, in accordance with the instructions of the prompt shown in (2-1) above, and therefore considers Pattern 1 and Pattern 2 to be the same.

[0101] Let's explain the case where the prompt includes "(2-2) The difference between "-" (hyphen) and "ノ" in the address is considered the same," as shown in Figure 13. For example, the judgment model 63 determines that for "Pattern 1. Saitama Prefecture, Kawagoe City, F25-1" output from the response model 61 and "Pattern 2. Saitama Prefecture, Kawagoe City, F25ノ1" output from the search model 62, Pattern 1 has "-" between the numbers in the address, etc., and Pattern 2 has "ノ" between the numbers in the address, etc. Here, the judgment model 63 considers the difference between "-" and "ノ" to be the same, in accordance with the instructions of the prompt shown in (2-3) above, and therefore considers Pattern 1 and Pattern 2 to be the same.

[0102] As described above, the information processing device 1 either prepares the text notation in advance and inputs it to the prompt, or inputs a definition to be considered identical to the prompt. This improves the accuracy of determining whether or not they are identical.

[0103] [Embodiment 4] Now, embodiments of the present disclosure have been described, but the present disclosure may be implemented in various other forms besides those described above.

[0104] (Numerical values, etc.) The numerical values ​​and threshold values ​​(for example, any numerical value) used in the above embodiment are merely examples and can be changed as desired.

[0105] [Hardware Configuration] Furthermore, the information processing device 1 according to Embodiments 1 to 4 described above is realized by a computer 1000 having the configuration shown in Figure 14, for example. The following will be explained using the information processing device 1 as an example. Figure 14 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device 1 of Embodiments 1 to 4. The computer 1000 has a CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I / F) 1500, input / output interface (I / F) 1600, and media interface (I / F) 1700.

[0106] The CPU 1100 operates based on programs stored in the ROM 1300 or HDD 1400, and controls various parts. The ROM 1300 stores boot programs executed by the CPU 1100 when the computer 1000 starts up, as well as programs that depend on the computer 1000's hardware.

[0107] The HDD 1400 stores programs executed by the CPU 1100, and data used by such programs. The communication interface 1500 receives data from other devices via a predetermined communication network and sends it to the CPU 1100, and transmits data generated by the CPU 1100 to other devices via the predetermined communication network.

[0108] The CPU 1100 controls output devices such as displays and printers, and input devices such as keyboards and mice, via the input / output interface 1600. The CPU 1100 acquires data from input devices via the input / output interface 1600. The CPU 1100 also outputs the generated data to output devices via the input / output interface 1600.

[0109] The media interface 1700 reads a program or data stored in the recording medium 1800 and provides it to the CPU 1100 via the RAM 1200. The CPU 1100 loads the program from the recording medium 1800 onto the RAM 1200 via the media interface 1700 and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase Change Rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

[0110] For example, when the computer 1000 functions as an information processing device 1 according to Embodiments 1 to 6, the CPU 1100 of the computer 1000 realizes the functions included in the control unit 5 of the information processing device 1 by executing a program loaded on the RAM 1200. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, these programs may be obtained from another device via a predetermined communication network.

[0111] [Other] Furthermore, among the processes described in each of the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically by known methods. In addition, the processing procedures, specific names, and information including various data and parameters shown in the above document and drawings can be changed at will unless otherwise specified. For example, the various information shown in each figure is not limited to the information shown.

[0112] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions.

[0113] Furthermore, the above embodiments can be combined as appropriate, provided that the processing content is not contradictory.

[0114] Although some embodiments of the present invention have been described in detail above with reference to the drawings, these are illustrative examples, and the present disclosure can be implemented in various other forms with modifications and improvements based on the knowledge of those skilled in the art, starting with the embodiments described in the disclosure section.

[0115] Furthermore, the terms "section, module, unit" mentioned above can be replaced with "means" or "circuits." For example, a generation unit can be replaced with a generation means or a generation circuit.

[0116] 1. Information processing device 2. Communication unit 3. Input unit 4. Output unit 5. Control unit 51. Generation unit 52. First acquisition unit 53. Second acquisition unit 54. Verification unit 6. Storage unit 61. Response model 62. Search model 63. Judgment model

Claims

1. An information processing device comprising: a first acquisition unit that acquires a response sentence to a user request generated by a first trained language model; a second acquisition unit that inputs a first prompt including search conditions for searching an official homepage that provides first information contained in the response sentence to a second trained language model and acquires the search results; and a verification unit that verifies the first information contained in the response sentence using the second information contained in the search results.

2. The information processing apparatus according to claim 1, characterized in that, if the second acquisition unit fails to acquire the search results from the second trained language model, it generates a second prompt including search conditions for searching for a platform that provides the first information, and inputs the generated second prompt to the second trained language model to acquire the retrieved search results.

3. The information processing apparatus according to claim 1, characterized in that the second acquisition unit generates a third prompt including search conditions for searching the official homepage and the platform providing the first information, and inputs the generated third prompt into the second trained language model to acquire the searched results.

4. The information processing apparatus according to claim 1, characterized in that the verification unit determines whether the first information is correct or incorrect based on the second information, and corrects the first information to the second information if it determines that the first information is incorrect.

5. The information processing device according to claim 1, characterized in that the verification unit generates a fourth prompt for determining whether the first information is correct based on the second information, and inputs the generated fourth prompt to a third trained language model to verify the first information.

6. The information processing apparatus according to claim 1, characterized in that the first acquisition unit acquires the name of the first facility and the address of the first facility as first information; the second acquisition unit acquires the name of the second facility and the address of the second facility from the official website acquired as the search result; and the verification unit verifies the first information based on whether the name of the first facility and the name of the second facility refer to the same facility, and whether the address of the first facility and the address of the second facility refer to the same facility.

7. An information processing method performed by an information processing device, comprising: a first acquisition step of acquiring a response sentence to a user request generated by a first trained language model; a second acquisition step of inputting a first prompt including search conditions for searching an official homepage that provides first information contained in the response sentence to a second trained language model and acquiring the search results; and a verification step of verifying the first information contained in the response sentence using the second information contained in the search results.

8. An information processing program characterized by causing a computer to execute: a first acquisition procedure for acquiring a response sentence to a user request generated by a first trained language model; a second acquisition procedure for inputting a first prompt including search conditions for searching an official homepage that provides the first information contained in the response sentence into a second trained language model and acquiring the search results; and a verification procedure for verifying the first information contained in the response sentence using the second information contained in the search results.

Images