Method for generating a summary of a financial document

Abstract

The present invention aims to generate context-aware and user-demand-aligned financial document summaries by seamlessly integrating large language models (LLM) with structured knowledge graphs. The system employs a three-stage approach: first, it parses financial documents to construct a comprehensive knowledge graph encompassing entities, relationships, and quantitative indicators; second, it queries the knowledge graph and utilizes LLMs for information synthesis, interpreting user intent to generate customized subsummaries; finally, it integrates these subsummaries into a unified final summary, elucidating cross-document insights and incorporating a self-reflection mechanism to ensure accuracy and coherence. This innovative solution overcomes the inherent shortcomings of existing summarization techniques, including their static nature, limited cross-document analysis capabilities, and frequent misalignment with user-specific priorities. By providing dynamic, structured, and adaptable financial insights, this process effectively meets the diverse analytical needs of users.

Description

Technical Field

[0001] This invention relates to the field of financial information analysis technology, specifically to a method for generating financial document summaries based on the collaboration of a large-scale language model and a dynamic knowledge graph. Background Technology

[0002] Financial documents (such as SEC filings, annual reports, and earnings call transcripts) are characterized by: dense domain-specific terminology, complex inter-entity relationships (e.g., relationships between subsidiaries and stakeholders), and massive amounts of quantitative data. Traditional summarization tools, relying on rigid rules or unstructured natural language processing (NLP) techniques, struggle to parse these complexities. Furthermore, a diverse user base, ranging from investors to compliance officers, requires customized summaries for different objectives (e.g., risk assessment, competitive analysis), necessitating summaries that dynamically adapt to specific scenarios.

[0003] Large Language Models (LLMs) are gaining increasing attention in the field of financial document summarization, with their core advantage lying in their ability to process massive amounts of unstructured data and generate coherent outputs. However, existing solutions using LLMs to process complex financial texts often suffer from illusions (i.e., the model generates inaccurate or fabricated details). In the financial field, where accuracy is paramount, such inaccuracies can damage the credibility of summaries and pose significant risks to the decision-making process.

[0004] Methods exist for summarizing regulatory documents through structured processing, but these methods lack real-time knowledge graph updates and user intent recognition mechanisms, limiting their adaptability to dynamic financial environments and failing to meet the diverse needs of stakeholders. Furthermore, there are schemes that utilize static knowledge graphs to map market trends and systematically identify correlations. However, the static nature of these graphs limits the integration of the latest information with document-specific details, potentially leading to outdated or incomplete summaries. Schemes focusing on the conciseness of automatic summarization often fail to provide customized outputs for specific user needs (such as risk assessment or compliance verification).

[0005] In summary, these existing studies all share common shortcomings: they are prone to inaccuracies, lack flexibility in data processing, and have limited ability to personalize solutions to meet the diverse needs of different users. These limitations highlight the necessity of developing a more reliable solution—one that ensures accuracy, dynamic responsiveness, and a user-centric design philosophy in financial document summarization. Summary of the Invention

[0006] This invention proposes a document summarization method that addresses the aforementioned shortcomings by synergistically leveraging the core capabilities of large-scale language models and knowledge graphs. In the first stage, financial documents are parsed with the assistance of a large-scale language model to construct a dynamic knowledge graph. This graph systematically encodes entities such as enterprises and financial instruments, relationships such as equity and transactions, and indicators such as revenue and debt-to-equity ratio. This structured representation lays a solid foundation for subsequent analysis. In the second stage, the system interprets user intent (derived from explicit user queries or inferred user preferences) and retrieves relevant subgraphs from the knowledge graph accordingly. Subsequently, the large-scale language model synthesizes these subgraphs into sub-summaries, ensuring precise alignment with specific user analytical needs (whether risk factor analysis or performance indicator analysis). In the final stage, relying on the inherent relational and temporal frameworks of the knowledge graph, the sub-summaries are integrated into a comprehensive summary, clarifying cross-document patterns (such as quarterly performance trends and competitive landscape). Furthermore, the system incorporates a self-reflection loop mechanism: the large-scale language model performs factual accuracy, logical coherence, and user intent checks on the comprehensive summary. Figure 1 Consistency checks and iterative optimizations are conducted to improve the reliability and relevance of the final abstract.

[0007] This invention proposes three innovative improvements to overcome the shortcomings of existing technologies.

[0008] First, by integrating large-scale language models with dynamically constructed knowledge graphs, this invention enables structured, machine-readable representations of financial entities (e.g., firms, financial instruments), relationships (e.g., equity hierarchies), and quantitative indicators (e.g., risk ratios). Unlike methods that use static knowledge graphs for market trend detection, this invention constructs domain-specific knowledge graphs in real-time during document parsing, continuously incorporating entities, relationships, and indicators from new documents and maintaining an immediate responsiveness to changes in the financial environment. This allows the system to capture document-specific contextual information (e.g., mergers and acquisitions described in 10-K filings) and inter-document relationships (e.g., liquidity trends in quarterly reports), capabilities not found in rule-based methods or purely large-scale language model-driven approaches.

[0009] Secondly, the user intent recognition mechanism of this invention can dynamically interpret user queries (e.g., "summarize the compliance risks of Company X") to guide knowledge graph traversal. This ensures that the summary is tailored to the user role (e.g., investor and auditor) and core needs (e.g., risk assessment and performance benchmark analysis). In contrast, generic summaries generated by existing technologies often overemphasize irrelevant parts or omit key details.

[0010] Third, the self-reflection mechanism uses a large language model to verify the factual accuracy, logical coherence, and consistency with user intent of the output. For example, the system verifies whether the summary related to "debt obligations" covers all entities in the knowledge graph related to high-risk loans. This iterative optimization process mitigates illusions and ensures compliance with domain-specific constraints (such as regulatory reporting standards).

[0011] On the one hand, the present invention provides a method for generating financial document summaries, comprising the following steps: Parse one or more financial documents to extract entities, relationships, and metrics; Construct a knowledge graph representing the entities, relationships, and indicators; Identify user intent by interpreting user queries or preset preferences; Based on the identified user intent, sub-summaries are generated by querying the knowledge graph and utilizing a large language model. Based on the cross-document relationships identified in the knowledge graph, the sub-summaries are integrated into a final summary; and The final summary is validated through a self-reflection process driven by a large language model.

[0012] In embodiments of the present invention, the knowledge graph incorporates temporal metadata to support cross-document trend analysis.

[0013] In embodiments of the present invention, the entities include enterprises, financial instruments, subsidiaries, and stakeholders; the relationships include equity, transactions, equity tiers, and mergers and acquisitions; and the metrics include revenue, debt-to-equity ratio, risk ratio, and liquidity trend.

[0014] In an embodiment of the present invention, generating multiple sub-summaries by querying the knowledge graph and utilizing a large language model based on the identified user intent includes: Based on the identified user intent, relevant sub-graphs are retrieved from the knowledge graph; The related sub-graphies are synthesized into sub-summaries using the large-scale language model.

[0015] In embodiments of the present invention, verifying the final summary through a self-reflection process driven by a large language model includes: The large-scale language model is used to assess the factual accuracy, logical coherence, and user intent of the final summary. Figure 1 Consistency is iteratively verified.

[0016] On the other hand, the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the method according to the present invention.

[0017] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method according to the present invention. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which: Figure 1 This is a flowchart of a financial document summary generation method according to an embodiment of the present invention. Figure 2 An exemplary knowledge graph built from financial documents is shown, demonstrating how entities, relationships, and metrics are organized into a network structure that provides a structured foundation for summary generation. Figure 3 It shows how user queries are mapped to specific subgraphs in the knowledge graph, providing guidance for the summary generation process. Figure 4 This is a block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions of the present invention, exemplary embodiments of the present invention are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] Where there is no conflict, the various embodiments of the present invention and the features thereof may be combined with each other.

[0021] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0022] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Terms such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0023] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having the meaning consistent with their meaning in the context of the relevant art and the invention, and will not be interpreted as having an idealized or overly formal meaning unless expressly so defined herein.

[0024] This invention provides a multi-step process for generating summaries of financial documents by utilizing large-scale language models and knowledge graphs. This process aims to address the inherent complexity of financial terminology and output summaries tailored to specific user needs. Figure 1 As shown, the process includes three core stages:

[0025] Stage 100, Document Preprocessing Stage. In embodiments of the present invention, the document preprocessing stage includes document parsing and knowledge graph construction.

[0026] The core objective of this phase is to transform unstructured financial documents (such as SEC filings and annual reports) into machine-readable structured knowledge graphs, providing a precise data foundation for subsequent summary generation.

[0027] Large-scale language models parse financial documents, extracting key entities, relationships, and metrics. These elements are systematically organized into a knowledge graph, providing a structured, machine-readable representation of the document content. In embodiments of this invention, entities include, for example, companies (e.g., "Company A," "Company B"), financial instruments (e.g., stocks, bonds), subsidiaries, and stakeholders; relationships include, for example, equity, transactions, equity tiers, and mergers and acquisitions; and metrics include, for example, revenue, risk ratio, and liquidity trends.

[0028] In embodiments of the present invention, the knowledge graph incorporates temporal metadata to support cross-document trend analysis.

[0029] In this invention, "time metadata" refers to a specific time tag associated with an entity or metric. In the "Revenue Growth Comparison" embodiment described below, the time metadata includes the fiscal year extracted from the annual report, such as 2021 or 2022. In the "Compliance Monitoring" embodiment described below, the time metadata includes the identified quarter, such as the third quarter of 2022. By incorporating knowledge graphs into the time metadata, cross-document trend analysis can be supported. This structure enables the concatenation of data scattered across different documents along a time dimension, for example, accurately calculating the 25% increase in Company A's revenue from $400 million in 2021 to $500 million in 2022, and generating a cross-document trend analysis summary based on this.

[0030] In the document preprocessing stage of this invention, the knowledge graph is updated in real time through the instant parsing and structured reorganization of input financial documents using a large-scale language model. Specifically, when receiving unstructured documents such as SEC filings or annual reports, the large-scale language model simultaneously identifies and extracts specific entities (such as newly established subsidiaries), dynamic relationships (such as newly disclosed M&A transactions), and quantitative indicators (such as quarterly revenue data). This extraction process is concurrent with the knowledge graph construction process. The extracted elements are encoded into the graph structure in real time, thereby constructing or updating a domain-specific knowledge graph while parsing the document. This dynamic construction mechanism enables the capture of contextual information specific to the document, such as details of specific M&A transactions described in 10-K filings or liquidity trend changes in quarterly reports, ensuring that the content of the knowledge graph remains synchronized with the latest input documents. In this way, this invention overcomes the problem of outdated or incomplete information caused by the inability to update static knowledge graphs in a timely manner, ensuring that the generated summary accurately reflects the dynamic changes in the financial environment and the details of specific documents.

[0031] Stage 200, Summary Generation Stage. In an embodiment of the present invention, the summary generation stage includes: step S201, identifying user intent.

[0032] Large-scale language models interpret user queries or preset preferences, clarify users' specific information needs (such as risk factors, performance indicators, compliance details, etc.), and ensure that the generated summaries match the user's core needs.

[0033] For example, users can directly input their requirements, such as "compare the revenue growth of Company A and Company B in 2021-2022".

[0034] If the user does not explicitly query, their intent can be inferred from historical interaction data or preset roles.

[0035] In an embodiment of the present invention, the summary generation stage further includes: step S202, generating a sub-summary.

[0036] For each document, data related to the identified user intent is obtained by querying the knowledge graph (sub-graph), and then the large language model synthesizes this data into a concise and focused sub-summary.

[0037] In an embodiment of the present invention, the abstract generation stage further includes: step S203, aggregating sub-abstracts to generate a final abstract.

[0038] Based on knowledge graph-based entity associations and temporal metadata, this system identifies relationships between sub-summaries of multiple documents and aggregates these sub-summaries. Knowledge graphs facilitate the identification of relationships between documents (such as trend changes across multiple reports). A large-scale language model integrates these core insights to generate a logically coherent final summary.

[0039] Phase 300, Abstract Confirmation Phase.

[0040] The final summary undergoes a self-reflection process, in which a large language model evaluates its factual accuracy, logical coherence, and consistency with user intent, and optimizes the output as needed to improve the quality and relevance of the summary.

[0041] In embodiments of the present invention, the final summary can be iteratively verified through three core dimensions: 1. Factual accuracy: verifying whether the summary content is consistent with the entities, relationships, and indicators in the knowledge graph; 2. Logical coherence: checking whether the sentence logic and information arrangement of the summary conform to the analysis logic of financial documents; 3. User intent. Figure 1 Consistency: Confirm whether the summary accurately matches the user's needs. If not, iterate and correct it.

[0042] The financial sector demands extremely high levels of accuracy. A single large-scale language model is prone to "illusion" (generating fictitious details), which may lead to decision-making risks. A self-reflection mechanism, through linkage and verification with a knowledge graph, significantly reduces factual errors, ensures that the summary meets regulatory reporting standards, and enhances the credibility of financial decisions. This is a quality assurance that a single large-scale language model solution cannot achieve.

[0043] Advantages or improvements of this invention compared to existing technologies

[0044] Compared to existing summary generation methods, this invention achieves a series of significant improvements, providing enhanced functionality and higher accuracy for financial document analysis. The construction of a dynamic knowledge graph enables the system to capture complex relationships and connections that are unattainable by unstructured natural language processing (NLP) methods, thereby improving the depth and accuracy of the summaries. This structured representation facilitates a refined understanding of financial scenarios, overcoming the limitations of static keyword extraction techniques. Furthermore, the method of this invention possesses the ability to interpret and adapt to user intent, ensuring that the summaries accurately match specific objectives (such as risk assessment or compliance monitoring), addressing a common deficiency in existing methods—which often generate generic outputs lacking customized relevance.

[0045] This invention leverages the relational and temporal structures of knowledge graphs to aggregate sub-summaries from multiple documents, aiding in the identification of key trends and patterns crucial for comprehensive financial analysis (such as quarterly performance changes and competitive landscape). This cross-document analysis capability significantly differs from the isolated methods of traditional systems. Furthermore, by incorporating a self-reflection mechanism, large language models can iteratively optimize their output, reducing the risk of factual errors and ensuring the coherence of summaries and consistency with user intent. This quality assurance process represents a significant advancement over summarization techniques solely based on large language models (which are prone to errors and inconsistencies), thus setting a new benchmark for the reliability of financial summaries.

[0046] To further illustrate the implementation effects of the present invention, the specific implementation methods of the present invention will be described below in conjunction with two financial field scenarios: "revenue growth comparison" and "compliance monitoring".

[0047] Example 1: Revenue Growth Comparison

[0048] In this embodiment, the system imports multiple annual reports, extracts "company (A / B), fiscal year (2021 / 2022), and revenue data" through large-scale language model parsing, and then processes them according to... Figure 2 Knowledge graphs are constructed in a manner that... For example... Figure 3 As shown, based on the user intent "generate a summary comparing the revenue performance of Company A and Company B in 2021-2022", relevant sub-graphs are retrieved ( Figure 3(The relevant subgraphs are highlighted with solid black lines). The large language model calculates the revenue growth rates of the two companies based on the relevant subgraphs (A: (500-400) / 400=25%; B: (300-250) / 250=20%), and then generates customized summaries: for example, "Company A's revenue increased from $400 million to $500 million in 2021 to 2022, an increase of 25%, while Company B's revenue increased from $250 million to $300 million in the same period, an increase of 20%". The large language model then compares with the knowledge graph to confirm the accuracy of "$400M→$500M (25%)" and "$250M→$300M (20%)", and checks whether the summaries match the query requirement of "revenue growth comparison" - ultimately confirming that the summaries are both accurate and relevant to the requirements.

[0049] The above process accurately extracted the revenue data of companies A and B for two years. The knowledge graph clearly linked "company-year-revenue" and the generated summary focused on "growth comparison" without any redundant information.

[0050] The generated summaries are concise and customizable, supporting investment analysis by highlighting performance trends.

[0051] Example 2: Compliance Monitoring

[0052] In this embodiment, the system processes regulatory filings, extracts compliance-related entities (such as disclosure information and penalty records) using a large language model, and organizes them into a knowledge graph. The large language model generates summaries that identify compliance issues (e.g., "Company C disclosed in the third quarter of 2022 that it was fined $2 million for violating environmental regulations"). A self-reflection step ensures the coherence and relevance of the summaries.

[0053] The summaries generated through the above process focus on compliance, providing key insights for regulatory oversight and helping to quickly identify compliance issues.

[0054] Figure 4 This is a block diagram of an electronic device provided in an embodiment of the present invention.

[0055] Reference Figure 4 The present invention provides an electronic device comprising: at least one processor 401; at least one memory 402; and one or more I / O interfaces 403 connected between the processor 401 and the memory 402; wherein the memory 402 stores one or more computer programs executable by the at least one processor 401, the one or more computer programs being executed by the at least one processor 401 to enable the at least one processor 401 to perform the above-described method for self-calibrating an omnidirectional radiation field.

[0056] The modules in the aforementioned electronic devices can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0057] This invention also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor / processor core, implements the aforementioned financial document digest generation method. The computer-readable storage medium can be volatile or non-volatile.

[0058] This invention also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described financial document digest generation method.

[0059] Each module in the above system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0060] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0061] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0062] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0063] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0064] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0065] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0066] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0067] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0068] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0069] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A method for generating financial document summaries, comprising the following steps: Parse one or more financial documents to extract entities, relationships, and metrics; Construct a knowledge graph representing the entities, relationships, and indicators; Identify user intent by interpreting user queries or preset preferences; Based on the identified user intent, sub-summaries are generated by querying the knowledge graph and utilizing a large language model. Based on the cross-document relationships identified in the knowledge graph, the sub-summaries are integrated into the final summary; as well as The final summary is validated through a self-reflection process driven by a large language model.

2. The method according to claim 1, wherein, The knowledge graph incorporates temporal metadata to support cross-document trend analysis.

3. The method according to claim 1, wherein, The entities mentioned include enterprises, financial instruments, subsidiaries, and stakeholders; The relationships mentioned include equity, transactions, equity tiers, and merger and acquisition transactions; The metrics include revenue, debt-to-equity ratio, risk ratio, and liquidity trend.

4. The method according to claim 1, wherein, Based on the identified user intent, multiple sub-summaries are generated by querying the knowledge graph and utilizing a large language model, including: Based on the identified user intent, relevant sub-graphs are retrieved from the knowledge graph; The related sub-graphies are synthesized into sub-summaries using the large-scale language model.

5. The method according to claim 1, wherein, The final summary was validated through a self-reflection process driven by a large language model, including: The large-scale language model is used to iteratively verify the factual accuracy, logical coherence, and consistency with user intent of the final summary.

6. An electronic device, comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, wherein, The computer program, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images