Financial document question answering method and device based on label hierarchical knowledge graph

Abstract

The embodiment of the application provides a financial document question and answer method and device based on a label hierarchical knowledge graph, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring original question data input by a target object; constructing a label hierarchical knowledge graph based on a preset financial field resource information base; performing label query on the original question data based on the label hierarchical knowledge graph to obtain a target graph label; performing document query on the financial field resource information base based on the target graph label to obtain an initial query document; performing document screening on the initial query document to obtain a target query document; performing text generation based on the target query document to obtain target financial reply content; and replying to the target object based on the target financial reply content. The embodiment of the application can improve the efficiency and accuracy of question and answer generation in the financial field.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a financial document question-answering method and apparatus based on a tag-hierarchical knowledge graph. Background Technology

[0002] Question answering is an artificial intelligence technology that uses knowledge bases, algorithmic models, or manual processing to understand and analyze user questions and generate accurate and relevant answers. It is widely used in various scenarios such as intelligent customer service and search engine optimization. For example, in the financial sector, it can provide financial Q&A based on customer-input financial questions.

[0003] Currently, in financial scenarios, question-answering methods primarily involve directly analyzing user questions to locate relevant subgraphs in the knowledge graph and generating answers directly, or using RAG question-answering technology to retrieve relevant content from the entire document library and generate answers. However, in practical applications, when faced with large-scale financial documents, both of these methods suffer from slow retrieval speeds and inaccurate retrieval content, impacting the efficiency and accuracy of question-answer generation.

[0004] Therefore, improving the efficiency and accuracy of question-and-answer generation in the financial field has become an urgent technical problem to be solved. Summary of the Invention

[0005] The main objective of this application is to propose a financial document question-answering method and apparatus based on a tag-hierarchical knowledge graph, aiming to improve the efficiency and accuracy of question-answer generation in the financial field.

[0006] To achieve the above objectives, a first aspect of this application proposes a financial document question-answering method based on a tag-hierarchical knowledge graph, the method comprising: Obtain the original question data input by the target object; A hierarchical knowledge graph based on a pre-set financial sector resource information database is constructed. The original question data is queried based on a tag-level knowledge graph to obtain the target graph tags. Based on the target map tags, a document query is performed on the financial sector resource information database to obtain the initial query document; The initial query document is filtered to obtain the target query document; Based on the target query document, text is generated to obtain the target financial response content; A response will be sent to the target recipient based on the content of the target financial response.

[0007] In some embodiments, the financial domain resource information database includes financial domain documents; the construction of a tag-based hierarchical knowledge graph based on the preset financial domain resource information database includes: Obtain the document tags for the aforementioned financial documents; Generate graph hierarchy child nodes based on the document tags; Generate hierarchical relationships based on preset hierarchical information; Based on the graph hierarchy, the sub-nodes of the graph hierarchy are used to construct the graph, thereby obtaining the tag hierarchy knowledge graph.

[0008] In some embodiments, the document tags of the financial documents are obtained in the following ways: Obtain the original financial documents; The original financial document is extracted to obtain financial text data; The financial text data is structured to obtain the financial domain document; Entity extraction is performed on the financial text data to obtain financial entity data; The document tags for the financial documents are generated based on the financial entity data.

[0009] In some embodiments, the step of filtering the initial query document to obtain the target query document includes: The similarity score is obtained by calculating the similarity between the initial query document and the original question data; Obtain the tag level of the target map tags; The target score of the initial query document is obtained by weighting the tag level, the similarity score, the importance score of the initial query document, and the publication score. Based on the target score, the initial query document is processed to obtain the target query document.

[0010] In some embodiments, the step of weighting the initial query document's target score based on the tag level, the similarity score, the importance score of the initial query document, and the publication score to obtain the target score includes: Obtain target weight data, wherein the target weight data includes first weight data, second weight data, and third weight data; A first score is obtained by performing aggregation calculations based on the label hierarchy and the first weight data; A second score is obtained by aggregating and calculating the similarity score, the importance score, and the second weight data. The third score is obtained by aggregating and calculating the publication score and the third weight data. The target score of the initial query document is obtained by summing the first score, the second score, and the third score.

[0011] In some embodiments, the step of performing a tag query on the original question data based on a tag-hierarchical knowledge graph to obtain target graph tags includes: The original question data is subjected to intent recognition to obtain the original question intent data; Based on the original question intent data, a graph query is performed on the tag hierarchy knowledge graph to obtain graph node data; The target map label is generated based on the map node data.

[0012] In some embodiments, the step of generating text based on the target query document to obtain the target financial response content includes: Based on the target query document, the response content is generated to obtain the original financial response content; Obtain the document link of the target query document; Inserting the document link into the original financial response content yields the target financial response content.

[0013] To achieve the above objectives, a second aspect of this application proposes a financial document question-answering device based on a tag-hierarchical knowledge graph, the device comprising: The problem data acquisition module is used to acquire the raw problem data input by the target object; The knowledge graph construction module is used to construct a tag-hierarchical knowledge graph based on a pre-set financial sector resource information database; The graph tag query module is used to perform tag queries on the original question data based on the tag hierarchy knowledge graph to obtain the target graph tags; The document query module is used to perform document queries on the financial sector resource information database based on the target map tags to obtain the initial query document; The document filtering module is used to filter the initial query documents to obtain the target query documents; The response content generation module is used to generate text based on the target query document to obtain the target financial response content. The response module is used to respond to the target object based on the target financial response content.

[0014] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0015] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0016] This application proposes a financial document question-answering method and apparatus based on a tag-hierarchical knowledge graph. It acquires raw question data input by the target object and constructs a tag-hierarchical knowledge graph based on a pre-set financial domain resource information database. This systematically organizes domain knowledge and forms a structured knowledge system. Next, it performs tag queries on the raw question data based on the tag-hierarchical knowledge graph to obtain target graph tags, enabling rapid association between the question and key nodes in the knowledge graph, improving information matching efficiency. Furthermore, it performs document queries on the financial domain resource information database based on the target graph tags to obtain initial query documents, narrowing the information retrieval scope. Document filtering of the initial query documents yields the target query document, further removing irrelevant information and improving information quality. Finally, it generates the target financial response content based on the target query document and responds to the target object based on this response, effectively improving the efficiency and accuracy of information processing and response. Attached Figure Description

[0017] Figure 1 This is a flowchart of a financial document question-answering method based on a tag-hierarchical knowledge graph provided in an embodiment of this application; Figure 2 This is a flowchart of a financial document question-answering method based on a tag-hierarchical knowledge graph provided in another embodiment of this application; Figure 3 yes Figure 1 The flowchart of step S102 in the document; Figure 4 yes Figure 1 The flowchart of step S103 in the process; Figure 5 yes Figure 1 The flowchart of step S105 in the process; Figure 6 yes Figure 5 The flowchart of step S503 in the process; Figure 7 yes Figure 1 The flowchart of step S106 in the process; Figure 8 This is a schematic diagram of the structure of the financial document question-answering device based on a tag-hierarchical knowledge graph provided in this application embodiment; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] First, let's analyze some of the terms used in this application: Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0022] Natural Language Processing (NLP): NLP uses computers to process, understand, and utilize human language (such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary field of computer science and linguistics, often referred to as computational linguistics. NLP includes syntactic analysis, semantic analysis, and discourse understanding. It is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, intent recognition, information extraction and filtering, text classification and clustering, sentiment analysis, and opinion mining. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computation.

[0023] Information extraction is a text processing technique that extracts factual information such as entities, relationships, and events from natural language text and outputs it as structured data. Information extraction is a technique for extracting specific information from text data. Text data is composed of specific units, such as sentences, paragraphs, and chapters. Text information is composed of smaller, specific units, such as characters, words, phrases, sentences, paragraphs, or combinations of these units. Extracting noun phrases, names of people, and place names from text data is an example of text information extraction. Of course, text information extraction techniques can extract information of various types.

[0024] A knowledge graph is a structured semantic network that organizes and represents knowledge graphically. It uses entities (such as people, things, and concepts) as nodes and relationships between entities as edges, constructing a vast, interconnected network. Knowledge graphs can clearly present the complex connections between different entities, revealing the knowledge logic hidden within massive amounts of data. In numerous fields such as intelligent search, question-answering systems, recommendation algorithms, and medical diagnosis, knowledge graphs play a crucial role in helping to achieve more accurate and efficient information processing and decision support.

[0025] Tag-based hierarchical knowledge graphs are a graph-structured knowledge organization method. Nodes represent entities (such as people, events, and concepts), edges represent relationships between entities (such as "belongs to" or "contains"), and a tagging system is introduced to classify and label entities at multiple levels. Its core lies in constructing a hierarchical tagging system.

[0026] Retrieval-Augmented Generation (RAG) is a technical architecture that combines information retrieval and text generation. First, it uses a retrieval model to accurately locate information fragments relevant to the user's query from a large-scale knowledge base or document collection. Then, it uses the retrieved content as context to input into the generation model, assisting in generating more accurate, comprehensive, and factual answers. This architecture effectively overcomes the limitations of traditional generation models that rely solely on their own parameter knowledge. It can utilize massive amounts of external information to improve the quality and reliability of generated content, demonstrating significant advantages in many scenarios that require the integration of external knowledge, such as intelligent question answering, knowledge reasoning, and content creation.

[0027] Question answering is an artificial intelligence technology that uses knowledge bases, algorithmic models, or manual processing to understand and analyze user questions and generate accurate and relevant answers. It is widely used in various scenarios such as intelligent customer service and search engine optimization.

[0028] Currently, knowledge graph-based question answering methods primarily generate answers by directly analyzing user questions and locating relevant subgraphs within the knowledge graph. However, in practical applications, when faced with a large volume of documents, issues such as slow retrieval speed and inaccurate retrieval content arise, impacting the efficiency and accuracy of question answer generation. Furthermore, the pre-extraction of entity relationships and the construction of the knowledge graph present dual challenges in terms of human resources and retrieval efficiency.

[0029] In addition, existing RAG question answering technologies are usually based on text vectorization to directly retrieve content from the entire document. When faced with large-scale data, they also suffer from slow retrieval speed, incomplete content recall, and insufficient reasoning ability in specific domains (such as causal relationships of financial events, supply chain analysis, etc.).

[0030] Based on this, embodiments of this application provide a financial document question-answering method and apparatus based on a tag-hierarchical knowledge graph, aiming to improve the efficiency and accuracy of question-answer generation in the financial field.

[0031] The financial document question answering method and apparatus based on tag-hierarchical knowledge graph provided in this application are specifically described through the following embodiments. First, the financial document question answering method based on tag-hierarchical knowledge graph in this application is described.

[0032] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0033] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0034] The financial document question-answering method based on a tag-hierarchical knowledge graph provided in this application relates to the field of artificial intelligence technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the tag-hierarchical knowledge graph-based financial document question-answering method, but is not limited to the above forms.

[0035] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0036] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0037] Figure 1 This is an optional flowchart of a financial document question-answering method based on a tag-hierarchical knowledge graph provided in this application embodiment. Figure 1The method may include, but is not limited to, steps S101 to S107.

[0038] Step S101: Obtain the original problem data input by the target object; Step S102: Construct a tag-hierarchical knowledge graph based on a pre-set financial sector resource information database; Step S103: Perform tag query on the original question data based on the tag hierarchical knowledge graph to obtain the target graph tags; Step S104: Based on the target map tags, perform document queries on the financial sector resource information database to obtain the initial query documents; Step S105: Filter the initial query documents to obtain the target query documents; Step S106: Generate text based on the target query document to obtain the target financial response content; Step S107: Respond to the target recipient based on the content of the target financial response.

[0039] Steps S101 to S107 of this embodiment involve acquiring the original question data input by the target object and constructing a hierarchical knowledge graph based on a pre-defined financial domain resource information database. This systematically organizes domain knowledge and forms a structured knowledge system. Next, a tag query is performed on the original question data based on the hierarchical knowledge graph to obtain target graph tags. This allows for rapid association between the question and key nodes in the knowledge graph, improving information matching efficiency. Furthermore, a document query is performed on the financial domain resource information database based on the target graph tags to obtain initial query documents, narrowing the information retrieval scope. Document filtering of the initial query documents yields the target query document, further removing irrelevant information and improving information quality. Finally, text generation is performed based on the target query document to obtain the target financial response content. This response is then sent to the target object, effectively improving the efficiency and accuracy of information processing and response.

[0040] In step S101 of some embodiments, the target object refers to the object that asks a question in the intelligent customer service consultation window of a specific platform, and the original question data is the unprocessed initial question content input by the target object in the form of natural language or the like.

[0041] For example, in a financial Q&A platform, users can access an intelligent customer service window through a pre-defined entry point to ask financial questions. The original question data is the content input by the target user, which can be in string format, such as querying relevant information about a company or searching for documents. Additionally, the original question data is in voice format and needs to be converted to text to obtain the final original question data.

[0042] In some embodiments, the financial sector resource information repository includes financial sector documents; these financial sector documents contain document tags.

[0043] The pre-built financial sector resource information database is a collection of various information related to financial business scenarios, such as text and data, which are collectively referred to as financial sector documents. Financial sector documents are document data related to securities issuance, financial information, prospectuses, annual reports, announcements, etc., within the financial context. Document tags are used to indicate the type, region, industry, event, and concept of financial sector documents.

[0044] It should be noted that the number of documents contained in the financial sector resource information database is in the tens of billions or even higher. Using conventional knowledge graphs or RAG question answering methods results in slow retrieval speed and incomplete content recall. Therefore, it is necessary to construct a tag-hierarchical knowledge graph for the financial sector documents in the financial sector resource information database to improve the efficiency and accuracy of retrieval.

[0045] Please see Figure 2 In some embodiments, document tags for financial documents can be obtained in the manner indicated by steps S201 to S205: Step S201: Obtain the original financial documents; Step S202: Extract financial data from the original financial document to obtain financial text data; Step S203: Perform structured processing on the financial text data to obtain financial documents; Step S204: Extract entities from the financial text data to obtain financial entity data; Step S205: Generate document tags for financial documents based on financial entity data.

[0046] Steps S201 to S205, as illustrated in this embodiment, involve extracting financial text data from the original financial document, then performing structured processing on the financial text data to obtain financial domain documents, facilitating storage and retrieval. Next, entity extraction is performed on the financial text data to obtain financial entity data, accurately locating key elements. Based on the financial entity data, document tags for financial domain documents are generated, enabling rapid and accurate document identification, improving the quality and efficiency of financial document processing, enhancing the utilization value of financial information, and contributing to improved retrieval efficiency.

[0047] In step S201 of some embodiments, the original financial document refers to various documents and materials that are unprocessed and directly derived from the financial business scenario, covering a variety of information such as securities issuance, financial information, prospectuses, annual reports, and announcements, and containing rich financial information.

[0048] Specifically, original financial documents can be obtained from public channels (such as financial news reports, corporate websites, financial websites, etc.) to provide basic materials for subsequent processing, ensuring the integrity and originality of the information and avoiding deviations in subsequent analysis due to missing or incorrect information.

[0049] In step S202 of some embodiments, financial text data is obtained by extracting documents from the original financial document. The financial text data is the extracted plain text content, which includes, but is not limited to, document text content, image text content, table text content, and formula text content.

[0050] For text content, natural language processing techniques can be used to extract the text content and obtain the document text content; For image content, optical character recognition (OCR) methods can be used to identify the image content and obtain the image text content; the image content can include financial statement analysis charts, transaction data visualization graphics, etc.

[0051] For table content, machine learning-based methods can be used to identify and parse the table structure and content to obtain the table text content; the tables can include balance sheets, income statements, cash flow statements, etc. from financial statements.

[0052] For formula content, various formula templates can be predefined. The formula content is matched with the templates to find the most similar template, thereby identifying the structure and symbols of the formula and obtaining the formula text content. Among them, the formulas can include formulas for calculating compound interest and present value.

[0053] In step S203 of some embodiments, the document text content, image text content, table text content, and formula text content in the financial text data are combined according to a certain structure and rules to obtain financial documents, making the data more organized, improving the readability and usability of the data, and facilitating data mining and decision support.

[0054] In some embodiments, certain structures and rules can be combined based on the order of the original financial documents to ensure that the extracted financial documents contain the same content as the original financial documents.

[0055] In step S203 of some other embodiments, paragraphs can be merged to form text blocks according to the business domain adaptation, and these text blocks can be vectorized to obtain financial domain documents.

[0056] Specifically, for different types of documents, the methods for dividing text blocks include, but are not limited to, the following three: (1) The recursive merging method based on word count and table of contents hierarchy is suitable for long documents with rich content and complex table of contents hierarchy, such as annual reports of listed companies. The table of contents of such documents usually has more than three levels, and the table content is relatively long. For such documents, the recursive merging method of table of contents headings can be used to merge text paragraphs under the same heading into one block as much as possible, and while preserving the complete information of the paragraphs, the word count variance between different blocks should be minimized.

[0057] (2) The title + summary block method is suitable for documents with simple content, such as some temporary announcements of listed companies. The temporary announcements disclosed by listed companies on a daily basis, such as shareholders' meetings, equity incentives, additional issuances, and major transactions, usually only contain specific information. The directory structure of these announcements is relatively simple, and the article content is relatively short. For this type of document, you can directly block it using the title + summary method, so that the blocks retain the main content of the original article.

[0058] (3) For policy and regulation documents, divide the text into blocks according to the clauses, and keep the complete clause paragraphs in one text block.

[0059] In step S204 of some embodiments, natural language processing techniques, such as named entity recognition (NER) algorithms, can be used to identify various entities in financial text data, accurately locate key elements, facilitate the classification, statistics and analysis of financial information, and uncover potential value.

[0060] Among them, financial entity data refers to entities with specific significance in financial scenarios, such as document ID, document disclosure time, disclosing institution, stock abbreviation, listing sector, document title, author, region, industry, event, concept, etc.

[0061] It should be noted that named entity recognition algorithms need to be trained specifically for financial scenarios to ensure the accuracy of the algorithm in recognizing financial text data.

[0062] In step S205 of some embodiments, after obtaining financial entity data, the financial entity data is tagged to obtain document tags for financial documents; wherein, the document tags are used for a brief description and classification of financial documents, which can enable quick location and understanding of document content, facilitate subsequent quick retrieval and document Q&A, and improve information acquisition efficiency.

[0063] Please see Figure 3 In some embodiments, step S102 may include, but is not limited to, steps S301 to S304: Step S301: Obtain document tags for financial documents; Step S302: Generate graph hierarchy child nodes based on document tags; Step S303: Generate a map hierarchy relationship based on preset map hierarchy information; Step S304: Based on the hierarchical relationship of the graph, construct the graph of the hierarchical child nodes to obtain the tag-level knowledge graph.

[0064] Steps S301 to S304, as illustrated in this embodiment, obtain document tags from financial documents to accurately locate key information within those documents. Based on these document tags, hierarchical sub-nodes are generated, laying the foundation for knowledge graph construction. Next, hierarchical relationships are generated based on preset hierarchical information, clarifying the relationships between nodes. Furthermore, the hierarchical sub-nodes are constructed based on these relationships to obtain a tag-based hierarchical knowledge graph. This allows for the systematic and accurate construction of a knowledge graph using financial domain knowledge, facilitating efficient retrieval.

[0065] In step S302 of some embodiments, in the knowledge graph, the graph hierarchy child nodes represent points of specific knowledge concepts. These points are organized according to a certain hierarchical structure. The child nodes are nodes at a relatively low level in the hierarchical structure, providing a basis for establishing relationships between nodes in the future.

[0066] For example, if a document in the financial field has document tags such as "securities", "IPO", and "a certain prefecture-level city", then a graph-level sub-node corresponding to "IPO", a graph-level sub-node corresponding to "a certain prefecture-level city", and a graph-level sub-node corresponding to "securities" will be generated based on the document tags.

[0067] In some embodiments, the preset graph hierarchy information needs to be set according to the actual application scenario. For example, in a financial scenario, the knowledge graph uses a multi-level hierarchical structure design, which may include, but is not limited to: First level: Industry (such as "securities", "banking", "trust", "insurance"), region (such as "city A", "city B", "city C"); Second level: Business segments (such as "underwriting and sponsorship"); The third level: events / concepts (such as "IPO", "rights issue"); The fourth layer includes disclosure content (such as financial information) and document content classification (such as prospectus).

[0068] In step S302 of some embodiments, the hierarchical relationship between each graph level sub-node is determined according to pre-defined graph hierarchy information. This hierarchical relationship can be represented using the relation definition method in a graph database. This ensures that the hierarchical structure of the knowledge graph is reasonable, logically clear, and conforms to the characteristics and rules of domain knowledge. Furthermore, it improves the scalability and maintainability of the knowledge graph. When a new knowledge node needs to be added, its position and relationship with other nodes can be quickly determined according to the pre-defined hierarchical relationship, facilitating the updating and optimization of the knowledge graph.

[0069] In step S304 of some embodiments, the hierarchical relationships of the knowledge graph are utilized to connect and organize the child nodes of the graph hierarchy, forming a complete tag-based hierarchical knowledge graph. Specifically, graph database tools can be used to store nodes and relationships in a database, and visualization tools can be used to display the knowledge graph. Intelligent querying and reasoning are supported; based on the hierarchical structure and relationships of the knowledge graph, more efficient query and reasoning functions can be achieved. For example, users can quickly obtain knowledge about related financial markets, financial institutions, and risk characteristics by querying relevant information about a financial product.

[0070] Understandably, the construction of a tag-based hierarchical knowledge graph stores only abstract knowledge, rather than detailed document content. It simulates human thinking by starting with a knowledge system and then searching for corresponding documents, reducing the storage size and query difficulty of the knowledge graph, and improving query efficiency. Furthermore, because it separates abstract tags into different levels, the content of upper-level tags can be adjusted at any time according to business and market needs without reconstructing a massive document tag library. Additionally, in resource-constrained scenarios, large-scale document knowledge base retrieval faces problems such as slow retrieval speed, high noise in retrieved content, and incomplete recall. Through a predefined tag-based hierarchical knowledge graph, a precise and dynamic topic knowledge base can be constructed to provide an accurate recall pool, improving retrieval speed and effectiveness.

[0071] Please see Figure 4 In some embodiments, step S103 may include, but is not limited to, steps S401 to S403: Step S401: Perform intent recognition on the original question data to obtain the original question intent data; Step S402: Based on the original question intent data, perform a graph query on the tag hierarchy knowledge graph to obtain graph node data; Step S403: Generate target map labels based on map node data.

[0072] Steps S401 to S403, as illustrated in this embodiment, accurately grasp the core of the problem and obtain the original problem intent data by identifying the intent of the original problem data. Based on this, a graph query is performed on the tag-level knowledge graph based on the original problem intent data, which can quickly locate the graph node data. Finally, target graph tags are generated based on the graph node data, which can efficiently and accurately associate the problem with the knowledge graph, improving the efficiency and accuracy of information processing.

[0073] In some embodiments, step S401 may include, but is not limited to, the following steps: Preprocess the original problem data to obtain initial problem data; Intent identification is performed on the initial question data to obtain the original question intent data.

[0074] Specifically, preprocessing may include, but is not limited to: removing punctuation marks, spaces, special symbols, stop words, word segmentation, correcting common typos, and unifying synonyms, thereby standardizing the original problem data, eliminating interference from irrelevant information, and helping to improve the efficiency and accuracy of intent recognition.

[0075] After obtaining the initial question data, techniques such as natural language processing and machine learning can be used to analyze the semantic information in the initial question data, understand the purpose and needs of the target audience in raising the question, and determine the information that needs to be obtained.

[0076] In step S402 of some embodiments, the original question intent data is transformed into a query statement, and matching and retrieval are performed in the tag-level knowledge graph to find the node that matches the original question intent data. This can quickly and accurately locate the knowledge related to the intent, improving the efficiency and accuracy of information retrieval.

[0077] In step S403 of some embodiments, the map node data is analyzed and refined to extract key features and information, and concise and accurate target map labels are generated according to certain rules and templates to facilitate subsequent processing and application, and to help improve the efficiency and accuracy of information processing.

[0078] In step S104 of some embodiments, financial documents all have corresponding document tags, and the graph level sub-nodes are constructed based on document tags. Therefore, the target graph tags can be used as keywords to tag documents in the financial resource information database, obtain financial documents containing these keywords, and obtain the initial query document, thereby improving the efficiency of information acquisition.

[0079] For example, if the target map tags are "City A", "insurance", "rights issue", and "prospectus", then the financial field documents in the financial field resource information database will be searched based on these target map tags to obtain the documents whose document tags match, which are the initial query documents.

[0080] Please see Figure 5 In some embodiments, step S105 may include, but is not limited to, steps S501 to S504: Step S501: Calculate the similarity between the initial query document and the original question data to obtain a similarity score; Step S502: Obtain the tag level of the target map tag; Step S503: The target score of the initial query document is obtained by weighting the tag level, similarity score, importance score of the initial query document, and publication score. Step S504: Perform document processing on the initial query document based on the target score to obtain the target query document.

[0081] Steps S501 to S504, as illustrated in this embodiment, calculate the similarity score between the initial query document and the original question data, providing a basis for assessing relevance. Next, the tag level of the target graph tags is obtained, and a weighted calculation is performed based on the tag level, similarity score, importance score of the initial query document, and publication score to obtain the target score of the initial query document, comprehensively measuring its value. Finally, document processing is performed on the initial query document based on the target score to obtain the target query document, accurately selecting target query documents that better meet the needs and have higher value.

[0082] In step S501 of some embodiments, methods such as cosine similarity algorithm and Jaccard similarity algorithm can be used to calculate the similarity between the initial query document and the original question data, and obtain a similarity score. This can quickly measure the relevance between the initial query document and the original question data, and provide a basis for subsequent document screening.

[0083] In step S502 of some embodiments, the tag hierarchy knowledge graph is multi-level, and its level is determined by the graph hierarchy information. The target graph tags are selected from the tag hierarchy knowledge graph. Therefore, the tag level of the target graph tags can be determined based on the graph hierarchy information. In the subsequent weighted calculation, considering the tag level can make the selection results more consistent with the logic of the knowledge structure and give priority to documents with more specific and relevant content.

[0084] Please see Figure 6 In some embodiments, step S503 may include, but is not limited to, steps S601 to S605: Step S601: Obtain target weight data, wherein the target weight data includes first weight data, second weight data and third weight data; Step S602: Perform aggregation calculation based on label level and first weight data to obtain the first score; Step S603: Based on the similarity score, importance score, and second weight data, perform aggregate calculation to obtain the second score; Step S604: Perform aggregation calculation based on the publication score and the third weight data to obtain the third score; Step S605: Summing the first score, second score, and third score yields the target score for the initial query document.

[0085] Steps S601 to S605, as illustrated in this embodiment, involve acquiring target weight data, which includes first weight data, second weight data, and third weight data. This provides a flexible and targeted weight foundation for subsequent calculations across different dimensions. Next, an aggregation calculation is performed based on the tag level and the first weight data to obtain a first score, which measures the document's positional value within the knowledge structure. An aggregation calculation is then performed based on the similarity score, importance score, and second weight data to obtain a second score, which comprehensively evaluates the document's matching degree with the question and its inherent importance. An aggregation calculation is then performed based on the publication score and the third weight data to obtain a third score, which reflects the document's timeliness value. Finally, the first, second, and third scores are summed to obtain the target score for the initially queried document. This comprehensively integrates information from multiple key dimensions such as knowledge structure, content matching, importance, and timeliness, enabling the target score to more accurately and comprehensively measure the value of the initially queried document, thereby providing a reliable basis for subsequently selecting documents that better meet the requirements.

[0086] In step S601 of some embodiments, the target weight data is preset according to the business scenario. The target weight data includes first weight data, second weight data and third weight data. The first weight data is used to adjust the position value of the document in the knowledge graph, the second weight data is used to adjust the matching degree and intrinsic importance of the document with the question, and the third weight data is used to adjust the timeliness value of the document.

[0087] It's important to note that the importance score is assigned by business personnel based on the actual application scenario for financial documents. For example, financial information published by the company itself, such as prospectuses, can be assigned a higher score, while other information, such as assessment reports provided by financial investment companies, is assigned a lower weight. The publication score is based on the publication date of the financial document and is used to assess its timeliness. Recently published financial documents can be assigned a higher publication score, while older documents can be assigned a lower score.

[0088] In some embodiments, the target score of the initial query document is shown in formula (1); (1); in, The data with the highest weight. This is the second weighted data. This is the third weighted data.

[0089] In other embodiments, in step S604, in addition to the publication score, version score, region score, etc., can be added to enrich the evaluation dimensions of the initial query document.

[0090] Understandably, rich metadata provides the ability to filter documents based on factors such as importance, timeliness, and validity, further improving the question-and-answer effect in complex financial scenarios and allowing for flexible application to different business scenarios.

[0091] In step S504 of some embodiments, the initial query documents can be sorted based on the target score, and then the top few with the highest target scores can be selected to obtain the target query documents. Alternatively, the initial query documents can be filtered based on a preset threshold to obtain the target query documents.

[0092] You can also sort the initial query documents based solely on the target score without filtering to obtain the target query documents.

[0093] Please see Figure 7 In some embodiments, step S106 may include, but is not limited to, steps S701 to S703: Step S701: Generate response content based on the target query document to obtain the original financial response content; Step S702: Obtain the document link of the target query document; Step S703: Insert the document link into the original financial response content to obtain the target financial response content.

[0094] Steps S701 to S703, as illustrated in this embodiment, generate response content based on the target query document to obtain the original financial response content, making the information more organized and readable. Next, the document link of the target query document is obtained and inserted into the original financial response content to obtain the target financial response content, facilitating users to quickly locate the original text and improving information processing efficiency and accuracy.

[0095] In step S701 of some embodiments, a pre-trained large model can be used to generate response content for the target query document. This typically involves extracting key information from the target query document, such as titles, paragraphs, lists, and tables, and organizing them according to a hierarchical structure, for example, using titles as first-level nodes and paragraphs as second-level nodes. This makes the content of the target query document clearer and more organized, facilitating users to quickly understand and obtain the information they need.

[0096] In step S702 of some embodiments, the document link is a unique identifier used to locate and access the specific location of the target query document on the network / financial resource information database. By clicking the link, the user can directly access the original target query document.

[0097] Document links provide a convenient way for the target audience to access the original document, allowing them to view the full content and learn more details by clicking the link. Additionally, document links help verify the source and accuracy of information.

[0098] In step S703 of some embodiments, a document link is inserted into the original financial response content, such as at the end of the document or next to relevant information (e.g., in the form of a footnote). Adding a document link improves the efficiency and convenience of information retrieval and also enhances the credibility and traceability of the response content.

[0099] In step S107 of some embodiments, the final generated target financial response content is presented to the target object in the intelligent customer service consultation window on a specific platform, so that the target object can view the answer corresponding to the original question data.

[0100] In other embodiments, in addition to backtracking document links, "knowledge transmission" paths can also be identified, that is, the target graph tags in the knowledge graph can be displayed. For example, the path "xx policy-new energy-lithium battery-mining" in the knowledge graph can be displayed according to the knowledge path as "xx policy benefits new energy → drives lithium battery demand → affects upstream mining stocks", reducing the black box operation of large model question answering and improving user experience.

[0101] The financial document question-answering method based on a tag-based hierarchical knowledge graph provided in this application can implement business functions by combining different tags according to financial business scenarios. This includes, but is not limited to: (1) Risk warning function: When the retrieved document contains the specified risk tag (such as "significant decline in performance", "case filed for investigation", "xx policy change", etc.), the tag can be a single tag or a combination of tags. When the business risk tag is triggered, the warning reminder is triggered in combination with market data.

[0102] (2) Compliance inspection function: The document to be published is segmented, and the document is searched in the knowledge graph and knowledge base. The policy, regulations, clauses, cases, regions, industries and other tags in the knowledge graph are associated and identified. The question and answer function is used to identify potential violations, generate a compliance report and suggest corrective measures.

[0103] (3) Intelligent investment function: Based on tags such as company, disclosure events, region, and industry, multi-dimensional data fusion is realized to construct investment strategies (such as event-driven investment). (4) Personalized service function: Based on the tags involved in the user's query history, push the latest announcements, research reports, etc. in relevant industries / regions / announcement categories, realize the linkage between user profile and knowledge graph tags, provide accurate services, and increase user stickiness.

[0104] Please see Figure 8 This application also provides a financial document question-answering device based on a tag-hierarchical knowledge graph, which can implement the above-mentioned financial document question-answering method based on a tag-hierarchical knowledge graph. The device includes: Problem data acquisition module 801 is used to acquire the original problem data input by the target object; The knowledge graph construction module 802 is used to construct a tag-hierarchical knowledge graph based on a pre-set financial domain resource information database; The graph tag query module 803 is used to perform tag queries on the original question data based on the tag hierarchy knowledge graph to obtain the target graph tags. The document query module 804 is used to perform document queries on the financial sector resource information database based on the target graph tags to obtain the initial query document; The document filtering module 805 is used to filter the initial query documents to obtain the target query documents; The response content generation module 806 is used to generate text based on the target query document to obtain the target financial response content; The response module 807 is used to respond to the target object based on the content of the target financial response.

[0105] The specific implementation of this tag-based hierarchical knowledge graph-based financial document question answering device is basically the same as the specific implementation of the tag-based hierarchical knowledge graph-based financial document question answering method described above, and will not be repeated here.

[0106] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned financial document question-answering method based on a tag-hierarchical knowledge graph. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0107] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 to implement the tag-based hierarchical knowledge graph-based financial document question-answering method of this application embodiment. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0108] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described financial document question-answering method based on a tag-hierarchical knowledge graph.

[0109] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0110] The financial document question-answering method and apparatus based on a tag-hierarchical knowledge graph provided in this application acquires the original question data input by the target object and constructs a tag-hierarchical knowledge graph based on a preset financial domain resource information database. This systematically organizes domain knowledge and forms a structured knowledge system. Next, the original question data is tagged using the tag-hierarchical knowledge graph to obtain target graph tags, which can quickly associate the question with key nodes in the knowledge graph, improving information matching efficiency. Furthermore, the financial domain resource information database is tagged with the target graph to obtain initial query documents, narrowing the information retrieval scope. Document filtering of the initial query documents yields the target query document, further removing irrelevant information and improving information quality. Finally, text is generated based on the target query document to obtain the target financial response content, and a response is sent to the target object based on this content, effectively improving the efficiency and accuracy of information processing and response.

[0111] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0112] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0113] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0115] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0116] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0117] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0118] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0120] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] The software tools or components not belonging to our company that appear in the embodiments of this application are for illustrative purposes only and do not represent actual use.

[0122] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A financial document question-answering method based on a tag-based hierarchical knowledge graph, characterized in that, The method includes: Obtain the original question data input by the target object; A hierarchical knowledge graph based on a pre-set financial sector resource information database is constructed. The original question data is queried based on a tag-level knowledge graph to obtain the target graph tags. Based on the target map tags, a document query is performed on the financial sector resource information database to obtain the initial query document; The initial query document is filtered to obtain the target query document; Based on the target query document, text is generated to obtain the target financial response content; A response will be sent to the target recipient based on the content of the target financial response; The step of filtering the initial query document to obtain the target query document includes: The similarity score is obtained by calculating the similarity between the initial query document and the original question data; Obtain the tag level of the target map tags; The target score of the initial query document is obtained by weighting the tag level, the similarity score, the importance score of the initial query document, and the publication score. Based on the target score, the initial query document is processed to obtain the target query document; The step of calculating the target score of the initial query document by weighting the tag level, the similarity score, the importance score of the initial query document, and the publication score includes: Obtain target weight data, wherein the target weight data includes first weight data, second weight data, and third weight data; A first score is obtained by performing aggregation calculations based on the label hierarchy and the first weight data; A second score is obtained by aggregating and calculating the similarity score, the importance score, and the second weight data. The third score is obtained by aggregating and calculating the publication score and the third weight data. The target score of the initial query document is obtained by summing the first score, the second score, and the third score.

2. The method according to claim 1, characterized in that, The financial sector resource information database includes financial sector documents; the construction of a tag-based hierarchical knowledge graph based on the pre-set financial sector resource information database includes: Obtain the document tags for the aforementioned financial documents; Generate graph hierarchy child nodes based on the document tags; Generate hierarchical relationships based on preset hierarchical information; Based on the graph hierarchy, the sub-nodes of the graph hierarchy are used to construct the graph, thereby obtaining the tag hierarchy knowledge graph.

3. The method according to claim 2, characterized in that, The document tags for the financial documents are obtained in the following way: Obtain the original financial documents; The original financial document is extracted to obtain financial text data; The financial text data is structured to obtain the financial domain document; Entity extraction is performed on the financial text data to obtain financial entity data; The document tags for the financial documents are generated based on the financial entity data.

4. The method according to any one of claims 1 to 3, characterized in that, The tag query of the original question data based on the tag-hierarchical knowledge graph to obtain the target graph tags includes: The original question data is subjected to intent recognition to obtain the original question intent data; Based on the original question intent data, a graph query is performed on the tag hierarchy knowledge graph to obtain graph node data; The target map label is generated based on the map node data.

5. The method according to any one of claims 1 to 3, characterized in that, The text generation based on the target query document to obtain the target financial response content includes: Based on the target query document, the response content is generated to obtain the original financial response content; Obtain the document link of the target query document; Inserting the document link into the original financial response content yields the target financial response content.

6. A financial document question-answering device based on a tag-based hierarchical knowledge graph, characterized in that, The device includes: The problem data acquisition module is used to acquire the raw problem data input by the target object; The knowledge graph construction module is used to construct a tag-hierarchical knowledge graph based on a pre-set financial sector resource information database; The graph tag query module is used to perform tag queries on the original question data based on the tag hierarchy knowledge graph to obtain the target graph tags; The document query module is used to perform document queries on the financial sector resource information database based on the target map tags to obtain the initial query document; The document filtering module is used to filter the initial query documents to obtain the target query documents; The response content generation module is used to generate text based on the target query document to obtain the target financial response content. The response module is used to respond to the target object based on the content of the target financial response; The step of filtering the initial query document to obtain the target query document includes: The similarity score is obtained by calculating the similarity between the initial query document and the original question data; Obtain the tag level of the target map tags; The target score of the initial query document is obtained by weighting the tag level, the similarity score, the importance score of the initial query document, and the publication score. Based on the target score, the initial query document is processed to obtain the target query document; The step of calculating the target score of the initial query document by weighting the tag level, the similarity score, the importance score of the initial query document, and the publication score includes: Obtain target weight data, wherein the target weight data includes first weight data, second weight data, and third weight data; A first score is obtained by performing aggregation calculations based on the label hierarchy and the first weight data; A second score is obtained by aggregating and calculating the similarity score, the importance score, and the second weight data. The third score is obtained by aggregating and calculating the publication score and the third weight data. The target score of the initial query document is obtained by summing the first score, the second score, and the third score.

7. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 5.

Images