Information processing apparatus, information processing method, and program

Abstract

To provide a technique for improving a user interface for interactive processing with a generation model using a reference database.SOLUTION: An aspect of the present disclosure relates to an information processing device including a reception unit that receives question data for a generation model, a prompt creation unit that extracts reference data related to the question data from a reference database and creates prompt data using the extracted reference data, a model interface unit that inputs the prompt data to the generation model and acquires answer data from the generation model, and a display unit that displays the answer data in association with the reference data.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] With the recent advances in deep learning, machine learning technology has been applied to a wide range of technical fields. For example, in the technical field of natural language processing, progress has been made in research and development of machine learning models used in information extraction, information retrieval, summary generation, machine translation, speech recognition, dialogue processing, etc. Natural language processing using generative models, such as the recently released ChatGPT (registered trademark), has attracted attention.

[0003] In dialogue processing using generative models, several approaches have been proposed to obtain more appropriate answers from a generative model in response to a question. One such approach is Retrieval-Augmented Generation (RAG). RAG extracts information related to the question from a reference database that holds accurate information, and then creates prompt data to be input to a generative model from the extracted related information and the question. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] In-Context Retrieval-Augmented Language Models(https: / / arxiv.org / abs / 2302.00083) Summary of the Invention [Problem to be solved by the invention]

[0005] Conventionally, an answer generated by a generative model may display a link to the source web page used to generate the answer. The displayed source web page is typically information used to train the generative model. A user can check the answer and also navigate to the displayed link to refer to the source web page, thereby being able to confirm the information on which the answer is based.

[0006] An object of the present disclosure is to provide a technique for improving the user interface for interacting with generative models that utilize a reference database. [Means for solving the problem]

[0007] One aspect of the present disclosure relates to an information processing device having a reception unit that receives question data for a generative model, a prompt creation unit that extracts reference data related to the question data from a reference database and creates prompt data using the extracted reference data, a model interface unit that inputs the prompt data into the generative model and acquires answer data from the generative model, and a display unit that displays the answer data in association with the reference data. [Effects of the Invention]

[0008] According to the present disclosure, it is possible to provide a technique for improving the user interface for interacting with a generative model using a reference database. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram illustrating an interaction process for a generative model using a reference database according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a schematic diagram illustrating reference data according to one embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating a hardware configuration of an information processing device according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a block diagram illustrating a functional configuration of an information processing device according to an embodiment of the present disclosure. [Figure 6] FIG. 6 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 7] FIG. 7 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 8] FIG. 8 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 10] FIG. 10 is a diagram illustrating an interaction screen with a generative model according to one embodiment of the present disclosure. [Figure 11] FIG. 11 is a flowchart illustrating an interaction process with a generative model according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0011] In the following embodiment, an information processing device is disclosed that realizes interactive processing for a generative model by using reference data extracted from a reference database.

[0012] [Summary of the Disclosure] In the interactive processing for a generative model according to an embodiment described below, prompt data is generated using information stored in a reference database 40. For example, as shown in Fig. 1, when a user device 20 of a user who uses interactive processing for a generative model 60 provides question data for the generative model 60 to an information processing device 100, the information processing device 100 transmits a search query to the reference database 40, which stores reference data that the generative model 60 should refer to in generating answer data, in order to extract information related to the question data.

[0013] For such reference data, a document file or the like obtained from one of the data sources is divided into text data called "chunks" according to a predetermined division method, and each divided text data is stored as reference data in the reference database 40. For example, as shown in Fig. 2, the reference database 40 stores reference data #X1, #X2, ... as text data extracted from a document file or the like obtained from one of the data sources. Note that tabular data included in the document file can be converted into text data in CVS format, JSON format, or the like using the Markdown method.

[0014] When the information processing device 100 extracts reference data corresponding to the search query from the reference database 40, it uses the question data acquired from the user device 20 and the reference data extracted from the reference database 40 to create prompt data for the generative model 60. For example, the creation of prompt data using such reference data may be performed in accordance with a known RAG or the like.

[0015] The information processing device 100 transmits the generated prompt data to the generative model 60, and acquires answer data for the prompt data from the generative model 60. Then, the information processing device 100 transmits the acquired answer data to the user device 20.

[0016] In an embodiment described later, the information processing device 100 displays answer data on an interactive screen for the generative model 60 on the user terminal 20, and embeds the reference data used to create the prompt data in the interactive screen. Specifically, when an answer to a question is displayed, the answer and reference data #1 to #4 detected for the question are displayed in a hidden state on the interactive screen. When the user performs an operation such as clicking on any of the reference data, the data content of the reference data changes from a hidden state to a visible state. For example, as shown in FIG. 3, when a user performs an operation on "reference data #3," the data content embedded in reference data #3 on the interactive screen is displayed.

[0017] When reference data #1 to #4 are clicked on the illustrated interactive screen, the data contents of reference data #1 to #4 that may be the basis for the answer sentence are changed from hidden to visible, and the user can check the data contents of the selected reference data #1 to #4. In this way, the user can easily check the data contents of the reference data related to the answer sentence, and the user interface for interactive processing with the generative model 60 can be improved.

[0018] Here, the information processing device 100 may be realized by a computing device such as a server, a personal computer (PC), a smartphone, or a tablet, and may have a hardware configuration such as that shown in Fig. 4. That is, the information processing device 100 has a drive device 101, a storage device 102, a memory device 103, a processor 104, a user interface (UI) device 105, and a communication device 106, which are interconnected via a bus B.

[0019] Programs or instructions that realize various functions and processes in the information processing device 100 may be stored in a removable storage medium such as a CD-ROM (Compact Disk-Read Only Memory) or flash memory. When the storage medium is set in the drive device 101, the programs or instructions are installed from the storage medium into the storage device 102 or memory device 103 via the drive device 101. However, the programs or instructions do not necessarily have to be installed from the storage medium, and may be downloaded from any external device via a network or the like.

[0020] The storage device 102 is realized by a hard disk drive or the like, and stores installed programs or instructions as well as files, data, etc. used to execute the programs or instructions.

[0021] The memory device 103 is realized by a random access memory, a static memory, or the like, and when a program or instruction is activated, reads and stores the program, instruction, data, or the like from the storage device 102. The storage device 102, the memory device 103, and the removable storage medium may be collectively referred to as a non-transitory storage medium.

[0022] The processor 104 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc., which may be composed of one or more processor cores, and performs various functions and processes of the information processing device 100 in accordance with programs, instructions, data such as parameters required to execute the programs or instructions, etc. stored in the memory device 103.

[0023] The user interface (UI) device 105 may be composed of input devices such as a keyboard, a mouse, a camera, a microphone, etc., output devices such as a display, a speaker, a headset, a printer, etc., and input / output devices such as a touch panel, and realizes an interface between a user and the information processing device 100. For example, a user operates the information processing device 100 by operating a GUI (Graphical User Interface) displayed on a display or a touch panel using a keyboard, a mouse, etc.

[0024] The communication device 106 is realized by various communication circuits that execute wired and / or wireless communication processing with external devices, the Internet, a LAN (Local Area Network), a cellular network, or other communication networks.

[0025] However, the above-described hardware configuration is merely an example, and the information processing device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.

[0026] [Information processing device] An information processing device 100 according to an embodiment of the present disclosure will be described. Fig. 5 is a block diagram showing a functional configuration of the information processing device 100 according to an embodiment of the present disclosure. As shown in Fig. 5, the information processing device 100 includes a reception unit 110, a prompt creation unit 120, a model interface unit 130, and a display unit 140. For example, one or more functional units of the reception unit 110, the prompt creation unit 120, the model interface unit 130, and the display unit 140 may be realized by one or more processors 104 executing one or more programs or instructions.

[0027] The receiving unit 110 receives question data for the generative model 60. Specifically, when a user inputs a question into an interactive screen for the generative model 60 displayed on the user terminal 20, the receiving unit 110 receives the question input into the interactive screen as question data and provides the received question data to the prompt creating unit 120.

[0028] The prompt creation unit 120 extracts reference data related to question data from the reference database 40 and creates prompt data using the extracted reference data. Specifically, the prompt creation unit 120 performs any known language processing on the question data to extract words included in the question data. Then, the prompt creation unit 120 sends the words extracted from the question data and / or words similar to the extracted words to the reference database 40 as a search query, and extracts reference data related to the question data from the reference database 40. After obtaining reference data corresponding to the search query from the reference database 40, the prompt creation unit 120 creates prompt data to be input to the generative model 60 based on the question data and the reference data related to the question data, for example, according to RAG. Specifically, the prompt data may be configured to include an instruction statement that instructs the generative model 60 to answer the question data by referring to the reference data included in the prompt data when generating answer data for the question data. The prompt creation unit 120 provides the created prompt data to the model interface unit 130.

[0029] The model interface unit 130 inputs prompt data to the generative model 60 and acquires answer data from the generative model 60. Specifically, the model interface unit 130 inputs prompt data created from question data and one or more pieces of reference data related to the question data to the generative model 60. The model interface unit 130 then acquires answer data for the prompt data from the generative model 60 and provides the acquired answer data to the display unit 140.

[0030] Here, the generative model 60 may typically be any large-scale language model, any model that processes other types of data other than text data, such as images, videos, and audio, or a multimodal model that processes multiple types of data.

[0031] The display unit 140 displays the answer data in association with the reference data. Specifically, the display unit 140 displays the answer data on the user terminal 20 together with the reference data used when creating prompt data for the answer data.

[0032] For example, the display unit 140 may hide the data content of the reference data and display the answer data. Then, in response to a user operation, the display unit 140 may display the data content. Specifically, as shown in FIG. 6 , when the display unit 140 receives a question from a user, such as "How is the sales situation of the beverage department?", and acquires an answer from the generative model 60, such as "Sales situation of the beverage department decreased in the fiscal year ending March 2022, with sales reaching 49.7 billion yen (down 3.4% from the previous fiscal year)," the display unit 140 may display the acquired question and answer on the interactive screen, and may also hide the data content of reference data #1 to #4 extracted from the reference database 40 used in the prompt data input to the generative model 60 and embed the data content in the interactive screen.

[0033] For example, if the user wants to check reference data #2, the user clicks on "Reference Data #2" on the interactive screen, and the data content of reference data #2 is displayed on the interactive screen as shown in Figure 7. This allows the user to check the data content of reference database 40 that is the basis for the answer sentence.

[0034] In the illustrated interactive screen, each piece of reference data is displayed in an operable manner using the words "Reference Data #1" to "Reference Data #4." However, the words indicating each piece of reference data are not necessarily limited to these and may be displayed using part of the data content of the reference data, hashtags, etc. For example, in the interactive screen shown in FIG. 6, "Reference Data #2" may be displayed using words that provide an overview of the data content of the reference data, such as "Reference Data #2 #Financial Results Summary for the Fiscal Year Ending March 2021." This allows the user to grasp the overview of the data content of the reference data that is displayed in a hidden state.

[0035] The display unit 140 may also display a link destination for accessing the source file of the reference data. Then, the display unit 140 may display the contents of the source file in response to a user operation on the link destination. As described above, the reference data is a part of a reference file, such as a PDF document file, and is extracted as a chunk from the reference file and converted into text. Therefore, the display unit 140 may display a link to the PDF document file from which the reference data was created on an interactive screen, allowing the user to easily confirm the source file of the reference data.

[0036] For example, as shown in FIG. 8, the display unit 140 may display on the interactive screen a link "dnp_kessan.pdf (p. 7)" to the PDF file that created "Reference Data #2," which has been displayed by a user operation. By clicking on the link, the user can confirm the PDF file that created "Reference Data #2," as shown in FIG. 9. The reference data is converted to text, and it may be difficult for the user to visually recognize the tabular data in the reference data. By displaying the link to the original file, the user can easily confirm the original file by clicking on the link as necessary.

[0037] Here, the display unit 140 may display the first content of the source file. That is, the display unit 140 may display the PDF file from the first page of the PDF file of the link "dnp_kessan.pdf(p.7)".

[0038] Alternatively, the display unit 140 may display the content of the source file corresponding to the reference data. That is, the display unit 140 may display the content portion corresponding to "Reference Data #2" in the PDF file linked to "dnp_kessan.pdf (p. 7)." Furthermore, as shown in FIG. 10, the display unit 140 may emphasize the content of the source file corresponding to the reference data by displaying it in bold, underlining it, highlighting it, or the like. This allows the user to easily confirm the content corresponding to the reference data in the source file.

[0039] To achieve this, each piece of reference data may include metadata indicating which content portion of the source file it corresponds to, or the metadata for each piece of reference data may be stored in a management table that registers the association between each piece of reference data and the metadata. That is, display unit 140 can identify the content of the source file corresponding to the reference data based on the metadata stored in association with the reference data.

[0040] According to the information processing device 100 described above, the user can easily check the data content of the reference data related to the answer sentence, and the user interface for interactive processing with the generative model 60 can be improved.

[0041] [Interaction with Generative Model 60] Next, an interactive process with the generative model 60 according to an embodiment of the present disclosure will be described. The interactive process can be realized by the information processing device 100, more specifically, by the processor 103 of the information processing device 100 executing a program stored in the memory device 102. Fig. 11 is a flowchart showing the interactive process with the generative model 60 according to an embodiment of the present disclosure.

[0042] 11 , in step S101, the information processing device 100 receives question data for the generative model 60. Specifically, when a user inputs a question into an interactive screen for the generative model 60 displayed on the user terminal 20, the information processing device 100 receives question data indicating the question input by the user from the user terminal 20.

[0043] In step S102, the information processing device 100 extracts reference data related to the question data from the reference database 40. Specifically, the information processing device 100 extracts words from the question sentence by applying any known language processing. Then, the information processing device 100 searches the reference database 40 using the extracted words and / or words similar to the extracted words as a search query, and extracts reference data related to the question sentence from the reference database 40.

[0044] In step S103, the information processing device 100 creates prompt data using the extracted reference data. Specifically, the information processing device 100 creates prompt data including question data and reference data extracted from the reference database 40 in response to the search query. The prompt data may include a directive that instructs the generative model 60 to answer the question data by referring to the reference data included in the prompt data when generating answer data to the question data.

[0045] In step S104, the information processing device 100 inputs the prompt data into the generative model 60 and acquires answer data from the generative model 60. The generative model 60 may typically be any large-scale language model, any model that processes other types of data such as images, videos, and audio other than text data, or a multimodal model that processes multiple types of data.

[0046] In step S105, the information processing device 100 displays the answer data in association with the reference data. Specifically, the information processing device 100 displays on the user terminal 20 an interactive screen including not only the answer data acquired from the generative model 60 but also the reference data used in creating the prompt data. At this time, the data content of each reference data may be embedded in the interactive screen, but may be hidden on the interactive screen. In response to a user operation on the interactive screen, the data content of the reference data specified by the user may be displayed.

[0047] The information processing device 100 may also display a link for accessing the source file of the reference data, and display the contents of the source file in response to a user operation on the displayed link. The source file may be displayed from its first page, or the page of the source file corresponding to the reference data may be displayed. The content portion corresponding to the reference data may be emphasized by bolding, underlining, highlighting, or the like, so that the user can easily identify it.

[0048] According to the above-described interactive processing with the generative model 60, the user can easily check the data content of the reference data related to the answer sentence, thereby improving the user interface for interactive processing with the generative model 60.

[0049] Although the examples of the present disclosure have been described in detail above, the present disclosure is not limited to the specific embodiments described above, and various modifications and variations are possible within the scope of the gist of the present disclosure as set forth in the claims. [Explanation of symbols]

[0050] 20 User Device 40 Reference Databases 60 Generative Models 100 Information processing device 110 Reception 120 Prompt Creation Department 130 Model Interface Section 140 Display section

Claims

1. a reception unit that receives query data for the generative model; a prompt creation unit that extracts reference data related to the question data from a reference database and creates prompt data using the extracted reference data; a model interface unit that inputs the prompt data into the generative model and acquires answer data from the generative model; a display unit that displays the response data in association with the reference data; An information processing device having the above.

2. The information processing device according to claim 1 , wherein the display unit displays the response data while hiding the data content of the reference data.

3. The information processing device according to claim 2 , wherein the display unit displays the data content in response to a user operation.

4. The information processing apparatus according to claim 3 , wherein the display unit displays a link destination for accessing a source file of the reference data.

5. The information processing device according to claim 4 , wherein the display unit displays the content of the source file in response to a user operation on the link destination.

6. The information processing device according to claim 5 , wherein the display unit displays a leading content of the source file.

7. The information processing device according to claim 5 , wherein the display unit displays the content of the source file corresponding to the reference data.

8. The information processing device according to claim 5 , wherein the display unit highlights the content of the source file corresponding to the reference data.

9. The information processing device according to claim 5 , wherein the display unit identifies the content of the source file corresponding to the reference data based on metadata stored in association with the reference data.

10. Accepting query data for the generative model; extracting reference data related to the question data from a reference database, and generating prompt data using the extracted reference data; inputting the prompt data into the generative model and obtaining answer data from the generative model; displaying the response data in association with the reference data; An information processing method executed by a computer.

11. Accepting query data for the generative model; extracting reference data related to the question data from a reference database, and generating prompt data using the extracted reference data; inputting the prompt data into the generative model and obtaining answer data from the generative model; displaying the response data in association with the reference data; A program that causes a computer to execute the following.

Images