Information processing device, information processing method

Abstract

To provide a technology that facilitates the selection of desired search results from a set of document search results. [Solution] Based on a question from the user, the information processing device identifies a section in a document containing information related to the answer to that question as a source. The information processing device then obtains difference information showing the differences between the identified source sections, or the category of each of the identified source sections, and outputs the difference information or the category.

Description

Technical Field

[0001] The present invention relates to a technique for responding to queries.

Background Art

[0002] The creation of in-house document databases has progressed, and the importance of document search systems for efficiently retrieving documents required for business operations is increasing. In the field of search systems, a technique has been proposed that facilitates selection of a desired search result from a group of search results by presenting the characteristics of each source of the search results.

[0003] Patent Document 1 discloses a technique for displaying identification words of a document, which are determined based on a preset priority rule from a group of words characterizing the document acquired for each document to be searched, in an identifiable manner in association with the document.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In the method of Patent Document 1, when the groups of words characterizing the documents to be searched are similar between documents, it is difficult to select a desired search result from the group of search results.

Means for Solving the Problems

[0006] One aspect of the present invention is characterized by comprising: an identification means for identifying a portion of a document containing information related to the answer to a question, based on a question from a user; an acquisition means for acquiring difference information indicating the differences among a plurality of source locations identified by the identification means, or the respective categories of the plurality of source locations identified by the identification means; and an output means for outputting the difference information or the categories. [Effects of the Invention]

[0007] According to the present invention, it is possible to easily select a desired search result from a group of document search results. [Brief explanation of the drawing]

[0008] [Figure 1] A diagram illustrating the system's usage scenarios. [Figure 2] (a) is a block diagram showing an example of the hardware configuration of the tablet 103, and (b) is a block diagram showing an example of the hardware configuration of the search system 104. [Figure 3] A block diagram showing example functional configurations for the tablet 103 and the search system 104. [Figure 4] Flowcharts of the processes performed by tablet 3 and search system 104, respectively. [Figure 5] A diagram showing an example of a prompt template. [Figure 6] A diagram showing an example of presentation using the presentation device 104. [Figure 7] A diagram showing an example of a prompt template. [Figure 8] A diagram showing an example of presentation using the presentation device 104. [Figure 9] A diagram showing an example of the configuration of the display screen. [Figure 10] (a) is a diagram showing an example of a prompt template, and (b) is a diagram showing an example of presentation by the presentation device 104. [Figure 11] (a) is a diagram showing an example of a prompt template, and (b) is a diagram showing an example of presentation by the presentation device 104. [Modes for carrying out the invention]

[0009] The embodiments will be described in detail below with reference to the attached drawings. Note that the following embodiments do not limit the invention to the claims. While the embodiments describe multiple features, not all of these features are essential to the invention, and the features may be combined in any way. Furthermore, in the attached drawings, the same or similar configurations are given the same reference numerals, and redundant descriptions are omitted.

[0010] [First Embodiment] First, the usage scenario of the system according to this embodiment will be explained using Figure 1. In the usage scenario in Figure 1, a problem has occurred with the refrigerator 101. The problem with the refrigerator 101 is, for example, a water leak. A service technician 102 (user) called to repair the refrigerator 101 uses a tablet 103, which is an example of an information processing device, to find a solution to the problem. First, the service technician 102 operates the tablet 103 to input a question, and after inputting the question, instructs the tablet to send the question. When instructed to send the question, the tablet 103 transmits the input question to a search system 104, which is an example of an information processing device, via a wireless network. Based on the question received from the tablet 103, the search system 104 identifies a location in a document containing information related to the answer to the question as a source, obtains difference information showing the differences between the multiple identified source locations, and transmits the obtained difference information to the tablet 103. The tablet 103 receives the difference information received from the search system 104 and presents it to the service technician 102. Service technician 102 identifies a solution to the problem with refrigerator 101 based on the provided discrepancy information.

[0011] Next, an example of the hardware configuration of the tablet 103 will be explained using the block diagram in Figure 2(a). Note that the hardware configuration example shown in Figure 2(a) is just one example of a hardware configuration applicable to the tablet 103, and can be modified or changed as appropriate.

[0012] The CPU 201 executes various processes using computer programs and data stored in the RAM 205. Accordingly, the CPU 201 controls the overall operation of the tablet 103, and also executes or controls various processes described as processes performed by the tablet 103.

[0013] The ROM 202 stores setting data of the tablet 103, computer programs and data related to startup of the tablet 103, computer programs and data related to basic operations of the tablet 103, and the like. For example, the ROM 202 stores a BIOS (Basic Input / Output System) program, a boot program, and the like.

[0014] The input device 203 is a user interface such as a keyboard, a mouse, or a touch panel, and can input various instructions and information to the tablet 103 through user operations.

[0015] The presentation device 204 is an interface for presenting the processing result of the tablet 103 to the user. For example, the presentation device 204 has a liquid crystal screen or a touch panel screen, and can display the processing result from the CPU 201 as an image or text. Also, for example, the presentation device 204 has a speaker, and can output the processing result from the CPU 201 as sound.

[0016] The RAM 205 has an area for storing computer programs and data loaded from the ROM 202 or the storage device 206, and an area for storing computer programs and data received from an external apparatus via the communication interface 207. The RAM 205 also has a work area used when the CPU 201 executes various processes. In this manner, the RAM 205 can appropriately provide various areas.

[0017] The storage device 206 is a non-volatile memory device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive). The storage device 206 stores an OS, computer programs and data for causing the CPU 201 to execute or control various processes described as processes performed by the tablet 103, and the like.

[0018] The communication interface 207 is an interface for the tablet 103 to perform data communication with an external apparatus via a network such as a LAN or the Internet. The network may be a wired network or a wireless network. Further, the type of the network is not limited to a specific type such as USB or serial communication.

[0019] The CPU 201, ROM 202, input device 203, presentation device 204, RAM 205, storage device 206, and communication interface 207 are all connected to a system bus 208.

[0020] Next, a hardware configuration example of the search system 104 will be described using the block diagram of Figure 2(b). Here, a case where the search system 104 is implemented by a single computer apparatus will be described, but the present invention is not limited to this, and the search system 104 may be implemented by a plurality of computer apparatuses.

[0021] The CPU 251 executes various processes using computer programs and data stored in the RAM 252. Thereby, the CPU 251 controls the overall operation of the search system 104, and executes or controls various processes described as processes performed by the search system 104.

[0022] RAM252 has areas for storing computer programs and data loaded from ROM253 and storage device254, and areas for storing computer programs and data received from external devices via the communication interface255. Furthermore, RAM252 has a work area used by the CPU251 when executing various processes. In this way, RAM252 can provide various areas as appropriate.

[0023] ROM253 stores configuration data for the search system 104, computer programs and data related to the startup of the search system 104, computer programs and data related to the basic operation of the search system 104, and so on.

[0024] The storage device 254 is a non-volatile memory device such as an SSD or HDD. The storage device 254 stores computer programs and data that cause the CPU 251 to execute or control various processes, which will be described as processes performed by the OS and the search system 104.

[0025] The communication interface 255 is an interface for the search system 104 to communicate data with external devices via a network such as a LAN or the Internet. The network may be a wired network or a wireless network. Furthermore, the type of network is not limited to a specific type such as USB or serial communication. The CPU 251, RAM 252, ROM 253, storage device 254, and communication interface 255 are all connected to the system bus 256.

[0026] Examples of the functional configurations of the tablet 103 and the search system 104 are shown in the block diagrams of Figure 3. In this embodiment, we will describe a case in which each functional part of the tablet 103 shown in Figure 3 (excluding the display device 204) is implemented by software (computer program). In the following, each functional part of the tablet 103 shown in Figure 3 (excluding the display device 204) may be described as the main processing unit, but in reality, the function of the functional part is realized by the CPU 201 executing the computer program corresponding to that functional part.

[0027] Furthermore, in this embodiment, we will describe a case in which the search unit 310 of the search system 104 shown in Figure 3 is implemented by software (computer program). In the following, the search unit 310 may be described as the main processing unit, but in reality, the functions of the search unit 310 are realized by the CPU 251 executing the computer program corresponding to the search unit 310.

[0028] The processes performed by Tablet 3 and the search system 104 to present the user with difference information showing the discrepancies in the content of the answers to the questions entered by the user across multiple sources will be explained according to the flowchart in Figure 4. The following explanation will use the case where the question is textual information as an example.

[0029] In step S401, the question acquisition unit 301 acquires the question entered in response to user operation. The method by which the question acquisition unit 301 acquires the question is not limited to a specific method. For example, the question acquisition unit 301 may acquire character information entered by the user by operating the keyboard or mouse on the input device 203 as a question. Alternatively, for example, the question acquisition unit 301 may recognize the voice input (spoken) by the user to the microphone on the input device 203 and acquire the character information resulting from this recognition as a question. Furthermore, the question acquisition unit 301 may acquire still images or moving images entered by the user by operating the keyboard or mouse on the input device 203 as a question. The acquisition unit 302 then transmits the question acquired by the question acquisition unit 301 to the search system 104 via the communication interface 207.

[0030] In step S451, the search unit 310 receives a question sent from the tablet 103 via the communication interface 255. The search system 104 has a database 306, which is a document-oriented database in which character data such as text (for example, information on the maintenance of the refrigerator 101, information on the product design of the refrigerator 101, knowledge information, etc., documents in the field to be searched in order to obtain answers to the user's query, accumulated both inside and outside the company) is vectorized and stored. The database 306 is stored, for example, in the storage device 254. The documents stored in the database 306 must be in a common document file format such as a text format or document format that can be vectorized. The documents stored in the database 306 are not limited to common document file formats, but may also include non-character information or element data that contains a mixture of character and non-character information, such as images, audio, video, diagrams, and tables.

[0031] The search unit 310 then converts the question received from the tablet 103 into vector data (question vector data), and calculates the cosine similarity between the question vector data and the vector data of the text in each paragraph of one or more documents registered in the database 306. The search unit 310 then identifies the paragraphs for which it has calculated a cosine similarity of a threshold or higher as the location in the document containing information related to the answer to the question (source location), and for each of the identified paragraphs, it generates a set of source location information consisting of the paragraph ID and the document ID to which the paragraph belongs. The search unit 310 then transmits the generated source location information to the tablet 103 via the communication interface 255.

[0032] In step S402, the acquisition unit 302 receives (acquires) source location information transmitted from the search system 104 via the communication interface 207. In step S403, the prompt creation unit 303 creates a prompt to inquire about the differences in content between each source location identified by the source location information received in step S402.

[0033] For example, the prompt creation unit 303 creates a prompt by inserting the source information obtained in step S402 into a prompt template, which is a predetermined string. Prompt templates are pre-stored in the ROM 202 or storage device 206. An example of a prompt template is shown in Figure 5. Prompts are written in Markdown format.

[0034] In the prompt template 500, description 501 is a fixed sentence (inquiry sentence) for querying the differences in content between sources. Description 502 is a sentence that specifies the group of sources. The prompt creation unit 303 inserts the source information received in step S402 as a string list into the "context" part of description 502. The prompt creation unit 303 then sends the created prompt to the search system 104 via the communication interface 207.

[0035] In step S452, the search unit 310 receives a prompt transmitted from the tablet 103 via the communication interface 255. The search unit 310 then replaces the source information in the received prompt with the source corresponding to that source information (the text of the paragraph in the document identified by the paragraph ID and document ID indicated by the source information), inputs the replaced prompt into the large-scale language model 307, and processes the large-scale language model 307 to generate difference information indicating the differences in content between each source.

[0036] The large-scale language model 307 is a general-purpose large-scale language model that takes a question sentence called a prompt as input and generates an answer sentence. The information output by the large-scale language model 307 is textual information, and because the prompt instructs it to express the difference in source locations in one sentence, a list of sentences expressing the difference in one sentence for each source location is output as difference information. The large-scale language model 307 uses the general-purpose model GPT-4 (trademark registered), which is a model that has been pre-trained in an unsupervised manner using a large amount of text data to learn grammar, word meanings, etc. The search unit 310 then transmits the generated difference information to the tablet 103 via the communication interface 255.

[0037] In step S404, the difference information acquisition unit 304 receives the difference information transmitted from the search system 104 via the communication interface 207. In step S405, the presentation content determination unit 305 determines each difference indicated by the difference information received in step S404 as content to be presented to the user (presentation content), and presents the presentation content to the user via the presentation device 204.

[0038] For example, as shown in Figure 6, the content determination unit 305 may display on the display screen of the presentation device 204 a screen containing a sentence 601 that expresses the difference between one source (first source) and other sources, a sentence 602 that expresses the difference between another source (second source) and other sources different from the first source, and a sentence 603 that expresses the difference between another source (third source) and other sources different from the first and second sources. Note that the method of presenting the content is not limited to display; for example, the content may be presented to the user by voice.

[0039] Thus, according to this embodiment, even when the sets of words that characterize the documents being searched are similar across documents, the user can easily select the desired search result from the set of search results.

[0040] <Modification 1 of the first embodiment> As described above, when a user inputs a question by voice into the microphone of the input device 203, the question acquisition unit 301 converts the input voice into text information, using a general-purpose speech recognition model (for example, Whisper®). This conversion process is performed in step S401 described above.

[0041] Thus, according to this modified example, even if the user has difficulty typing, they can enter a question into the tablet 103 and perform the processes from step S402 onward.

[0042] <Modification 2 of the first embodiment> In the first embodiment, the search system 104 allowed multiple documents to be used as search targets in response to a user's question. In contrast, this modification describes a case where the search target document in response to a user's question is a single document.

[0043] The database 306 contains vector data of the text of each paragraph in a single document (target document). Therefore, the search unit 310 calculates the cosine similarity between, for example, the question vector data and the vector data of the text of each paragraph in the target document. The search unit 310 then identifies the paragraphs for which it has calculated a cosine similarity equal to or greater than a threshold as source locations, and for each of the identified paragraphs, it generates a set of source location information consisting of the paragraph ID and the target document ID.

[0044] Thus, this modification allows users to select their desired search results even when multiple sources are presented from a single document and the content is similar, making it difficult for them to choose a result.

[0045] <Modification 3 of the first embodiment> In the first embodiment, the unit of a source was a paragraph. However, a source can be any information indicating a region within a document that contains a continuous string of characters, and its unit is not limited to a paragraph. For example, if the unit of a source is a section, the source information is a set of section ID and document ID. In addition to sections, chapters, pages, and characters are also possible units. Furthermore, the unit of a source is not limited to units within a document; the document itself can also be the unit. If the content of a single document is to be obtained as a source, the document itself becomes the unit of the source.

[0046] Furthermore, although the first embodiment used only one type of unit for the source location, it is not limited to this, and multiple types of units may be used in combination. In this case, each source location information is accompanied by an identifier indicating the type of source location unit.

[0047] Thus, according to this modified version, the appropriate source can be obtained by allowing the user to select the unit of source, such as a document unit or a paragraph unit. Furthermore, the unit of source can be selected by including the user's preferred unit in the question. This allows for obtaining source information and presenting differences according to the user's intent.

[0048] <Modification 4 of the first embodiment> In the first embodiment, the source was searched using a vector-based search (vector search), but the method for searching for the source is not limited to vector search as long as it is a search of a document database. For example, the source may be searched using search methods such as full-text search, keyword search, or faceted search.

[0049] Furthermore, multiple search methods may be used in combination to acquire exhibition locations. For example, when using a hybrid search combining vector search and full-text search, it is possible to achieve both the accuracy of keyword search and the semantic relationships of vector search, thereby obtaining highly accurate search results. In this modified example, the case in which a hybrid search combining vector search and full-text search is used will be described. The processing performed by the tablet 3 and the search system 104 to present difference information to the user will follow the flowchart in Figure 4, except for the following points.

[0050] In step S451, the search unit 310, similar to the first embodiment, converts the question received from the tablet 103 into question vector data, calculates the cosine similarity (vector search score) between the question vector data and the text vector data of each paragraph in one or more documents registered in the database 306, and performs a vector search to identify the paragraphs for which a cosine similarity of a certain threshold or higher has been calculated as the source.

[0051] Furthermore, the search unit 310 indexes each document stored in the database 306 and performs a full-text search within those documents to find paragraphs that match the question received from the tablet 103. In the full-text search, a score (full-text search score) is calculated for each paragraph found. Elasticsearch, a distributed search engine that efficiently performs full-text searches, is used for the full-text search, and paragraphs containing specific keywords are obtained as sources as a result of the full-text search.

[0052] The search unit 310 then calculates an integrated score for each searched paragraph by normalizing and weighting the vector search score and full-text search score calculated for that paragraph.

[0053] The weighting of the vector search score and the full-text search score is adjusted depending on how much emphasis is placed on semantic relevance or keyword match. Here, as an example of emphasizing semantic relevance, the weight of the vector search is set high. For example, if the weight of the vector search score is set to 0.6 and the weight of the full-text search score is set to 0.4, the combined score will be calculated as Combined score = 0.6 x Vector search score + 0.4 x Full-text search score.

[0054] The search unit 310 then identifies paragraphs that have calculated an integrated score above a threshold as locations within the document containing information related to the answer to the question (source locations), and for each identified paragraph, it generates a set of source location information consisting of the paragraph ID and the document ID to which the paragraph belongs. This allows for search results that take into account semantic relevance and keyword matching.

[0055] As demonstrated in this modified version, a hybrid search combining vector search and full-text search can obtain search results that simultaneously consider semantic relevance and keyword matching. This allows users to obtain more relevant information as sources.

[0056] <Modification 5 of the First Embodiment> In the first embodiment, the prompt creation unit 303 used a prompt template, which is a predetermined (fixed) string, to create a prompt. However, the prompt only needs to inquire about differences in source locations, and the prompt template is not limited to a predetermined (fixed) string.

[0057] For example, the prompt creation unit 303 displays multiple sentences that have been created in advance and stored in the storage device 206 on the display screen of the presentation device 204. The prompt creation unit 303 then selects a sentence from among these sentences by the user operating the input device 203 as the query sentence, and creates a prompt that includes the query sentence and source information. Alternatively, the prompt can be rewritten from outside the tablet 103, the tablet 103 can change the query sentence as appropriate, or the tablet 103 can generate the query sentence.

[0058] Furthermore, in the first embodiment, description 501 instructs the system to generate differences in content for all sources, but it is not limited to this, and the prompt may specify the number of sources to output. For example, the prompt may specify that differences be generated for 10 paragraphs.

[0059] Furthermore, in the first embodiment, description 501 instructs the system to generate a single sentence indicating the differences in the content of the source, but the system is not limited to this, and the prompt may specify the length of the difference information in multiple sentences or by the number of characters.

[0060] Furthermore, while the first embodiment presents differences in content for each source, it is not limited to this; for example, it may also present an overview of the source and information on the differences. In this case, the search unit 310 transmits an overview (part of the source) of each source to the tablet 103 in addition to the information on the differences. Alternatively, the entire source may be transmitted instead of just an overview.

[0061] Furthermore, the length of the difference information, the type of information presented regarding the source, and the manner of presentation may be fixed or variable. For example, these may be changeable based on the amount of information that can be displayed on the display screen of the presentation device 204. In this way, this modified example allows for adjustment of the balance between the ease of viewing information and the amount of information.

[0062] <Modification 6 of the First Embodiment> In the first embodiment, a general-purpose model is used as the large-scale language model 307. However, the large-scale language model 307 is not limited to a general-purpose model; for example, a dedicated model may be used as the large-scale language model 307. The dedicated model may be a model specialized for a particular field, a selection of multiple models with specific features, or a fine-tuned model obtained by retraining the general-purpose model.

[0063] For example, when using a retrained general-purpose model as the large-scale language model 307, a general-purpose natural language processing model that has been pre-trained unsupervised on grammar and word meanings using a large amount of text data is used as the base model, and then supervised learning of knowledge in a specific domain is performed. As the base model, GPT-4 (trademark registered) is used for training. Furthermore, the training dataset used for retraining is prepared by preparing a set of source information to be treated as the correct answer and a set of difference information, and then fine-tuned. In this modified example, by obtaining more accurate difference information, users can easily select the desired source from multiple sources.

[0064] <Modification 7 of the first embodiment> In the first embodiment, the content determination unit 305 presents all differences indicated by the difference information received in step S404 to the user, but it is also possible to present only a portion of the differences indicated by the difference information received in step S404 to the user. In this modified example, a configuration is described in which the user can select which differences to present to the user from among the multiple differences indicated by the difference information received in step S404. The processing performed by the tablet 3 and the search system 104 to present the difference information to the user follows the flowchart in Figure 4, except for the following points.

[0065] In step S451, the search unit 310 obtains a search score by normalizing the cosine similarity calculated for each source identified in the same manner as in the first embodiment to between 0 and 1. The search unit 310 then transmits the search score for each source and the source information generated in the same manner as in the first embodiment to the tablet 103 via the communication interface 255.

[0066] In step S402, the acquisition unit 302 receives (acquires) the search score and source information transmitted from the search system 104 via the communication interface 207.

[0067] In step S405, the content determination unit 305 determines, based on the search score above, which content to present to the user (presentation content) from among the differences indicated by the difference information received in step S404.

[0068] For example, the content determination unit 305 determines the differences between the source locations indicated by the difference information received in step S404 as the content to be presented, specifically those source locations whose search score is equal to or greater than a threshold.

[0069] In this case, the content determination unit 305 may display the differences in source locations with search scores above a threshold on the display screen of the presentation device 204, in descending order of search score. In this modified form, by controlling the number of differences presented to the user, the time it takes for the user to select the desired source location from the presented information can be reduced.

[0070] [Second Embodiment] In the following embodiments and modifications, including this embodiment, the differences from the first embodiment will be explained, and unless otherwise specified below, they will be the same as the first embodiment. In the first embodiment, the difference information of the source location was presented to the user in response to a question from the user. In contrast, this embodiment describes a case in which the source location is also presented to the user in addition to the difference information of the source location. The processing performed by the tablet 3 and the search system 104 in order to present the difference information and the source location to the user will follow the flowchart in Figure 4, except for the following points.

[0071] In step S403, the prompt generation unit 303 creates prompts to query each source identified by the source information received in step S402, and the differences in content between each of those sources.

[0072] For example, the prompt creation unit 303 creates a prompt by inserting the source information obtained in step S402 into a prompt template, which is a predetermined string. Prompt templates are pre-stored in the ROM 202 or storage device 206. An example of a prompt template is shown in Figure 7.

[0073] In prompt template 700, description 701 is a fixed sentence for inquiring about the source and the differences in content between the sources. Note that description 701 is not limited to a fixed sentence; it may be editable by the user operating input device 203, or it may be editable from an external device.

[0074] Description 702, like description 502, is a statement that specifies a group of sources. The prompt generation unit 303 inserts the source information received in step S402 as a string into the "context" part of description 702. The prompt generation unit 303 then sends the created prompt to the search system 104 via the communication interface 207.

[0075] In step S452, the search unit 310 transmits the generated difference information, along with the source location, to the tablet 103 via the communication interface 255. In step S404, the difference information acquisition unit 304 receives the difference information and source location transmitted from the search system 104 via the communication interface 207.

[0076] In step S405, the content determination unit 305 determines the differences for each source location indicated by the difference information received in step S404, along with the source locations themselves, as the content to be presented to the user (presented content), and presents the presented content to the user via the presentation device 204.

[0077] For example, as shown in Figure 8, the content determination unit 305 may display a list 801 on the display screen of the presentation device 204, which includes, for each source location, a set of the source location and a sentence that expresses the difference between that source location and other sources. The method of presenting the content is not limited to display; for example, the content may be presented to the user by voice. In this way, according to this embodiment, it is possible to make it easier to select the desired search result from the search result group.

[0078] [Third Embodiment] In the first and second embodiments, the search system 104 transmits the difference information to the tablet 103 after acquiring all the difference information for the source locations. In contrast, in this embodiment, the search system 104 transmits the difference information to the tablet 103 each time it acquires difference information for a predetermined number of source locations. The processing performed by the tablet 3 and the search system 104 to present the difference information to the user follows the flowchart in Figure 4, except for the following points.

[0079] In step S403, the prompt generation unit 303 queries for the differences in content between each source identified by the source information received in step S402, and creates a prompt requesting output for each specified number of source difference information. The specified number of source locations refers to source locations at specified divisions, such as one paragraph, two paragraphs, etc. This "specified number" may be a pre-set value, or it may be set / changed by the user by operating the input device 203. The following explanation will use the case where the specified number is 1 as an example.

[0080] In step S452, the search unit 310 receives a prompt sent from the tablet 103 via the communication interface 255. The search unit 310 then generates difference information in the same manner as in the first embodiment, and each time it generates difference information for a predetermined number of sources, it transmits the difference information generated for that predetermined number of sources to the tablet 103 via the communication interface 255.

[0081] As a result, the presentation device 204 can present the difference information to the user each time a difference information is generated for a predetermined number of sources, without having to wait for the generation of difference information for all sources.

[0082] Thus, according to this embodiment, the initial presentation of difference information to the user is accelerated, and difference information can be obtained sequentially. Therefore, when the information the user requests is presented, they can obtain the information without having to wait until the entire presentation of difference information is complete.

[0083] <Modified form of the third embodiment> In the method of the third embodiment, if additional difference information is acquired, the position of the difference information presented to the user may change. On the other hand, by inserting a predetermined space between the presentation positions of two different pieces of difference information, the change in the presentation position of the information when additional difference information is presented can be mitigated. The processing performed by the tablet 3 and the search system 104 to present the difference information to the user is the same as in the third embodiment, except for the following points.

[0084] In step S403, the prompt generation unit 303 inquires about the differences in content between each source identified by the source information received in step S402, requests output for each specified number of source difference information, and creates a prompt that specifies that the number of characters in the source difference information should be less than or equal to a specified number of characters. This "specified number of characters" may be a pre-set value, or it may be set / changed by the user by operating the input device 203.

[0085] In step S452, the search unit 310 receives a prompt sent from the tablet 103 via the communication interface 255. The search unit 310 then generates difference information in the same manner as in the third embodiment, but performs processing to fit the number of characters in the difference information to the number of characters specified in the prompt. For example, the search unit 310 deletes characters from the beginning of the difference information so that the number of characters from the beginning is the number of characters specified in the prompt, thereby generating "difference information where the number of characters from the beginning is the number of characters specified in the prompt". Each time the search unit 310 generates difference information for a predetermined number of sources, it transmits the difference information generated for that predetermined number of sources to the tablet 103 via the communication interface 255.

[0086] In step S404, the difference information acquisition unit 304 receives the difference information transmitted from the search system 104 via the communication interface 207. Here, the presentation content determination unit 305 configures a presentation screen by arranging display areas capable of arranging a string of more than the number of characters specified by the prompt, at predetermined distances, for the number of difference information items (number of paragraph IDs), and displays it on the display screen of the presentation device 204.

[0087] Then, in step S405, the presentation content determination unit 305 determines each of the differences indicated by the difference information received in step S404 as the content to be presented to the user (presentation content), and displays the differences in the corresponding display area.

[0088] Figure 9 shows an example of the display screen configuration. Figure 9(a) shows an example of the display before the additional difference information is displayed, and Figure 9(b) shows an example of the display after the additional difference information is displayed. Display area 901 displays the difference information for the first source. Display area 902 is the area where the difference information for the second and subsequent sources is displayed. A predetermined distance is provided between the display areas. Figure 9(b) shows the display screen after the difference information is displayed in display area 902. As shown in Figure 9(b), the display area for the difference information remains unchanged from Figure 9(a). The space between the display areas may be blank, or a message indicating "Generating" may be displayed if the difference information to be displayed in the display area is being generated. In this way, according to this modified example, the user can easily select the desired search result without confusing information about multiple different sources.

[0089] [Fourth Embodiment] In the first to third embodiments, differences in the content of the source material are obtained and presented to the user. In contrast, this embodiment describes a case in which the category of the content of the source material is presented.

[0090] The question acquisition unit 301 acquires the question entered in response to user operation. The acquisition unit 302 then transmits the question acquired by the question acquisition unit 301 to the search system 104 via the communication interface 207.

[0091] The search unit 310 receives the question sent from the tablet 103 via the communication interface 255. The search unit 310 then generates source information in the same manner as in the first embodiment and transmits the generated source information to the tablet 103 via the communication interface 255.

[0092] The acquisition unit 302 receives (acquires) source location information transmitted from the search system 104 via the communication interface 207. The prompt creation unit 303 creates prompts to query the category of the content of each source location identified by the source location information received from the search system 104.

[0093] For example, the prompt creation unit 303 creates a prompt by inserting the source information obtained in step S402 into a prompt template, which is a predetermined string. An example of a prompt template is shown in Figure 10(a). In prompt template 1000, description 1001 is a fixed sentence for querying the category of the content published in the source, but it may be editable as appropriate. Description 1002 is a sentence that specifies a group of source locations, similar to description 502 above. The prompt creation unit 303 inserts the source information received from the search system 104 as a string in the "context" part of description 1002. Then, the prompt creation unit 303 sends the created prompt to the search system 104 via the communication interface 207.

[0094] The search unit 310 receives prompts transmitted from the tablet 103 via the communication interface 255. The search unit 310 then replaces the source information in the received prompt with the source corresponding to that source information (the text of the paragraph identified by the paragraph ID and document ID indicated by the source information), inputs the replaced prompt into the large-scale language model 307, and processes the large-scale language model 307 to generate categories for the content of each source. The large-scale language model 307 in this embodiment is the same as the large-scale language model 307 in the above embodiment except for the input and output, and uses the general-purpose model GPT-4 (trademark registered), which is a model that has been pre-trained in an unsupervised manner using a large amount of text data to learn grammar, word meanings, etc. The search unit 310 then transmits the generated categories to the tablet 103 via the communication interface 255.

[0095] The difference information acquisition unit 304 receives the categories transmitted from the search system 104 via the communication interface 207. The presentation content determination unit 305 then determines the category of each source received from the search system 104 as the presentation content and presents this content to the user via the presentation device 204.

[0096] For example, the content determination unit 305 may display a screen containing the categories 1003a, 1003b, 1003c, and 1003d of the content published at each source on the display screen of the presentation device 204, as shown in Figure 10(b). The method of presenting the content is not limited to display; for example, the content may be presented to the user by voice.

[0097] In this embodiment, we have described a case in which the category of the source is presented to the user, but it is also possible to present the user with information on the differences in the source and the category. In that case, the prompt creation unit 303 creates a prompt using the prompt template exemplified in Figure 11(a).

[0098] In the prompt template 1100, description 1101 is a fixed sentence for querying the differences in content between sources and the categories of the sources, but it may be editable as needed. Description 1102 is a sentence that specifies the group of sources. The prompt creation unit 303 inserts the source information received in step S402 in the "context" part of description 1102 as a string list.

[0099] Furthermore, in addition to the difference information and category of the source location, Figure 11(b) shows an example of the configuration of the presentation screen that the presentation content determination unit 305 displays on the presentation device 204 when the source location is obtained as described above. Description 1103 indicates the category of the source location, and description 1104 indicates the source location and the difference information of said source location.

[0100] Furthermore, if the number of paragraph IDs (number of sources) indicated by the source information is greater than or equal to a threshold, the prompt generation unit 303 creates a prompt to inquire about the category of content for each source identified by the source information received from the search system 104.

[0101] On the other hand, if the number of paragraph IDs (number of sources) indicated by the source information is less than a threshold, the prompt generation unit 303 creates a prompt to inquire about the differences in content between each source identified by the source information received in step S402. In this way, according to this embodiment, the user can select the desired search result more quickly.

[0102] [Fifth Embodiment] The system usage scenarios described in the above embodiments and modifications are just examples of troubleshooting, and the system can be applied to other troubleshooting situations as well. Furthermore, the system's usage scenarios are not limited to troubleshooting.

[0103] Furthermore, the tablet 103 and the search system 104 may be located in a different place from the refrigerator 101, for example, on the maintenance company's side. In other words, various placements of the tablet 103 and the search system 104 are possible depending on the application scenario of the system.

[0104] Furthermore, although the above embodiments and modifications describe a case where the tablet 103 and the search system 104 are separate devices, the invention is not limited to this. For example, a single information processing device may be configured in which the tablet 103 and the search system 104 are combined. In this case, the information processing device executes or controls the various processes (excluding communication processes between devices) described as being performed by the tablet 103 and the search system 104, respectively.

[0105] Furthermore, the above embodiments and modifications describe a case where the tablet 103 has a question acquisition unit 301, an acquisition unit 302, a prompt creation unit 303, a difference information acquisition unit 304, and a presentation content determination unit 305, and the search system 104 has a database 306, a large-scale language model 307, and a search unit 310. However, it is not limited to this case. For example, the tablet 103 may also have a search unit 310. Also, the database 306 and the large-scale language model 307 may be provided by the tablet 103. In addition, one or more of the functional units shown in Figure 3 (excluding the presentation device 204) may be implemented in hardware.

[0106] The numerical values, processing timing, processing order, processing entity, data (information) structure / acquisition method / destination / source / storage location, etc., used in the above embodiment are given as examples for the purpose of providing a concrete explanation, and are not intended to limit the scope to such examples.

[0107] Furthermore, some or all of the embodiments described above may be used in appropriate combinations. Alternatively, some or all of the embodiments described above may be used selectively.

[0108] (Other embodiments) The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. It can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0109] The inventions described herein include the following information processing devices, information processing methods, and computer programs. (Item 1) A means for identifying a source location in a document that contains information related to the answer to a question, based on a question from a user. An acquisition means for obtaining difference information indicating the differences between multiple source locations identified by the identification means, or each category of the multiple source locations identified by the identification means, Output means for outputting the difference information or the category An information processing device characterized by comprising: (Item 2) The information processing device according to item 1, characterized in that the aforementioned identification means searches for the source location using a plurality of search methods. (Item 3) The information processing device according to item 1 or 2, characterized in that the identifying means identifies the source location from one or more documents. (Item 4) The information processing device according to any one of items 1 to 3, characterized in that the acquisition means acquires the difference information based on a prompt to query the difference information. (Item 5) The information processing device according to item 4, characterized in that the acquisition means acquires the difference information by inputting the prompt to a model that outputs differences for each source and processing the model. (Item 6) The information processing device according to item 4 or 5, characterized in that the prompt specifies the number of sources from which to output differences. (Item 7) The information processing device according to any one of items 4 to 6, characterized in that the length of the difference information is specified in the prompt. (Item 8) The information processing device according to any one of items 1 to 7, characterized in that the output means outputs the difference information and the source location or an overview of the source location. (Item 9) The information processing device according to any one of items 1 to 7, characterized in that the output means outputs the degree of similarity between the question and the source, and the difference information. (Item 10) The information processing device according to any one of items 1 to 7, characterized in that the output means outputs difference information of source locations among the difference information acquired by the acquisition means, the similarity to the question being equal to or greater than a threshold. (Item 11) The information processing device according to any one of items 1 to 7, characterized in that the output means outputs the difference information and the source location. (Item 12) The information processing device according to any one of items 1 to 7, characterized in that the output means outputs the difference information each time the acquisition means acquires a predetermined number of source locations. (Item 13) The information processing device according to any one of items 1 to 3, characterized in that the acquisition means acquires the category based on a prompt to query the category. (Item 14) The information processing device according to any one of items 1 to 3, characterized in that the acquisition means acquires the difference information and the category based on a prompt to query the difference information and the category. (Item 15) The information processing device according to any one of items 1 to 3, characterized in that the acquisition means acquires the category based on a prompt to query the category if the number of sources is equal to or greater than a threshold, and acquires the difference information based on a prompt to query the difference information if the number of sources is less than a threshold. (Item 16) The information processing device described in any one of items 1 to 15, characterized in that the aforementioned source is a unit of one of the following: a document, a paragraph in a document, a section in a document, a chapter in a document, a page in a document, or a character in a document. (Item 17) An information processing method performed by an information processing device, The identification means of the information processing device includes an identification step of identifying a location in a document containing information related to the answer to a question, based on a question from a user, The acquisition means of the information processing device includes an acquisition step of acquiring difference information indicating the differences between multiple source locations identified in the specified step, or the respective categories of the multiple source locations identified in the specified step, The output means of the information processing device includes an output step of outputting the difference information or the category. An information processing method characterized by comprising: (Item 18) A computer program that causes a computer to function as one of the means of an information processing device described in any one of items 1 through 16.

[0110] The invention is not limited to the embodiments described above, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, claims are attached to disclose the scope of the invention. [Explanation of Symbols]

[0111] 103: Tablet 104: Search system 204: Presentation device 301: Question acquisition unit 302: Acquisition unit 303: Prompt creation unit 304: Difference information acquisition unit 305: Presentation content determination unit 306: Database 307: Large-scale language model 310: Search unit

Claims

1. A means for identifying a source location in a document that contains information related to the answer to a question, based on a question from a user. An acquisition means for obtaining difference information indicating the differences between multiple source locations identified by the identification means, or each category of the multiple source locations identified by the identification means, Output means for outputting the difference information or the category An information processing device characterized by comprising:

2. The information processing device according to claim 1, characterized in that the aforementioned identification means searches for the source location using a plurality of search methods.

3. The information processing apparatus according to claim 1, characterized in that the identifying means identifies the source location from one or more documents.

4. The information processing apparatus according to claim 1, wherein the acquisition means acquires the difference information based on a prompt to query the difference information.

5. The information processing apparatus according to claim 4, characterized in that the acquisition means acquires the difference information by inputting the prompt to a model that outputs differences for each source and processing the model.

6. The information processing device according to claim 4, characterized in that the prompt specifies the number of sources from which to output differences.

7. The information processing device according to claim 4, characterized in that the length of the difference information is specified in the prompt.

8. The information processing apparatus according to claim 1, characterized in that the output means outputs the difference information and the source location or an overview of the source location.

9. The information processing apparatus according to claim 1, characterized in that the output means outputs the degree of similarity between the question and the source, and the difference information.

10. The information processing apparatus according to claim 1, characterized in that the output means outputs difference information of source locations among the difference information acquired by the acquisition means, the similarity to the question being equal to or greater than a threshold.

11. The information processing apparatus according to claim 1, characterized in that the output means outputs the difference information and the source location.

12. The information processing apparatus according to claim 1, wherein the output means outputs the difference information each time the acquisition means acquires a predetermined number of source locations.

13. The information processing apparatus according to claim 1, characterized in that the acquisition means acquires the category based on a prompt to query the category.

14. The information processing apparatus according to claim 1, wherein the acquisition means acquires the difference information and the category based on a prompt to query the difference information and the category.

15. The information processing apparatus according to claim 1, characterized in that the acquisition means acquires the category based on a prompt to query the category if the number of sources is equal to or greater than a threshold, and acquires the difference information based on a prompt to query the difference information if the number of sources is less than a threshold.

16. The information processing apparatus according to claim 1, characterized in that the aforementioned source is any of the units of a document, a paragraph in a document, a section in a document, a chapter in a document, a page in a document, or a character in a document.

17. An information processing method performed by an information processing device, The identification means of the information processing device includes an identification step of identifying a location in a document containing information related to the answer to a question, based on a question from a user, The acquisition means of the information processing device includes an acquisition step of acquiring difference information indicating the differences between multiple source locations identified in the specified step, or the respective categories of the multiple source locations identified in the specified step, The output means of the information processing device includes an output step of outputting the difference information or the category. An information processing method characterized by comprising:

18. A computer program for causing a computer to function as one of the means of an information processing apparatus described in any one of claims 1 to 16.

Images