Dialogue system

Abstract

To improve accuracy of responses of a dialogue system.SOLUTION: A dialogue system comprises: designating means for designating a database used for searching from multiple databases based on user's questions; creation means for creating a prompt inputted to a creation AI based on a search result searched by using the designated database and the questions; and output means for outputting the creation AI for the prompt as responses to the questions.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the technical field of dialogue systems. [Background technology]

[0002] As a technology used in this type of system, for example, a method has been proposed that includes inputting an initial input determined based on input data from a user into a deep learning model, and determining a second input to be input into the deep learning model based on the initial input and a first intermediate result (see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-182707 Summary of the Invention [Problem to be solved by the invention]

[0004] When generative AI (artificial intelligence) is used in dialogue systems, hallucination, in which the generative AI outputs content that is inconsistent with the facts or unrelated to the context, is an issue. To address this issue, a technology called RAG (Retrieval Augmented Generation) has been proposed, which combines generative AI with a search system to generate answers that reflect specialized knowledge and the latest knowledge. However, even if a product has the same name, its specifications may differ depending on, for example, the year of manufacture. For this reason, even when RAG is used, there is a technical problem in that answers to inquiries about the product may be inaccurate.

[0005] The present invention has been made in consideration of the above problems, and an object of the present invention is to provide a dialogue system that can improve the accuracy of answers. [Means for solving the problem]

[0006] A dialogue system according to one embodiment of the present invention comprises a designation means for designating a database to be used for a search from among a plurality of databases based on a user's question, a generation means for generating a prompt to be input to a generation AI based on the search results of the search using the designated database and the question, and an output means for outputting the output of the generation AI in response to the prompt as an answer to the question. [Brief explanation of the drawings]

[0007] [Figure 1] 1 is a block diagram showing a configuration of a dialogue system according to a first embodiment. [Figure 2] 4 is a flowchart showing the operation of the dialogue system according to the first embodiment. [Figure 3] FIG. 10 is a block diagram showing the configuration of a dialogue system according to a second embodiment. [Figure 4] 10 is a flowchart showing the operation of the dialogue system according to the second embodiment. [Figure 5] FIG. 10 is a block diagram showing the configuration of a dialogue system according to a third embodiment. [Figure 6] 10 is a flowchart showing the operation of the dialogue system according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] First Embodiment A first embodiment of the dialogue system will be described with reference to FIGS. 1 and 2. In FIG. 1, the dialogue system 1 includes an input device 10, an information processing device 20, a storage device 30, an output device 40, and a generated AI 50. The input device 10 is a device that can be operated by a user, such as a keyboard, a mouse, a touch panel, or a microphone. The information processing device 20 may include a processor, such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The information processing device 20 may further include memory, such as a RAM (Random Access Memory) or a ROM (Read Only Memory). The storage device 30 may include a recording medium, such as a hard disk drive or an SSD (Solid State Drive). The output device 40 is a device, such as a display or a speaker, that can output information to the outside of the information processing device 20.

[0009] The generated AI 50 may be configured to be available via a network such as the Internet. That is, the generated AI 50 may be provided as a cloud service. The generated AI 50 may be configured to be executable on a device that can be directly operated by a user, such as a personal computer, after being downloaded to the device.

[0010] The information processing device 20 may be, for example, a personal computer. In this case, the information processing device 20 and the storage device 30 may be housed in the same housing. However, the storage device 30 may be a device connected to the information processing device 20 via a network, such as a network drive. The input device 10, the information processing device 20, the storage device 30, and the output device 40 may be housed in the same housing, such as in a notebook-sized personal computer. The information processing device 20 may be implemented by a server device (e.g., a cloud server). In this case, the input device 10 and the output device 40 may be implemented by a terminal device (e.g., a personal computer, a tablet terminal, a smartphone, etc.) capable of communicating with the information processing device 20.

[0011] The information processing device 20 has a question receiving function 21, a first document DB (DataBase) switching function 22, a document DB 23, a prompt inquiry function 24, and an answer output function 25. The storage device 30 has a plurality of document DBs (e.g., document DB1, document DB2, document DB3, ...). The storage device 30 further has a switching setting file 31 and a document DB information file 32.

[0012] The following is a specific description of a case where the dialogue system 1 answers a user's question about an automobile. The question reception function 21 of the information processing device 20 receives the user's question input via the input device 10 as text data. When the user inputs a question via a microphone, which is an example of the input device 10 (i.e., in the case of voice input), the question reception function 21 may generate text data related to the question based on the voice data related to the question. The question reception function 21 outputs the text data to the first document DB switching function 22.

[0013] The first document DB switching function 22 detects keywords indicating, for example, vehicle type, drive system, etc. from the text data. The first document DB switching function 22 identifies regular expressions for the detected keywords by referring to the switching setting file 31 in the storage device 30. A regular expression is one of the expression methods used for pattern matching of character strings. The switching setting file 31 will be described with reference to Table 1. As shown in Table 1, the switching setting file 31 defines the correspondence between regular expressions and synonyms. For example, the regular expression for "Prius (registered trademark)" and "Prius (registered trademark)" is "vehicle type T1." For example, the regular expression for "HEV," "HV," and "hybrid" is "drive system V1."

[0014] [Table 1]

[0015] For example, a user's question may be, "What is the model of the battery in the smart key for a 2018 Prius PHEV?" In this case, the first document DB switching function 22 may detect "Prius" and "PHEV" as keywords from the question as text data. The first document DB switching function 22 may refer to the switching setting file 31 and identify "vehicle model T1," which is a regular expression for "Prius," and "drive system V2," which is a regular expression for "PHEV." Note that "2018 model" is a regular expression related to the model year.

[0016] The first document DB switching function 22 identifies a document DB to be used for search from multiple document DBs based on a regular expression and the document DB information file 32 in the storage device 30. The document DB information file 32 will be described with reference to Table 2. As shown in Table 2, the document DB information file 32 defines the correspondence between a combination of vehicle model, drive system, and model year, a document, and a document DB related to the document. Here, product manuals and product catalogs are given as examples of documents. Note that the documents are not limited to product manuals and product catalogs, and may be, for example, a main specification table, a function operation guide, a multimedia instruction manual, a maintenance procedure manual, etc. The document list in the document DB file 32 may register not only file names indicating electronic data related to the documents, but also uniform resource locators (URLs) where the documents are published.

[0017] [Table 2]

[0018] For example, if the user's question is, "What is the battery model of the smart key for a 2018 Prius PHEV?", the first document DB switching function 22 may identify document DB3 as the DB to use for search based on the regular expressions "vehicle model T1," "drive system V2," and "2018 model" and the document DB information file 32. The first document DB switching function 22 outputs information indicating the identified document DB (e.g., document DB3) to the document DB search function 23. Note that the first document DB switching function 22 outputting the identified document DB to the document DB search function 23 can be considered as the first document DB switching function 22 specifying the document DB to be searched by the document DB search function 23.

[0019] The document DB search function 23 refers to the user's question and searches the document DB identified by the first document DB switching function 22. The document DB search function 23 may perform a vector search. Each of the multiple document DBs (e.g., document DB1, document DB2, document DB3, ...) may include a vector index so that the document DB search function 23 can perform the vector search quickly. In this case, documents (e.g., product manuals, product catalogs, etc.) may be vectorized. Vector data may be generated by vectorizing the documents. The vector index may refer to a mechanism for efficiently searching vectorized documents (i.e., vector data) included in each document DB. Note that various existing methods can be applied to vectorizing documents, and detailed description thereof will be omitted. For example, the document DB search function 23 may vectorize the user's question (i.e., document). The document DB search function 23 may calculate the cosine similarity between the vectorized question and the vectorized document included in the document DB identified by the first document DB switching function 22. The document DB search function 23 may search for documents (in other words, information) related to the user's question based on the cosine similarity, and outputs the search results to the prompt query function 24.

[0020] The prompt inquiry function 24 creates a prompt to be input to the generation AI 50 based on the search results from the document DB search function 23 and the user's question. An example of a prompt created by the prompt inquiry function 24 will be specifically described. The prompt may include an instruction, prerequisite knowledge, and a question. For example, the instruction may include the role of the generation AI 50 and the processing content. The prerequisite knowledge may include the search results from the document DB search function 23. The question is a user's question. The user's question may be, "What model battery does the smart key for a 2018 Prius PHEV have?" The search results from the document DB search function 23 may be, "What to do when the Prius smart key stops working," "Replacing the smart key battery (lithium battery CR2032)," and "How to unlock the doors with the smart key." In this case, the prompt inquiry function 24 may create a prompt such as the following:

[0021] Instructions: You are an excellent sales staff member at a Toyota dealership. Please create an answer to the question below based on the prerequisite knowledge. Prerequisite knowledge: Search result 1: What to do when your Prius smart key stops working Search result 2: Smart key battery replacement (Lithium battery CR2032) Search result 3: How to unlock the door with a smart key Question: What model smart key battery does the 2018 Prius PHEV have? In the above instruction, "You are an excellent sales staff member at a Toyota dealership" is an example of the role of the generation AI 50. Also, "Please create an answer to the question: based on the following prerequisite knowledge:" is an example of processing content.

[0022] The prompt inquiry function 24 may acquire the answer of the generated AI 50 to the prompt (in other words, the output of the generated AI). The prompt inquiry function 24 may output the answer of the generated AI 50 to the answer output function 25. Note that the answer output function 25 may acquire the answer of the generated AI 50 to the prompt instead of the prompt inquiry function 24. The answer output function 25 transmits the answer of the generated AI 50 to the output device 40. As a result, the output device 40 presents the answer of the generated AI 50 to the user as the answer to the user's question.

[0023] The operation of the dialogue system 1 will be further described with reference to the flowchart of FIG. 2. In FIG. 2, the question receiving function 21 of the information processing device 20 performs a question receiving process to receive a user's question as text data (step S101). The first document DB switching function 22 of the information processing device 20 performs a document DB switching process to specify a document DB to be used for search from multiple document DBs based on the text data (i.e., the user's question) received in the process of step S101 (step S102). The document DB search function 23 of the information processing device 20 performs a document DB search process to search the document DB specified in the process of step S102 (step S103). The prompt inquiry function 24 of the information processing device 20 performs a prompt inquiry process to create a prompt to be input to the generation AI 50 based on the result of the process of step S103 (i.e., the search result of the document DB search function 23) and the user's question (step S104). The answer output function 25 of the information processing device 20 performs an answer output process to transmit the answer of the generation AI 50 to the prompt to the output device 40 (step S105).

[0024] (Technical Effects) For example, in response to the question, "What model battery does your Prius smart key have?", the correct answer for a 2018 Prius is "CR2032," but the correct answer for a 2023 Prius is "CR2450." Thus, even for the same question, different products may produce different answers. Even in a system that uses RAG, it is difficult for the system to correctly answer a question unless an appropriate information source is searched. In contrast, in the dialogue system 1 according to this embodiment, the first document DB switching function 22 specifies a document DB to be used for search from multiple document DBs. Then, the document DB search function 23 searches the specified document DB. In this way, the dialogue system 1 searches for a document DB (i.e., an information source) appropriate for the user's question. Therefore, the dialogue system 1 can improve the accuracy of answers to user questions.

[0025] (Variation) A user's question may not include at least one of "drive system" and "year of model." Therefore, initial values ​​for "drive system" and "year of model" may be set in advance. For example, the initial value for "drive system" may be the drive system with the highest sales volume. For example, the initial value for "year of model" may be the year of the latest model. With this configuration, even if a user's question does not include at least one of "drive system" and "year of model," the first document DB switching function 22 can specify the document DB to be used for search.

[0026] Second Embodiment A second embodiment of the dialogue system will be described with reference to Figures 3 and 4. The dialogue system 2 according to the second embodiment is similar to the dialogue system 1 according to the first embodiment, except for a portion of the configuration of the information processing device 20. Therefore, for the second embodiment, explanations that overlap with the explanation for the first embodiment will be omitted as appropriate. Furthermore, in the drawings, parts that are common to the first embodiment are denoted by the same reference numerals.

[0027] 3, the dialogue system 2 includes an input device 10, an information processing device 20, a storage device 30, an output device 40, and a generation AI 50. The information processing device 20 includes a question receiving function 21, a first document DB switching function 22, a document DB 23, a prompt inquiry function 24, an answer output function 25, and a second document DB switching function 26.

[0028] For example, as shown in Table 2, the document DB information file 32 in the storage device 30 may define the correspondence between a combination of vehicle model, drive system, and model year, a document, and a document DB related to the document. If the user's question does not include at least one of vehicle model, drive system, and model year, the first document DB switching function 22 may have difficulty identifying the document DB to use for the search.

[0029] If the user's question lacks information to identify the document DB to be used for the search (for example, at least one of the vehicle model, drive system, and model year), the second document DB switching function 26 may ask the user for confirmation. "Asking the user for confirmation" may be rephrased as "requesting the user for additional information."

[0030] For example, if the user's question is "What is the model of the battery in the Prius smart key?", the drive system and model year are unknown. In other words, in this case, the "drive system" and "model year" as information for identifying the document DB to be used for the search are unknown. In this case, the second document DB switching function 26 may ask, "Is the drive system an HEV or a PHEV?" and "What is the model year?"

[0031] When the user responds to the second document DB switching function 26's request for confirmation, the first document DB switching function 22 may identify a document DB to be used for the search based on the user's question and the user's response (i.e., additional information). Specifically, the first document DB switching function 22 may detect keywords from each of the user's question and the user's response. The first document DB switching function 22 may identify a regular expression for the detected keyword by referring to the switching setting file 31 in the storage device 30. The first document DB switching function 22 may identify a document DB to be used for the search from multiple document DBs based on the regular expression and the document DB information file 32 in the storage device 30.

[0032] The operation of the dialogue system 2 will be further explained with reference to the flowchart in Fig. 4. In Fig. 4, after the process of step S101, it is determined whether the user's question contains sufficient information to identify the document DB to be used for the search (step S201). The process of step S201 may be performed by the first document DB switching function 22 or the second document DB switching function 26.

[0033] In the process of step S201, if it is determined that the user's question contains sufficient information to identify the document DB to be used for the search (step S201: Yes), the process of step S102 is performed. On the other hand, in the process of step S201, if it is determined that the user's question does not contain sufficient information to identify the document DB to be used for the search (step S201: No), the second document DB switching function 26 performs a re-asking process to ask the user again (step S202). After the process of step S202 is performed, if the user answers the re-asking, the process of step S102 is performed.

[0034] (Technical Effects) In the dialogue system 2 according to this embodiment, if a user's question lacks information for identifying a document DB to be used for the search, the second document DB switching function 26 asks the user for a reconsideration question. The first document DB switching function 22 may identify a document DB to be used for the search based on the user's question and the user's response to the reconsideration question. This configuration makes it possible to appropriately identify a document DB to be used for the search. Therefore, the dialogue system 2 can improve the accuracy of answers to user questions.

[0035] <Third embodiment> A third embodiment of the dialogue system will be described with reference to Figures 5 and 6. The dialogue system 3 according to the third embodiment is similar to the dialogue system 1 according to the first embodiment, except that the information processing device 20 and the storage device 30 are partially different in configuration. Therefore, for the third embodiment, descriptions that overlap with those of the first embodiment will be omitted as appropriate. Furthermore, in the drawings, parts that are common to the first embodiment are denoted by the same reference numerals.

[0036] 3, the dialogue system 2 includes an input device 10, an information processing device 20, a storage device 30, an output device 40, and a generation AI 50. The information processing device 20 includes a question receiving function 21, a first document DB switching function 22, a document DB 23, a prompt inquiry function 24, an answer output function 25, and a third document DB switching function 27. The storage device 30 includes a plurality of document DBs, a switching setting file 31, a document DB information file 32, and CRM (Customer Relationship Management) data 33. The CRM data 33 stores information about users.

[0037] For example, as shown in Table 2, the document DB information file 32 in the storage device 30 may define the correspondence between a combination of vehicle model, drive system, and model year, a document, and a document DB related to the document. If the user's question does not include at least one of vehicle model, drive system, and model year, the first document DB switching function 22 may have difficulty identifying the document DB to use for the search.

[0038] If the user's question lacks information for identifying the document DB to be used for the search (for example, at least one of the vehicle model, drive system, and model year), the third document DB switching function 27 may acquire information about the user from the CRM data 33. The third document DB switching function 27 may acquire supplementary information from the acquired information as information for identifying the document DB to be used for the search.

[0039] The first document DB switching function 22 may identify a document DB to be used for search based on the user's question and the supplemental information acquired by the third document DB switching function 27. Specifically, the first document DB switching function 22 may detect keywords from each of the user's question and the supplemental information. The first document DB switching function 22 may identify a regular expression for the detected keyword by referring to the switching setting file 31 in the storage device 30. The first document DB switching function 22 may identify a document DB to be used for search from multiple document DBs based on the regular expression and the document DB information file 32 in the storage device 30.

[0040] The operation of the dialogue system 3 will be further explained with reference to the flowchart in Fig. 6. In Fig. 6, after the process of step S101, it is determined whether the user's question contains sufficient information to identify the document DB to be used for the search (step S301). Note that the process of step S301 may be performed by the first document DB switching function 22 or the third document DB switching function 27.

[0041] In the process of step S301, if it is determined that the user's question contains sufficient information to identify the document DB to be used for the search (step S301: Yes), the process of step S102 is performed. On the other hand, in the process of step S301, if it is determined that the user's question does not contain sufficient information to identify the document DB to be used for the search (step S301: No), the third document DB switching function 27 performs a CRM data acquisition process to acquire supplemental information from the CRM data 33 in the storage device 30 (step S302). Thereafter, the process of step S102 is performed.

[0042] (Technical Effects) In the dialogue system 3 according to this embodiment, if a user's question lacks information for identifying a document database to be used for the search, the third document database switching function 27 acquires supplemental information from the CRM data 33. The first document database switching function 22 may identify a document database to be used for the search based on the user's question and the supplemental information. This configuration makes it possible to appropriately identify a document database to be used for the search. Therefore, the dialogue system 3 can improve the accuracy of answers to user questions.

[0043] In the above-described embodiment, an automobile is given as an example of a product, but the product may also be, for example, computer software, computer hardware, etc. If the product is software, the document DB information file 32 may define the correspondence between a combination of an application name, a compatible OS (Operating System), and a version number, a document (for example, a software usage manual, a user's guide, a specification, etc.), and a document DB related to the document.

[0044] Aspects of the invention derived from the above-described embodiment and modifications will be described below.

[0045] A dialogue system according to one aspect of the invention comprises: a designation means for designating a database to be used for a search from among multiple databases based on a user's question; a generation means for generating a prompt to be input to a generation AI based on the search results of the search using the designated database and the question; and an output means for outputting the output of the generation AI in response to the prompt as an answer to the question. In the above-described embodiment, the "first document DB switching function 22" corresponds to an example of a "designation means," the "prompt inquiry function 24" corresponds to an example of a "generation means," and the "answer output function 25" corresponds to an example of an "output means."

[0046] The dialogue system may include a requesting means for requesting additional information from the user when the question lacks information for specifying a database to be used in the search. In this case, the specifying means may specify a database to be used in the search based on the question and the additional information. In the above-described embodiment, the "second document DB switching function 26" corresponds to an example of a "requesting means."

[0047] The dialogue system may include an acquisition unit that acquires supplemental information from pre-registered user information about the user when the question lacks information for specifying the database to be used in the search. In this case, the designation unit may designate the database to be used in the search based on the question and the supplemental information. In the above-described embodiment, the "third document DB switching function 27" corresponds to an example of the "acquisition unit."

[0048] In the dialogue system, each of the plurality of databases may include vector data in which character strings representing documents are vectorized.

[0049] The present invention is not limited to the above-described embodiments, but can be modified as appropriate within the scope of the claims and the gist or idea of ​​the invention as can be read from the entire specification, and dialogue systems involving such modifications are also included in the technical scope of the present invention. [Explanation of symbols]

[0050] 1, 2, 3...Dialogue system, 20...Information processing device, 21...Question reception function, 22...First document DB switching function, 23...Document DB search function, 24...Prompt inquiry function, 25...Answer output function, 30...Storage device, 31...Switching setting file, 32...Document DB information file, 50...Generation AI

Claims

1. a designation means for designating a database to be used for search from among a plurality of databases based on a user's query; A generation means for generating a prompt to be input to the generation AI based on the search results of the search using the specified database and the question; an output means for outputting an output of the generated AI in response to the prompt as an answer to the question; A dialogue system comprising:

2. a requesting means for requesting additional information from the user when the query lacks information for specifying a database to be used in the search; The designation means designates a database to be used for the search based on the question and the additional information.

2. The dialogue system according to claim 1 .

3. an acquisition means for acquiring supplemental information from pre-registered user information about the user when the question lacks information for specifying a database to be used for the search; The designation means designates a database to be used for the search based on the question and the supplemental information.

2. The dialogue system according to claim 1 .

4. 4. The interactive system according to claim 1, wherein each of the plurality of databases contains vector data in which character strings representing documents are vectorized.

Images