Information processing apparatus, information processing method, and program

Abstract

To provide a technique for efficiently classifying the details of descriptions in document data by using a generation model.SOLUTION: An aspect of the present disclosure relates to an information processing apparatus that has: an acquisition unit that acquires document data; a prompt generation unit that generates prompt data for classifying sentences in the document data into predetermined categories; and a classification unit that inputs the prompt data to a generation model and classifies the sentences on the basis of a result of output from the generation model.SELECTED DRAWING: Figure 1

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. Currently, natural language processing using generative models such as ChatGPT (registered trademark) is attracting attention.

[0003] For example, Patent Document 1 describes the use of a generation AI (Artificial Intelligence) system to crawl job posting data providing websites on the Internet and generate job posting data by formatting the original job posting data obtained from these websites. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7349219 Summary of the Invention [Problem to be solved by the invention]

[0005] Many companies create business manuals to ensure that employees perform their work properly, and have employees follow the manuals to carry out their work. Such business manuals are often created for each business department, and multiple business manuals often contain the same or similar content.

[0006] When consolidating and integrating multiple documents such as business manuals, workers currently have to visually inspect the documents and categorize the content of the sentences that make up the documents. This type of content classification is a costly manual process.

[0007] An object of the present disclosure is to provide a technology for efficiently classifying the content of document data using a generative model. [Means for solving the problem]

[0008] One aspect of the present disclosure relates to an information processing device having an acquisition unit that acquires document data, a prompt generation unit that generates prompt data for classifying sentences in the document data into predetermined categories, and a classification unit that inputs the prompt data into a generation model and classifies the sentences based on the output results from the generation model. [Effects of the Invention]

[0009] According to the present disclosure, a technique can be provided for efficiently classifying the content of document data using a generative model. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic diagram illustrating a text classification process according to one embodiment of the present disclosure. [Figure 2] FIG. 2 is a schematic diagram illustrating an information processing device according to an embodiment of the present disclosure. [Figure 3] FIG. 3 is a block diagram illustrating a hardware configuration of an information processing device according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating a functional configuration of an information processing device according to an embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram illustrating prompt data according to one embodiment of the present disclosure. [Figure 6] FIG. 6 is a diagram illustrating output results from a generative model according to one embodiment of the present disclosure. [Figure 7] FIG. 7 is a diagram illustrating the classification scores of each sentence according to one embodiment of the present disclosure. [Figure 8] FIG. 8 is a flowchart illustrating a text classification process according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0012] In the following embodiment, an information processing device for classifying text content is disclosed.

[0013] [Summary of the Disclosure] An information processing device 100 according to an embodiment described below determines into which category the content of a sentence to be processed in the document data can be classified when each piece of document data is composed of multiple sentences with a coherent meaning. Specifically, when the information processing device 100 acquires the document data, it extracts the sentence to be processed from the document data and generates prompt data for causing a generative model 50 to classify the extracted sentence into a predetermined category. The information processing device 100 then inputs the generated prompt data into the generative model 50 and determines the category of the sentence to be processed based on the output result from the generative model 50.

[0014] For example, when the information processing device 100 acquires document data of an "XX Insurance Application Manual" as shown in Fig. 1, it extracts the contents of "identification" as a sentence to be processed. For example, the sentence to be processed may be automatically extracted based on a heading, a table of contents, a document layout, etc., or may be specified by a user.

[0015] Once the sentence to be processed is identified, the information processing device 100 generates prompt data for causing the generative model 50 to classify the sentence to be processed into a predetermined category. In the example shown in FIG. 1, the prompt data first includes an instruction statement such as "Please classify the following sentence into specified categories. Please also output a score for each category in the range of 0 to 1.", followed by the sentence to be processed. Furthermore, the prompt data specifies six predetermined categories: "1. Data creation, 2. Confirmation of submitted documents, 3. Confirmation of application details, 4. Reception, 5. Review, and 6. Sending of contract-related notifications." Note that the prompt data shown in the figure is merely an example and is not limited thereto.

[0016] When the prompt data is generated, the information processing device 100 inputs the generated prompt data to the generative model 50 and obtains an output result from the generative model 50. For example, when the prompt data shown in FIG. 1 is input to the generative model 50, the information processing device 100 obtains an output result as shown in FIG. 1 and classifies the sentence to be processed based on the output result from the generative model 50. In the example shown in the figure, it is indicated that the classification score for "2. Confirm submitted documents" is the highest for the sentence to be processed, and the information processing device 100 can classify the sentence to be processed into the category of "2. Confirm submitted documents."

[0017] 2, when the information processing device 100 acquires document data, it extracts sentences to be processed from the acquired document data and generates prompt data for classifying the extracted sentences in the generative model 50. Then, the information processing device 100 inputs the generated prompt data to the generative model 50 and acquires an output result from the generative model 50. The information processing device 100 classifies the sentences to be processed based on the acquired output result.

[0018] In the illustrated embodiment, the generative model 50 may be, for example, a language model used in ChatGPT (registered trademark) or the like, and is typically operated and managed by a server or the like external to the information processing device 100. However, the present disclosure is not limited to this, and, for example, the generative model 50 may be stored and executed in the information processing device 100.

[0019] In this way, the information processing device 100 can use the generative model 50 to classify the contents of document data into predetermined categories.

[0020] 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, for example, a hardware configuration as shown in Fig. 3. 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.

[0021] 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.

[0022] 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.

[0023] 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.

[0024] 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.

[0025] 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.

[0026] 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.

[0027] 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.

[0028] [Information processing device] Fig. 4 is a block diagram showing a functional configuration of an information processing device 100 according to an embodiment of the present disclosure. As shown in Fig. 4, the information processing device 100 includes an acquisition unit 110, a prompt generation unit 120, and a classification unit 130. Each of the functional units of the acquisition unit 110, the prompt generation unit 120, and the classification unit 130 may be realized by the processor 104 executing a computer program stored in a memory device 103 of the information processing device 100.

[0029] The acquiring unit 110 acquires document data. Here, the document data is not limited to the business manual as described above, but may be any document data including text to be classified. The document data here is not limited to a file format such as a WORD file, but may be electronic data in any file format including text, such as text data, HTML (HyperText Markup Language) data, Excel files, PowerPoint files, or PDF (Portable Document Format) files. Furthermore, the document data may include not only characters such as alphanumeric characters, but also diagrams, images, and the like.

[0030] The prompt generation unit 120 generates prompt data for classifying sentences in the document data into predetermined categories. Specifically, the prompt generation unit 120 extracts sentences to be processed from the document data and generates prompt data for classifying the extracted sentences in the generation model 50.

[0031] In one embodiment, the prompt generator 120 may extract each sentence from the document data based on the layout of the document data. Generally, document data is composed of multiple sentences that are structured (layouted) by, for example, dividing the sentences into chapters based on the unity of meaning, and each sentence is provided with a heading or the like to indicate that it is a coherent description. The prompt generator 120 can identify the heading and layout of such document data using any known heading / layout extraction technique and extract each sentence from the document data.

[0032] When the text to be processed is identified, the prompt generation unit 120 generates prompt data for classifying the identified text into a predetermined category in the generative model 50. For example, when a description related to "identity verification" is identified as the text to be processed, the prompt generation unit 120 may generate prompt data that describes an instruction for classifying the document into a category, the text to be processed, the predetermined categories, an explanation of each category, words related to each category, and an example output format, as shown in FIG.

[0033] 5, the instruction statement is "Please divide the following sentences into specified categories. Please output a score for each category in the range of 0 to 1. Note: Be sure to output a score for all categories. Be sure to also clearly state the reason." This instruction statement allows the generative model 50 to recognize the instruction content in the prompt data.

[0034] Next, as the sentence to be processed, a copy of the sentence in the document data is written: "### sentence: (1) Identity verification Identity verification will be carried out based on the identity verification document submitted by the contractor. Please select from the pull-down menu in the identity verification field on the [Application Data Entry] screen. Examples of identity verification documents are as follows: driver's license, passport, My Number card." The "### sentence" can clearly specify the "sentence" in the instruction.

[0035] Next, enter the specified category: "###Category: 1. Data Creation 2. Confirmation of Submitted Documents" 3. Confirmation of application details 4. Reception 5. Review 6. Sending of contract-related notifications." The "### category" can be used to clearly define the "category" of the instruction.

[0036] Next, the following is written as an explanation of each category: "###Specific meaning of each category 1. Data creation: This describes specific actions such as operating the application database, input procedures, and how to write information. 2. Confirmation of submitted documents: This describes confirmation of documents such as application forms and identification cards. 3. Confirmation of application content: This explains how to confirm application data. 4. Reception: This explains the acceptance procedures for screening, and the specific steps involved. 5. Screening: This describes details regarding whether the insurance contract is approved. 6. Sending contract-related notifications: This describes how to send notifications to applicants, such as the screening results and contract terms." This description clearly defines the specific meaning of each category, which can improve the accuracy of classification.

[0037] Next, the following words are written as words related to each category: "### Words related to each category. If a sentence contains the following words or their synonyms, it is more likely to be classified into the category to which the words belong. 1. Data creation: application data, input, operation, database 2. Submitted document confirmation: application form, certificate, health certificate, authenticity confirmation, name 3. Application content confirmation: name, age, occupation, plan, period, amount, area, consent 4. Reception: reception number, reception procedure 5. Screening: applicant information review, risk assessment, health status, ability to pay 6. Delivery of contract-related notices: notice, guidance, insurance policy." In other words, the prompt generation unit 120 may write a list of keywords related to each category in the prompt data. This description clearly indicates what words and terms are used in each category, thereby improving classification accuracy.

[0038] Finally, as an output format example, "### Output format example 1. Data creation: 0.0 2. Confirmation of submitted documents: 1.0 3. Confirmation of application details: 0.0 4. Reception: 0.0 5. Review: 0.0 6. Sending of contract-related notification: 0.0 Reason: ○○○" is described. This description can specify the format of the output result from the generative model 50.

[0039] The prompt generation unit 120 generates prompt data as shown in Figure 5 and passes the generated prompt data to the classification unit 130. The prompt generation unit 120 may also input each sentence to be processed into the "### sentence~" portion of the prompt data described above, and process multiple sentences in the sentence data. Note that the prompt data shown in the figure is merely an example, and the prompt data may be composed of descriptive content in any other appropriate format.

[0040] Here, there are cases where the document data includes non-text data such as figures, tables, and images. In such cases, the prompt generation unit 120 may convert the non-text data in the document data into text data and write the text data in the prompt data. Specifically, the prompt generation unit 120 may use any known conversion tool, such as OCR or image recognition AI, to convert figures, tables, images, and the like into text data and include the text data in the sentence to be processed. Note that if the generative model 50 is a multimodal-type generative model capable of processing data in formats other than text data, the prompt generation unit 120 may generate prompt data that directly includes non-text data such as figures, tables, and images as the sentence to be processed.

[0041] The classification unit 130 inputs prompt data to the generative model 50 and classifies sentences based on the output result from the generative model 50. For example, when prompt data such as that shown in FIG. 5 is acquired from the prompt generation unit 120, the classification unit 130 inputs the acquired prompt data to the generative model 50. At this time, the generative model 50 generates an output result such as that shown in FIG. 6, and the classification unit 130 acquires the output result from the generative model 50. The output result describes a classification score for each category, along with the reason for determining the classification score.

[0042] The classification score here may indicate, for example, the probability (a real number between 0 and 1) that the text to be processed will be classified into the category. For example, it can be interpreted that the closer the probability is to 0, the less likely the text to be processed will belong to the category, and the closer the probability is to 1, the more likely the text to be processed will belong to the category. Therefore, according to the output result shown in FIG. 6, it is determined that the text to be processed is highly likely to fall under "2. Confirm submitted documents."

[0043] Upon receiving such an output result, the classification unit 130 may display the category with the highest classification score as the classification result. Alternatively, the classification unit 130 may display the classification score for each category as the classification result. For example, the classification unit 130 may present a combination of the text to be processed and the classification results for a predetermined category. For example, the classification unit 130 may format and display the classification results in a table format as shown in FIG. 7. According to the classification results displayed in the illustrated table format, for the document to be processed, "(1) Identity Verification...", the classification unit 130 may input the classification score for each category based on the output result shown in FIG. 6. In the classification result shown in FIG. 7, a classification score of "0.0" is input for each of the categories "Data Creation," "Application Content Confirmation," "Reception," "Screening," and "Sending Contract-Related Notification," and a classification score of "0.9" is input for the category "Submitted Document Confirmation." Furthermore, in this case, the classification unit 130 may display the classification score for each category for the text in a display format corresponding to the score value of the classification score. For example, the classification unit 130 may visually highlight the classification scores by coloring or grayscaling them according to the magnitude of the classification scores.

[0044] In one embodiment, the classifier 130 may classify a document based on a categorization result of the document using a categorization method. That is, the classifier 130 may classify a document to be processed by using both the output result from the generative model 50 and the categorization result from any known categorization method using natural language processing, such as TF-IDF (Term Frequency-Inverse Document Frequency) or a support vector machine.

[0045] For example, the classification unit 130 may classify the text to be processed based on the average value of a first classification score based on the result of categorizing the text using a categorization method such as TF-IDF or a support vector machine, and a second classification score based on the output result from the generative model 50. Specifically, the classification unit 130 derives a first classification score based on the result of categorizing the text to be processed using any categorization method such as TF-IDF or a support vector machine, and derives a second classification score based on the output result from the generative model 50. The classification unit 130 may then average the two derived classification scores for each category and determine the calculated average value for each category as the classification score for each category. This allows the text to be classified in consideration of the two classification results from any categorization method and the generative model 50.

[0046] Alternatively, the classification unit 130 may calculate a first classification score of a sentence according to a categorization method, and when the calculated first classification score is equal to or greater than a predetermined threshold, input prompt data for classifying the document into a predetermined category to the generative model 50. That is, the classification unit 130 may first classify the sentence to be processed according to a categorization method such as TF-IDF or a support vector machine, and when the classification score obtained from the classification result is equal to or greater than a predetermined threshold, perform category classification using the generative model 50. This allows further confirmation of the classification result obtained by the categorization method.

[0047] [Text classification processing] Next, a text classification process according to an embodiment of the present disclosure will be described with reference to Fig. 8. Fig. 8 is a flowchart showing the text classification process according to an embodiment of the present disclosure. The text classification process may be executed by, for example, an information processing device 100, and more specifically, may be realized by the processor 104 of the information processing device 100 executing a computer program or instructions stored in the memory device 103. The text classification process may also be realized by multiple information processing devices 100 that cooperate with each other via a network to execute each step.

[0048] In step S101, the information processing device 100 acquires document data. For example, the document data may be provided by a user or extracted from a database.

[0049] In step S102, the information processing device 100 generates prompt data for classifying sentences in the document data into predetermined categories. Such prompt data is typically generated as text data, and the text data may be configured to include the sentences to be processed, each category and / or their descriptions, and instructions for classifying the sentences into categories. Furthermore, the prompt data may be configured to include a list of keywords related to each category.

[0050] In step S103, the information processing device 100 inputs prompt data to the generative model 50. For example, if the generative model 50 is operated by a server external to the information processing device 100, the information processing device 100 may transmit the prompt data to the server. Alternatively, if the generative model 50 is stored within the information processing device 100, the information processing device 100 may input the prompt data to the generative model 50 and execute the generative model 50.

[0051] In step S104, the information processing device 100 acquires the output result from the generative model 50.

[0052] In step S105, the information processing device 100 classifies the sentences based on the output result. For example, the information processing device 100 may format the classification result as tabular data consisting of headings and / or chapter items of each sentence in the document data and classification scores of each category, as shown in Fig. 7, and display the classification result.

[0053] According to the above-described embodiment, the generative model 50 can be used to efficiently classify the contents of each sentence constituting document data into predetermined categories.

[0054] 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]

[0055] 50 Generative Models 100 Information processing device 110 Acquisition Department 120 Prompt Generation Unit 130 Classification Department

Claims

1. an acquisition unit that acquires document data; a prompt generation unit that generates prompt data for classifying sentences in the document data into predetermined categories; a classification unit that inputs the prompt data into a generative model and classifies the sentence based on an output result from the generative model; An information processing device having the above.

2. The information processing apparatus according to claim 1 , wherein the prompt generating unit extracts each sentence in the document data based on a layout of the document data.

3. The information processing device according to claim 1 , wherein the classification unit determines a classification score for each category for the sentence.

4. The information processing device according to claim 1 , wherein the classification unit presents a combination of the sentence and the classification result of the predetermined category.

5. The information processing device according to claim 3 , wherein the classification unit displays the classification scores of the sentences in each category in a display format corresponding to the score values ​​of the classification scores.

6. The information processing apparatus according to claim 1 , wherein the prompt generating unit describes a list of keywords related to each category in the prompt data.

7. The information processing apparatus according to claim 1 , wherein the prompt generating unit converts non-text data in the document data into text data, and describes the text data in the prompt data.

8. The information processing device according to claim 1 , wherein the classification unit classifies the text further based on a result of categorizing the text using a categorization technique.

9. The information processing device according to claim 8 , wherein the classification unit classifies the text based on an average value of a first classification score based on the category classification result and a second classification score based on the output result from the generative model.

10. 9. The information processing device according to claim 8, wherein the classification unit calculates a first classification score of the sentence according to the category classification technique, and inputs the prompt data to the generative model when the calculated first classification score is equal to or greater than a predetermined threshold.

11. Obtaining document data; generating prompt data for classifying sentences in the document data into predetermined categories; inputting the prompt data into a generative model and classifying the sentence based on an output from the generative model; A computer-implemented information processing method comprising:

12. Obtaining document data; generating prompt data for classifying sentences in the document data into predetermined categories; inputting the prompt data into a generative model and classifying the sentence based on an output from the generative model; A program that causes a computer to execute the following.

Images