Information providing device, method, and program

Abstract

A system, method, and program for providing information are capable of outputting more accurate output sentences and related reference-attached explanatory sentences in response to input sentences such as questions. [Solution] Based on the reference concept data 112 corresponding to the text data of the arcuate member 130 extracted by the concept extraction unit 103, reference expert knowledge data 116 is acquired, and then text data 118 of an explanatory text with references is acquired based on access to the generation AI system 108, thereby providing highly accurate answers, advice, explanations, etc. for the input text.

Description

[Technical Field]

[0001] The present invention relates to a system, method, and program for providing information. [Background technology]

[0002] Conventionally, there is known an information providing device that provides a means for providing accurate and easy-to-understand information for a variety of questions in the field of a target service, by comprising a question receiving unit configured to receive questions regarding labor, etc., an answer generating unit configured to cause a Generative Pre-trained Transformer (GPT) (a generative artificial intelligence system) to generate answers to the questions, a diagram etc. acquiring unit configured to acquire diagrams etc. related to the questions from the Web, and a report generating unit configured to generate a report including the answers generated by the answer generating unit and the diagrams etc. acquired by the diagram etc. acquiring unit (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7519138 Registration Publication Summary of the Invention [Problem to be solved by the invention]

[0004] However, simply having a generative artificial intelligence system generate answers to questions and searching for relevant charts and graphs on the Internet, etc., has not been able to provide sufficiently accurate answers to questions that require specialized knowledge, such as those related to human resources and labor. For example, when answering questions about specialized knowledge, it was difficult to accurately point to Supreme Court precedents or labor-related articles.

[0005] Therefore, an object of the present invention is to enable output of more accurate output sentences and related reference-attached explanatory sentences in response to input sentences such as questions. [Means for solving the problem]

[0006] An information providing device according to one embodiment includes a specialized knowledge database that stores specialized knowledge data, which is data related to specialized knowledge, as a database, a concept database that stores, for each concept constituting the specialized knowledge, concept vector values ​​that are vector values ​​corresponding to the concept, concept generation phrases that generated the concept, and one or more specialized knowledge identifiers that link and identify the concept with one or more specialized knowledge data in the specialized knowledge database, and a reference concept data processing unit that inputs text data of an input sentence related to the specialized knowledge, generates input vector values ​​that are vector values ​​corresponding to words or sentences that constitute the input sentence, and retrieves one or more concept data having concept vector values ​​similar to the input vector value from the concept database. a specialized knowledge data acquisition unit that acquires one or more pieces of specialized knowledge data corresponding to each of the specialized knowledge identifiers included in each piece of reference concept data from a specialized knowledge database as reference specialized knowledge data; a referenced explanatory text acquisition unit that acquires text data of an explanatory text explaining the reference specialized knowledge data by providing a reference prompt that instructs an explanation of the referenced specialized knowledge data to the generative artificial intelligence system, and acquires text data of a referenced explanatory text including the explanatory text and a reference to each of one or more pieces of specialized knowledge data in the specialized knowledge database that correspond to the referenced specialized knowledge data; and a display unit that outputs and displays the text data of the input sentence and the text data of the referenced explanatory text on a user's terminal device. [Effects of the Invention]

[0007] According to the present invention, it is possible to output a more accurate output sentence and an explanation sentence with related references in response to an input sentence such as a question. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 2 is a functional block diagram of an embodiment of an information providing device. [Figure 2] FIG. 2 is a functional block diagram of a concept extraction unit. [Figure 3] FIG. 10 is a diagram illustrating an example of the configuration of a specialized knowledge database (personnel and labor). [Figure 4] FIG. 10 is a diagram illustrating an example of the configuration of a conceptual database (personnel and labor). [Figure 5] 1 is a diagram illustrating the operation of the concept extraction unit. [Figure 6] FIG. 10 is a diagram showing a display example. [Figure 7] FIG. 2 is a hardware block diagram illustrating an example of the configuration of a computer that executes an embodiment of the information providing device. [Figure 8] 10 is a flowchart illustrating an example of a control operation of the information providing device. [Figure 9] 10 is a flowchart illustrating an example of a generation AI API access process. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. Fig. 1 is a functional block diagram of an information providing device 100 according to an embodiment of the present invention. The information providing device 100 is, for example, a server computer connected to a network 119 such as the Internet or a local network.

[0010] First, the information providing device 100 includes a specialized knowledge database 101. This specialized knowledge database 101 stores specialized knowledge data, which is data related to specialized knowledge such as personnel and labor management, as a database. FIG. 3 is a diagram showing an example of the configuration of the specialized knowledge database 101 when the specialized knowledge is, for example, personnel and labor management. The specialized knowledge data stored in the specialized knowledge database 101 is, for example, one or more of the following: Supreme Court precedent data 301 related to specialized knowledge of personnel and labor management; article and notice data 302 related to the specialized knowledge; related form data 303 of laws related to the specialized knowledge; news data 304 related to the specialized knowledge; web data 305 related to the specialized knowledge that is publicly available on the Internet; and practical information or message information 306 from emails, SNS, etc. related to the specialized knowledge.

[0011] Next, the information providing device 100 includes a concept database 102. FIG. 4 is a diagram showing an example of the data structure of the concept database 102. As shown in FIG. 4, the concept database 102 stores concept data 401 for each concept constituting specialized knowledge. The concept data 401 includes at least a concept vector value 402, which is a vector value corresponding to the concept, a concept-generating phrase 403, which is a word that generated the concept or a sentence in which words are strung together syntactically, and one or more specialized knowledge identifiers 404 indicating one or more specialized knowledge data 301, 302, ..., 306 (see FIG. 3) in the specialized knowledge database 101 that includes the concept. The concept data 401 may also include a concept summary 405 that explains the concept.

[0012] Moreover, the concept data 401 may further include parent link information 406 indicating a link to another concept data 401 that is conceptually in a parent relationship with the concept data 401 . More specifically, this parent link information 406 may be manually assigned. Alternatively, the concept data 401 may include, as parent link information 406, a link to another concept data 401 that has all of the concept-generating phrases 403 included in the concept data 401 as part of the other concept-generating phrases 403.

[0013] The concept extraction unit 103 inputs text data 110 of an input sentence related to specialized knowledge, generates input vector values ​​111, which are vector values ​​corresponding to the words or sentences that make up the input sentence, and extracts one or more concept data 401 (see Figure 4) having concept vector values ​​402 similar to the input vector values ​​111 from the concept database 102 as reference concept data 112.

[0014] In this case, the concept extraction unit 103 may extract, as reference concept data 112, one or more conceptual data 401 having a conceptual vector value 402 similar to the input vector value 111, as well as conceptual data 401 linked to the conceptual data 401 by parent link information 406 contained in the conceptual data 401. In the example of Figure 4, for example, when the concept extraction unit 103 extracts concept data 401(#1), concept data 401(#0) may also be extracted based on link A indicated by the parent link 406 included in the concept data 401(#1). For example, when the concept extraction unit 103 extracts concept data 401(#2), concept data 401(#1) may also be extracted based on link B indicated by the parent link 406 included in the concept data 401(#2), and further concept data 401(#0) may also be extracted based on link A indicated by the parent link 406 included in the concept data 401(#1).

[0015] Fig. 2 is a functional block diagram of the concept extraction unit 103 in Fig. 1. First, the morphological analysis unit 200 performs morphological analysis on the text data 110 of the input sentence, thereby separating the text data 110 of the input sentence into words. Fig. 5 is a diagram illustrating the operation of the concept extraction unit 103. For example, suppose the text data 110 of the input sentence is a question such as "An employee who left on good terms contacted me to ask me to pay his unpaid overtime wages." The morphological analysis unit 200 performs morphological analysis on the text data 110 of this input sentence, thereby separating the text into words as shown in 501 in FIG.

[0016] Next, the word / vector value conversion machine learning model unit 201 in FIG. 2 inputs the text data of each significant word (e.g., word 502 with a dark background color among 501 in FIG. 5) obtained by separating the text data 110 of the input sentence output from the morphological analysis unit 200, excluding particles, conjunctions, etc., and outputs a word vector value 210, which is a vector value corresponding to the word, such as V1, V2, V3, V4, V5, or V6 in FIG. 5, by inputting the text data of the word.

[0017] The word / vector value conversion machine learning model unit 201 operates based on, for example, the Word2vec (registered trademark) algorithm (patented) created and published by Google LLC, USA. The Word2vec algorithm is a series of models used to generate word embeddings. These models are shallow two-layer neural networks trained to reconstruct the linguistic context of words, and receive a large corpus to generate a vector space. This vector space typically has several hundred dimensions, and each word in the corpus is assigned to an individual vector in the vector space. Word vectors that share the same context in the corpus have close vector distances to each other in the vector space.

[0018] 5, the vector synthesis unit 202 extracts sentences in which words are linked together syntactically by syntactically analyzing the dependency relationships between words in the input sentence, and outputs a sentence vector value 211 by synthesizing each of the word vector values ​​210 extracted by the word / vector value conversion machine learning model unit 201 corresponding to each of the words that make up the sentence. Synthesis here refers to, for example, adding up the element values ​​of vectors. Finally, all synthesized vectors may be added up to calculate an even larger sentence vector value 211. The word vector values ​​210 output from the word / vector value conversion machine learning model unit 201 and the sentence vector values ​​211 output from the vector synthesis unit 202 become the input vector values ​​111 output by the concept extraction unit 103 in Fig. 1. As a result, for each significant word indicated by 502 in Fig. 5, the distance between each of the word vector values ​​210 of that word, such as V1, V2, V3, V4, V5, or V6 in Fig. 5, and each conceptual vector value 402 included in each conceptual data 401 in the conceptual database 102 described in Fig. 4 is calculated, and the higher-level conceptual data 401 with the closest distance is extracted as data indicating the concept of that word. Furthermore, for each sentence indicated by 503 in FIG. 5, the distance between each of the sentence vector values ​​211 of that sentence, such as V12, V34, V56, or Vs in FIG. 5, and each conceptual vector value 402 included in each conceptual data 401 in the conceptual database 102 described in FIG. 4 is calculated, and the higher-level conceptual data 401 having the closest distance is extracted as data indicating the concept of that sentence.

[0019] 2, the word / vector value conversion machine learning model training unit 203 trains the word / vector value conversion machine learning model unit 201 using a corpus of text data of specialized knowledge as training data 212. This is a typical machine learning process. The corpus of text data on specialized knowledge used as the training data 212 may be, for example, Supreme Court precedent data, article and notice data related to specialized knowledge, format data related to laws related to specialized knowledge, news data related to specialized knowledge, web data published on the Internet related to specialized knowledge, or practical information related to specialized knowledge or message information such as emails and SNS. In addition, the corpus of Wikipedia (a registered trademark of the Wikimedia Foundation, USA) may be used as the corpus of the training data 212.

[0020] After the learning is completed by the word / vector value conversion machine learning model learning unit 203, the concept database creation unit 204 stores the learning data 212 in the specialized knowledge database 101 of FIG. 1 and creates, for each word constituting the learning data 212, conceptual data 401 (see FIG. 4 ) in the conceptual database 102, which includes, as a conceptual vector value 402, the word vector value 210 output by inputting the text data of the word into the word / vector value conversion machine learning model unit 201, a phrase corresponding to the word as a concept generation phrase 403, and information linking and identifying the learning data 212 stored in the specialized knowledge database 101 as a specialized knowledge identifier 404. At the same time, the conceptual database creation unit 204 creates conceptual data 401 in the conceptual database 102, which includes, as conceptual vector values ​​402, the sentence vector values ​​211 output from the vector synthesis unit 202 corresponding to the learning data 212, and as concept-generating phrases 403, words corresponding to the sentences, and which includes, as specialized knowledge identifiers, information that links and identifies the learning data 212 stored in the specialized knowledge database 101.

[0021] Here, the specialized knowledge identifier 404 may be, for example, a record identifier of the learning data 212 stored as specialized knowledge data in the specialized knowledge database 101. When the specialized knowledge in the specialized knowledge database 101 corresponding to the concept data 401 is referenced, the specialized knowledge of the record having the same record identifier as the specialized knowledge identifier 404 in the concept data 401 is referenced. Alternatively, instead of providing a separate expert knowledge identifier 404, the conceptual vector value 402 may also serve as the expert knowledge identifier 404. In this case, the learning data 212 may be stored as expert knowledge data in the expert knowledge database 101, and a vector value identical to the conceptual vector value 402 of the corresponding conceptual data 401 may be stored as part of the expert knowledge data. When expert knowledge in the expert knowledge database 101 corresponding to the conceptual data 401 is referenced, the expert knowledge of a record in which the same vector value as the conceptual vector value 402 in the conceptual data 401 is stored may be referenced.

[0022] Returning to the explanation of Figure 1, the output sentence acquisition unit 104 acquires text data 114 of the output sentence from the generative AI system 108, which is a generative artificial intelligence system, by issuing instructions to the generative AI system 108 using an input prompt 113 composed of concept generation terms such as words and sentences that have generated the reference concept data 112 extracted by the concept extraction unit 103 and text data 110 of the input sentence. The input prompt 113 includes, for example, content instructing the user to respond with text data 114 of the output sentence, such as a main topic, summary, answer, advice, or explanatory text data pointing out points to note or problems, in response to the text data 110 of the input sentence. Examples of the generative AI system 108 include ChatGPT, GPT-3, GPT-4, and GPT-4o (each a registered trademark of OpenAI, Inc., USA), Gemini (a registered trademark of Google LLC, USA), and Claude (a registered trademark of Anthropic, PBC, USA). The generative AI system 108 may also operate by inputting a prompt to each of a plurality of sub-generative AI systems, obtaining a response from each of the sub-generative AI systems, and outputting an appropriate response based on the respective responses.

[0023] The text data 110 of the input sentence is, for example, news information related to specialized knowledge such as human resources and labor, and the corresponding text data 114 of the output sentence acquired by the output sentence acquisition unit 104 is, for example, text data representing the subject (or summary) of the news information. Furthermore, the text data 110 of the input sentence is, for example, question information regarding specialized knowledge such as personnel and labor matters sent from a user operating the user terminal device 209 in Figure 1, and the text data 114 of the output sentence acquired by the output sentence acquisition unit 104 in response thereto is, for example, summary information and answer information for the question information. The text data 110 of the input sentence is, for example, notice information acquired from the website of a government or related organization regarding specialized knowledge via the Internet 119 in FIG. The text data 110 of the input sentence is, for example, information on in-house applications, contracts, various documents, or formats related to specialized knowledge obtained by operating the user terminal device 209 in FIG. Furthermore, the text data 110 of the input sentence is, for example, information on various in-house rules related to specialized knowledge obtained by operating the user terminal device 209 in FIG.

[0024] The expert knowledge data acquisition unit 105 acquires, from the expert knowledge database 101, one or more pieces of expert knowledge data 301, 302, ..., 306 (see Figure 3) corresponding to each of the expert knowledge identifiers 404 included in each of the reference concept data 112 (hereinafter collectively referred to as the "reference expert knowledge identifiers 115" shown in Figure 1). In this case, one or more pieces of expert knowledge data are obtained as the reference expert knowledge data 116 from one or more records corresponding to one or more record identifiers corresponding to one or more expert knowledge identifiers 404 that are the reference expert knowledge identifiers 115 . As described above, the reference expertise identifier 115 may be a concept vector value 402 that also serves as the expertise identifier 404, rather than the expertise identifier 404. In this case, one or more pieces of expertise data are obtained as the reference expertise data 116 from one or more records each including one or more concept vector values ​​402 that are the reference expertise identifier 115.

[0025] The referenced commentary acquisition unit 106 provides a reference prompt 117 instructing an explanation of the referenced expert knowledge data 116 acquired by the expert knowledge data acquisition unit 105 to the generative AI system 108, which is a generative artificial intelligence system, thereby acquiring text data of a commentary that explains the referenced expert knowledge data 116 from the generative AI system 108, and acquiring text data 118 of a referenced commentary that includes the commentary and references to one or more expert knowledge data in the expert knowledge database 101 that correspond to the referenced expert knowledge data 116. The generation AI system 108 may be, for example, the same service as that used in the output sentence acquisition unit 104. Alternatively, it may be a different service. Also, as in the case of the output sentence acquisition unit 104, the generative AI system 108 may operate by inputting the prompt input to each of multiple sub-generative AI systems, thereby obtaining a response from each of the sub-generative AI systems, and outputting an appropriate response based on each of those responses.

[0026] The display unit 107 outputs and displays the input sentence text data 110, the output sentence text data 114, and the referenced commentary text data 118 on a display (not shown) of the user terminal device 109, for example, via the network 119. FIG. 6 is a diagram showing an example of a display displayed by the display unit 107 on the display of the user terminal device 109. The display area 601 displays text data, such as a consultation summary, acquired by the output sentence acquisition unit 104 of FIG. 1. The display area 602 displays the referenced commentary text data 118 acquired by the referenced commentary acquisition unit 106 of FIG. 1. The display area 603 displays reference information such as related legal precedents, explanations of the provisions, recent related news, and related form collections. The user of the user terminal device 109 can understand the content of the consultation by reading the commentary text in the display area 602, and can tap the reference button in the display area 603 to learn examples that can further deepen their knowledge and understanding.

[0027] Fig. 7 is a hardware block diagram showing an example of the configuration of a computer that executes the functions of the information providing device 100 in Fig. 1. This hardware comprises a server computer configuration in which a CPU (Central Processing Unit) 701, a memory 702 into which programs are loaded and executed, an external storage device 703 that stores programs, databases such as the expert knowledge database 101 and concept database 102 in Fig. 1, and other data, and a network interface circuit 704 that controls access to the Internet 120 in Fig. 1 are all interconnected by a system bus 705. Although not specifically shown, input devices such as a keyboard and a mouse and output devices such as a display may also be connected.

[0028] Fig. 8 is a flowchart showing an example of a control operation for realizing the functions of the information providing device 100 in Fig. 1. This flowchart shows a process in which the CPU 701 in Fig. 7 executes a control operation program written in, for example, the Python computer programming language ("Python" is a registered trademark of the Python Software Foundation, USA), which has been loaded from the external storage device 703 into the memory 702.

[0029] First, the CPU 701 sets a library for accessing the generation AI API and API authentication information as initial settings for the generation AI system 108 (step S801). This library is, for example, a ChatGPT API library provided by Python. This library is, for example, pre-installed in the external storage device 703 of Fig. 7. In step S801, the CPU 701 executes a process of importing the library from the external storage device 703 of Fig. 7 to the memory 702, for example, by executing an "import openai" command. In addition, in step S801, the CPU 701 executes, at the beginning of the program, a command to set API authentication information data acquired by registering in advance with, for example, the ChatGPT service.

[0030] Next, in step S802, the CPU 701 executes the concept extraction process described as the function of the concept extraction unit 103 in Fig. 1. As a result, the CPU 701 inputs text data 110 of an input sentence related to specialized knowledge, as described in Fig. 1, generates input vector values ​​111 which are vector values ​​corresponding to words or sentences that make up the input sentence, and extracts one or more pieces of concept data 401 having concept vector values ​​402 similar to the input vector values ​​111 from the concept database 102 as reference concept data 112.

[0031] Next, the CPU 701 executes the output sentence acquisition process described as a function of the output sentence acquisition unit 104 in Fig. 1. Specifically, the CPU 701 calls a program for the generation AI API access process using, as arguments, the query data of the input prompt 113 configured from the concept generation phrases such as words and sentences that generated the reference concept data 112 extracted in the concept extraction process in step S802 and the text data 110 of the input sentence (step S803). This process will be described later using the flowchart in Fig. 9.

[0032] Next, the CPU 701 executes the expert knowledge data acquisition process described as a function of the expert knowledge data acquisition unit 105 in Fig. 1 (step S804). Here, the CPU 701 acquires, as reference expert knowledge data 116, one or more pieces of expert knowledge data 301, 302, ..., 306 (see Fig. 3) corresponding to each of the expert knowledge identifiers 404 (reference expert knowledge identifiers 115) included in each of the reference concept data 112 extracted in the concept extraction process in step S802 from the expert knowledge database 101.

[0033] Thereafter, the CPU 701 executes the reference-attached commentary acquisition process described as a function of the reference-attached commentary acquisition unit 106 in Fig. 1. Specifically, the CPU 701 calls a program for the generation AI API access process using, as an argument, the query data of the reference prompt 117 configured from the reference expert knowledge data 116 acquired in the expert knowledge data acquisition process in step S804 (step S805). This process will be described later using the flowchart in Fig. 9.

[0034] Finally, CPU 701 executes the display process described as the function of display unit 107 in Fig. 1 (step S806). CPU 701 outputs and displays text data 110 of the input sentence, text data 114 of the output sentence acquired in step S803, and text data 118 of the explanatory text with references acquired in step S805 on a display (not shown) of user terminal device 109 via network 119, for example, in a format such as that shown in Fig. 6. Thereafter, the CPU 701 ends the control operation shown in the flowchart of FIG. 8 and waits for the input of the text data 110 of the next input sentence.

[0035] Figure 9 is a flowchart showing an example of processing of a generated AI API access program, which is a subroutine called from the processes of step S803 and step S805 in Figure 8. As with Figure 8, this flowchart also shows processing of executing a generated AI API access program written in, for example, the Python computer programming language, which the CPU 701 in Figure 7 has loaded from the external storage device 703 to memory 702.

[0036] 9, first, the CPU 701 sets a generative AI model (step S901). This is, for example, a command specifying ChatGPT-3.5 or ChatGPT-4o as the generative AI model.

[0037] Next, the CPU 701 sets the inquiry data (the data of the input prompt 113 or the reference prompt 117 described above) passed by the subroutine call from the processing of step S803 or step S805 in FIG. 8 to a predetermined variable on the memory 702 (step S902).

[0038] Then, the CPU 701 sets, for example, other control parameters for the ChatGPT API in other predetermined variable sets on the memory 702 (step S903).

[0039] After the processing of steps S901 to S903, the CPU 701 calls the response setting function in the API library read in step S803 or step S805 of FIG. 8 described above, using the text data or parameter group set in the variables in steps S901 to S903 as arguments (step S904).

[0040] Next, the CPU 701 calls the response acquisition function of the API library (step S905).

[0041] Finally, the CPU 701 returns the response from the generation AI system 108 (e.g., ChatGPT system, Figure 1) obtained by the call in step S905 as text data 114 of the output sentence or text data 118 of the explanatory text with references in Figure 1 described above, and returns to processing of step S803 or step S805 in Figure 8 described above (step S906).

[0042] According to the embodiment described above, the text data 110 of the input sentence is converted into conceptual data (reference conceptual data 112) by referring to the conceptual database 102 via the input vector values ​​111 corresponding to the words and sentences that make up the text data, and then the specialized knowledge data (reference specialized knowledge data 116) linked to the text data is referenced in the specialized knowledge database 101. This makes it possible to obtain more accurate specialized knowledge data for input sentences such as questions and news feeds. Furthermore, by providing a prompt (reference prompt 117) to the generation AI system 108 to instruct an explanation of the expert knowledge data (reference expert knowledge data 116) obtained in this manner, it is possible to obtain an appropriate explanation and display it on the user terminal device 109, etc. Furthermore, by displaying a referenced explanatory text that also includes a reference link to the corresponding expert knowledge data in the original expert knowledge database 101, it is possible to further improve the convenience of reference for users. In addition, by providing the generation AI system 108 with a prompt (input prompt 113) that instructs it to respond with text data of the main topic, summary, answer, advice, or explanation pointing out points of caution or problems for the text data 110 of the input sentence based on the text data 110 of the input sentence and its corresponding concept data (reference concept data 112), it is now possible to obtain text data 114 of the output sentence that appropriately indicates the main topic, summary, answer, advice, or explanation pointing out points of caution or problems for the text data 110 of the input sentence, and display it on the user terminal device 109, etc.

[0043] The above embodiment has been described using personnel and labor knowledge as an example of specialized knowledge, but the present invention can be applied to specialized knowledge such as accounting knowledge, legal knowledge, administrative knowledge, and industrial property knowledge. [Explanation of symbols]

[0044] 100 Information provision device 101 Expert Knowledge Database 102 Conceptual Database 103 Concept extraction part 104 Output text acquisition unit 105 Expert Knowledge Data Acquisition Department 106 Referenced commentary acquisition section 107 Display section 108 Generative AI System 109 User terminal equipment 110 Text data of input sentence 111 Input Vector Values 112 Reference Concept Data 113 Input Prompt 114 Text data of output sentences 115 Reference Expertise Identifier 116 Reference Expertise Data 117 Reference Prompt 118 Text data of explanatory text with references 119 Network 200 Morphological analysis section 201 Word / Vector Value Conversion Machine Learning Model 202 Vector Synthesis Unit 203 Word / Vector Value Conversion Machine Learning Model Training Unit 204 Conceptual Database Creation Department 210 Word Vector Values 211 Sentence Vector Values 301 Supreme Court Case Data 302 Labor Law Provisions and Notifications e-Gov Data 303 Labor Law Related Forms Data Set 304 Labor-related news data 305 Labor-related web article data 306 Labor Information Message Information 401 Conceptual Data 401′ Parent Concept Data 402 Concept Vector Values 403 Concept generation phrases 404 Expertise Identifier 405 Conceptual Overview 406 Parent Link

Claims

1. a specialized knowledge database that stores specialized knowledge data, which is data related to specialized knowledge, as a database; a concept database that stores, for each concept constituting the specialized knowledge, concept data including a concept vector value that is a vector value corresponding to the concept, a concept generating phrase that is a word or sentence that generated the concept, and one or more specialized knowledge identifiers that link and identify the concept with one or more specialized knowledge data in the specialized knowledge database; a concept extraction unit that receives input text data of an input sentence related to the specialized knowledge, generates input vector values ​​that are vector values ​​corresponding to words or sentences that constitute the input sentence, and extracts one or more concept data having the concept vector values ​​similar to the input vector values ​​from the concept database as reference concept data; a specialized knowledge data acquisition unit that acquires, as reference specialized knowledge data, one or more specialized knowledge data corresponding to each of the specialized knowledge identifiers included in each of the reference concept data from the specialized knowledge database; a referenced commentary acquisition unit that acquires text data of a commentary that explains the referenced expert knowledge data by providing a reference prompt that instructs an explanation of the referenced expert knowledge data to a generative artificial intelligence system, and acquires text data of a referenced commentary that includes the commentary and a reference to each of one or more of the expert knowledge data in the expert knowledge database that correspond to the referenced expert knowledge data; a display unit that outputs and displays the text data of the input sentence and the text data of the commentary with references on a user's terminal device; An information providing device comprising:

2. an output sentence acquisition unit that acquires text data of an output sentence by issuing an instruction to a generative AI system that is the same as or different from the generative AI system using an input prompt that is composed of the concept generation phrase that generated the reference concept data and the text data of the input sentence; the display unit further outputs and displays the text data of the output sentence together with the input sentence and the commentary with reference on the user's terminal device. The information providing device according to claim 1 .

3. The information providing device of claim 2, wherein the input prompt includes content instructing the device to respond with text data of the output sentence that is a main theme, summary, answer, advice, or commentary pointing out points of caution or problems regarding the text data of the input sentence.

4. The concept extraction unit a morphological analysis unit that performs morphological analysis on the text data of the input sentence to separate the text data of the input sentence into words; a word / vector value conversion machine learning model unit that, for each significant word obtained by segmenting the text data of the input sentence output from the morphological analysis unit, outputs a word vector value that is a vector value corresponding to the word by inputting the text data of the word; a vector synthesis unit that extracts the sentence by syntactically analyzing the dependency relationships between the words in the input sentence, and outputs a sentence vector value by synthesizing each of the word vector values ​​corresponding to each of the words that constitute the sentence; Equipped with The word vector value and the sentence vector value are defined as the input vector values. The information providing device according to claim 1 .

5. a word / vector value conversion machine learning model training unit that trains the word / vector value conversion machine learning model unit using a corpus of text data of the specialized knowledge as training data; a concept database creation unit that, after learning is completed by the word / vector value conversion machine learning model learning unit, stores the learning data in the specialized knowledge database, and creates, for each word constituting the learning data, conceptual data in the concept database, the conceptual data including, as the conceptual vector value, the input vector value output by inputting text data of the word to the word / vector value conversion machine learning model unit, and including, as the concept vector value, a word or phrase corresponding to the word, and including, as the concept-generating word or phrase, information linking and identifying the learning data stored in the specialized knowledge database, and the conceptual data in the concept database, the conceptual data including, as the conceptual vector value, the sentence vector value output from the vector synthesis unit corresponding to the learning data, and including, as the concept-generating word or phrase, information linking and identifying the learning data stored in the specialized knowledge database, and including, as the specialized knowledge identifier, The information providing device according to claim 4 , further comprising:

6. The conceptual database further includes, for each of the conceptual data stored in the conceptual database, parent link information indicating a link to other conceptual data that is conceptually in a parent relationship with the conceptual data, the concept extraction unit extracts, as the reference conceptual data, one or more conceptual data having the conceptual vector value similar to the input vector value, as well as the conceptual data that is in the parent relationship with the conceptual data linked to the conceptual data by the parent link information included in the conceptual data; The information providing device according to claim 1 .

7. 2. The information providing device according to claim 1, wherein the input text is news information related to the specialized knowledge.

8. The information providing device according to claim 1 , wherein the input sentence is question information relating to the specialized knowledge.

9. 2. The information providing device according to claim 1, wherein the input text is notice information from a government or related organization relating to the specialized knowledge.

10. 2. The information providing device according to claim 1, wherein the input text is information on an in-house application, contract, various documents, or format related to the specialized knowledge.

11. 2. The information providing device according to claim 1, wherein the input text is information about various in-house rules related to the specialized knowledge.

12. An information providing device as described in any one of claims 1 to 11, wherein the specialized knowledge data stored in the specialized knowledge database is one or more of the following: Supreme Court case law data related to the specialized knowledge, article and notice data related to the specialized knowledge, related format data of laws related to the specialized knowledge, news data related to the specialized knowledge, web data published on the Internet related to the specialized knowledge, or practical information or message information related to the specialized knowledge.

13. 12. The information providing device according to claim 1, wherein the specialized knowledge is human resources and labor knowledge.

14. An information provision device described in any one of claims 1 to 11, wherein the generative artificial intelligence system inputs an input prompt to each of a plurality of sub-generative artificial intelligence systems, obtains a response from each of the sub-generative artificial intelligence systems, and outputs an appropriate response based on the respective responses.

15. storing expert knowledge data, which is data relating to expert knowledge, in an expert knowledge database; storing, for each concept constituting said specialized knowledge, concept data in a concept database, the concept data including a concept vector value which is a vector value corresponding to said concept, a concept-generating phrase which is a word or sentence included in said concept, and one or more specialized knowledge identifiers which link and identify said concept with one or more of said specialized knowledge data in said specialized knowledge database; inputting text data of an input sentence related to the specialized knowledge, generating input vector values ​​that are vector values ​​corresponding to words or sentences that constitute the input sentence, and extracting one or more pieces of conceptual data having the conceptual vector values ​​similar to the input vector values ​​as reference conceptual data; acquiring one or more pieces of expert knowledge data corresponding to the expert knowledge identifiers included in each of the reference concept data from the expert knowledge database as reference expert knowledge data; obtaining text data of an explanatory statement explaining the reference expert knowledge data by providing a reference prompt to a generative artificial intelligence system instructing an explanation of the reference expert knowledge data, and obtaining text data of an explanatory statement with reference including the explanatory statement and a reference to each of one or more expert knowledge data in the expert knowledge database corresponding to the reference expert knowledge data; outputting and displaying the text data of the input sentence and the text data of the commentary with references on a user's terminal device; An information providing method executed by an information providing device.

16. A process of storing expert knowledge data, which is data related to expert knowledge, in an expert knowledge database; a process of storing, for each concept constituting said specialized knowledge, concept data in a concept database, the concept data including a concept vector value which is a vector value corresponding to said concept, a concept-generating phrase which is a word or sentence included in said concept, and one or more specialized knowledge identifiers which link and identify said concept with one or more specialized knowledge data in said specialized knowledge database; a process of inputting text data of an input sentence related to the specialized knowledge, generating input vector values ​​that are vector values ​​corresponding to words or sentences that constitute the input sentence, and extracting one or more of the concept data having the concept vector values ​​similar to the input vector values ​​as reference concept data; a process of acquiring one or more pieces of expert knowledge data corresponding to each of the expert knowledge identifiers included in each of the reference concept data from the expert knowledge database as reference expert knowledge data; a process of obtaining text data of an explanatory statement explaining the reference expert knowledge data by providing a reference prompt to a generative artificial intelligence system, the text data of an explanatory statement including the explanatory statement and a reference to each of one or more expert knowledge data in the expert knowledge database corresponding to the reference expert knowledge data; a process of outputting and displaying the text data of the input sentence and the text data of the commentary with references on a user's terminal device; A program that causes a computer to execute the following.

Images