Tracing report generation method and device, electronic equipment and storage medium

Abstract

The invention discloses a traceability report generation method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a target task for a research problem according to the research problem input by a user; creating a traceability knowledge base having an association relationship with the research problem; obtaining original text information of a to-be-executed step for executing the target task; metadata used for representing the logic relation between the original text information and the execution result of the to-be-executed step is determined; storing the original text information, the execution result and the metadata into a traceability knowledge base; and in response to a received traceability request of a user for the task report of the target task, extracting the original text information, the execution result and the metadata from the traceability knowledge base to generate a traceability report. According to the method, the original information, the intermediate execution result and the logic association metadata are synchronously stored in the task generation process, a complete mapping system from a fact manuscript to a final conclusion is constructed, and logic trace reserving and evidence path restoration of artificial intelligence generation content are realized.

Description

Technical Field

[0001] This invention relates to the field of traceability report generation technology, and in particular to a traceability report generation method, a traceability report generation device, an electronic device, and a readable storage medium. Background Technology

[0002] With the widespread application of generative artificial intelligence technology in professional fields such as investment research and consulting, intelligent agents can automatically break down tasks, retrieve information, and generate research reports based on user questions. However, existing report generation technologies still face significant challenges in terms of accuracy and reliability.

[0003] First, traditional generative models often directly output the final text when producing conclusions, failing to record the logical derivation process behind those conclusions. When users encounter specific numerical values ​​or viewpoints in a report, it is difficult to know which original data the conclusion was based on or what processing steps were involved, making it difficult to verify the report's authority.

[0004] Furthermore, in complex task processing, there is a lack of clear logical connections between raw data, intermediate processing steps, and final conclusions. When reports contain factual errors or logical deviations (i.e., "model illusions"), because the system fails to synchronously establish a mapping relationship from "raw materials" to "executed actions" and then to "output results" during the generation phase, users or the system cannot accurately trace back to the error points, nor can they effectively verify the logical consistency of the report content.

[0005] Finally, the traceability process suffers from insufficient efficiency and granularity. Existing simple retrieval technologies typically only provide relevant reference documents, failing to pinpoint specific text blocks and their roles in particular execution steps. This necessitates extensive manual intervention for secondary comparisons during the traceability process, making it difficult to meet the real-time, automated, and in-depth verification requirements of rigorous scenarios such as financial analysis. Summary of the Invention

[0006] The present invention provides a method, apparatus, electronic device, and readable storage medium for generating traceability reports to overcome or at least partially solve the above-mentioned problems.

[0007] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a method for generating a traceability report, including: Based on the research question input by the user, determine the target task for the research question; Create a source knowledge base that is related to the research question; Obtain the raw text information of the steps to be executed for performing the target task; Determine metadata used to characterize the logical relationship between the original text information and the execution result of the step to be executed; The original text information, the execution result, and the metadata are stored in the source tracing knowledge base; In response to receiving a user's request for tracing the source of a task report for the target task, the system extracts the original text information, the execution result, and the metadata from the tracing knowledge base to generate a tracing report.

[0008] Optionally, the source tracing knowledge base includes an original data knowledge base, a subtask data knowledge base, and a metadata knowledge base. The step of storing the original text information, the execution result, and the metadata into the source tracing knowledge base includes: The original text information is stored in the original data knowledge base; The execution result is stored in the subtask data knowledge base; The metadata is stored in the metadata knowledge base.

[0009] Optionally, the original data knowledge base includes a vector database and a structured search index, and the step of storing the original text information into the original data knowledge base includes: Generate a timestamp for the original text information and obtain the source path of the original text information; The original text information is semantically sliced ​​to generate multiple original text blocks; The original text block is converted into a high-dimensional feature vector, and the high-dimensional feature vector is stored in the vector database; The original text block, the timestamp, and the source path are stored in the structured search index.

[0010] Optionally, the step of storing the execution result in the subtask data knowledge base includes: Extract the execution logic description and preset output format corresponding to the step to be executed; The steps to be executed are performed according to the preset output format to obtain intermediate reasoning conclusions that are consistent with the execution logic description. Extract key values ​​and core fact fragments from the intermediate reasoning conclusions, and identify the key values ​​and core fact fragments as structured fact data; The intermediate reasoning conclusions are used to perform vectorization processing to generate conclusion feature vectors; The structured factual data and the conclusion feature vector are stored together in the subtask data knowledge base.

[0011] Optionally, the step of storing the metadata in the metadata knowledge base includes: When performing the steps to be performed, the target tool used to perform the steps to be performed is determined; Record the tool identifier of the target tool; Record the call index between the step to be executed and the original text block corresponding to the step to be executed; Generate a tool execution timestamp that records the tool identifier invocation action; The tool identifier, the call index, and the tool execution timestamp are encapsulated into metadata that characterizes the path mapping relationship between the original text block and the execution result.

[0012] Optionally, the step of generating a source tracing report by extracting the original text information, the execution result, and the metadata from the source tracing knowledge base in response to receiving a user's source tracing request for a task report on the target task includes: In response to receiving a user's selection operation for target data in a task report, and receiving a confirmation instruction from the user for the selection operation, determine whether the selected target data is valid data; When the selected target data is determined to be invalid, a prompt message is displayed to indicate that the selected target data does not have associated tracing results; When the selected target data is determined to be valid data, the original text information, the execution result, and the metadata are extracted from the traceability knowledge base to generate a traceability report.

[0013] Optionally, the step of generating a source tracing report by extracting the original text information, the execution result, and the metadata from the source tracing knowledge base in response to receiving a user's source tracing request for a task report on the target task includes: In response to receiving an automatic execution tracing instruction sent by the user, which indicates that an automatic tracing operation will be performed on the overall task report after the task report is generated, the original text information, the execution result, and the metadata are extracted from the tracing knowledge base to generate a tracing report.

[0014] Optionally, the step of extracting the original text information, the execution result, and the metadata from the source tracing knowledge base to generate a source tracing report includes: Using semantic blocks that characterize the target data or overall task report, the corresponding logical relation metadata is matched from the metadata knowledge base; Based on the logical relationship metadata, source tracing text blocks are extracted from the original data knowledge base, and source tracing execution results corresponding to the source tracing text blocks are extracted from the subtask data knowledge base; The tracing report is generated based on the tracing text block and the tracing execution result.

[0015] Optionally, after the step of generating a traceability report from the traceability knowledge base by extracting the original text information, the execution result, and the metadata in response to receiving a traceability request from a user for a task report on the target task, the method further includes: In response to receiving a verification request sent by a user, which indicates that a verification operation is performed on the source tracing report when the source tracing report is generated, a credibility score is calculated by comparing the semantic deviation between the source tracing text block and the source tracing execution result. When the credibility score is determined to be lower than a preset threshold, the abnormal task node corresponding to the source text block is determined based on the logical relationship metadata. Re-execute the pending steps that are associated with the abnormal task node.

[0016] Secondly, embodiments of this application provide a traceability report generation apparatus, comprising: The target task determination module is used to determine the target task for the research question based on the research question input by the user. The knowledge base creation module is used to create a knowledge base that is related to the research question. The raw text information acquisition module is used to acquire the raw text information of the steps to be executed for performing the target task; Metadata determination module, used to determine metadata that characterizes the logical relationship between the original text information and the execution result of the step to be executed; The source tracing knowledge base storage module is used to store the original text information, the execution result, and the metadata into the source tracing knowledge base; The traceability report generation module is used to extract the original text information, the execution result, and the metadata from the traceability knowledge base to generate a traceability report in response to receiving a user's traceability request for a task report on the target task.

[0017] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0018] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0019] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0020] The embodiments of the present invention have the following advantages: This invention, through its embodiments, determines a target task for a research question input by a user; creates a source knowledge base associated with the research question; acquires the original text information of the steps to be executed for the target task; determines metadata representing the logical relationship between the original text information and the execution results of the steps to be executed; stores the original text information, the execution results, and the metadata in the source knowledge base; and, in response to receiving a source tracing request from a user for a task report on the target task, extracts the original text information, the execution results, and the metadata from the source knowledge base to generate a source tracing report. This achieves simultaneous storage of original information, intermediate execution results, and their logically related metadata during task generation, constructing a complete mapping system from factual drafts to final conclusions, and realizing logical traceability and evidence path restoration of AI-generated content. Attached Figure Description

[0021] Figure 1 This is a flowchart of the steps of a source tracing report generation method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a knowledge base construction process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of another knowledge base construction process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a word-segmentation and source tracing process provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an automatic traceability process provided in an embodiment of the present invention; Figure 6 This is a flowchart illustrating a method for verifying traceability results provided in an embodiment of the present invention; Figure 7 This is a flowchart illustrating a source tracing and query mechanism provided in an embodiment of the present invention; Figure 8 This is a structural block diagram of a traceability report generation device provided in an embodiment of the present invention; Figure 9 This is a hardware structure block diagram of an electronic device provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of a computer-readable medium provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] Reference Figure 1 The diagram illustrates a flowchart of a source tracing report generation method provided in an embodiment of the present invention, which may specifically include the following steps: Step 101: Determine the target task for the research question based on the research question input by the user; Step 102: Create a source knowledge base that is related to the research question; Step 103: Obtain the original text information of the steps to be executed for performing the target task; Step 104: Determine metadata used to characterize the logical relationship between the original text information and the execution result of the step to be executed; Step 105: Store the original text information, the execution result, and the metadata in the source tracing knowledge base; Step 106: In response to receiving a user's tracing request for a task report on the target task, extract the original text information, the execution result, and the metadata from the tracing knowledge base to generate a tracing report.

[0026] The embodiments of the present invention can be applied to a traceability report generation system (hereinafter referred to as the system). The system can be configured with a user interface, through which users can control the system to generate traceability reports and display the traceability reports.

[0027] In a specific implementation, embodiments of the present invention can respond to receiving a research question input by a user, determine a target task for the research question, and perform semantic understanding and intent recognition on the natural language research question input by the user (such as "analyze the future valuation of XXX brand"), and transform it into a structured target that the system can process.

[0028] The target task refers to the structured research objectives that the system, after parsing, needs to accomplish in response to the research problem. In other words, the target task is to complete a task report addressing the research problem, which serves as the basis for all subsequent execution steps. For example, the structured decomposition result of the target task is as follows: Basic Information and Business Review: Subtask 1.1: Obtain historical financial data for the XXX brand (revenue, net profit, and gross profit margin for the past 5 years).

[0029] Subtask 1.2: Search for major strategic announcements and product line changes of the XXX brand over the past year.

[0030] Subtask 1.3: Analyze the market size and growth potential of the segment in which the XXX brand operates.

[0031] Core valuation metric calculation: Subtask 2.1: Calculate the current PE (price-to-earnings ratio), PS (price-to-sales ratio), and PB (price-to-book ratio) valuations for the XXX brand.

[0032] Subtask 2.2: Identify and obtain valuation data for the main comparable companies of the XXX brand (as a replacement target for the tool mapping).

[0033] Subtask 2.3: Calculate the equity value range of the XXX brand based on the DCF (Discounted Cash Flow) model.

[0034] Brand Value and Competitive Analysis: Subtask 3.1: Collect public opinion heat index and consumer sentiment index for brand XXX.

[0035] Subtask 3.2: Search for industry experts' analysis reports on the future competitiveness of the XXX brand.

[0036] Risk and Sensitivity Analysis: Subtask 4.1: Identify the core risk factors that will affect the future profitability of the XXX brand (such as regulatory changes and fluctuations in raw material costs).

[0037] Subtask 4.2: Conduct a sensitivity analysis to assess the impact of changes in key assumptions (such as perpetual growth rate and discount rate) on the valuation results.

[0038] Ultimately, the output of this objective will be a complete valuation report, including a suggested valuation range, supporting data, and key risk warnings.

[0039] In this embodiment of the invention, a traceability knowledge base related to the research question can be created to establish an independent digital storage space for the current specific research question, ensuring data isolation and targeted traceability.

[0040] A knowledge base for tracing origins can be a collection of databases used to uniformly store materials, conclusions, and logical relationships generated throughout the entire process of task execution.

[0041] In this embodiment of the invention, raw text information of the steps to be executed for performing the target task can be obtained. These steps can be specific operational steps necessary to achieve the target task, such as the sub-tasks in the example above. The raw text information can be unprocessed initial text material obtained from an external data source.

[0042] For example, suppose the step to be executed is subtask 1.1 (obtain historical financial data for brand XXX). The original text information can be the plain text converted from the PDF of brand XXX's 2023 annual report. The process of executing the step to be executed involves the system calling the "information extraction tool," searching for the keyword "operating revenue" in the annual report text, locating the "consolidated profit and loss statement" section, and identifying and extracting the corresponding value for 2023 (e.g., 10 billion yuan).

[0043] In a specific implementation, embodiments of the present invention can determine metadata used to characterize the logical relationship between the original text information and the execution result of the step to be executed, so as to establish a causal relationship record between the material and the output, eliminate logical gaps in the generation process, and enable the evidence chain to have closed-loop characteristics.

[0044] The execution result can be an intermediate conclusion or numerical fact produced by the step to be executed after processing the original text information.

[0045] Metadata can be descriptive data that describes the correspondence between raw text and execution results (such as indexes, timestamps, and tool identifiers).

[0046] In this embodiment of the invention, the original text information, the execution result, and the metadata can be stored in the traceability knowledge base to achieve automated traceability of "generation as evidence", and to synchronously solidify static data, dynamic output, and logical chain in physical storage.

[0047] In a specific implementation, embodiments of the present invention can respond to a user's request for tracing the source of a task report for the target task by extracting the original text information, the execution result, and the metadata from the tracing knowledge base to generate a tracing report, thereby making the stored implicit logic explicit and providing the user with a verification report supported by evidence, thus enhancing the transparency of the generated content.

[0048] A task report can be the final text output generated by the system based on the target task.

[0049] A source tracing request can be an instruction triggered automatically by a user or the system, requiring verification of the authenticity and logic of the report content.

[0050] A source tracing report can be a visual verification result that integrates the original evidence, execution logic, and intermediate conclusions.

[0051] This invention, through its embodiments, determines a target task for a research question input by a user; creates a source knowledge base associated with the research question; acquires the original text information of the steps to be executed for the target task; determines metadata representing the logical relationship between the original text information and the execution results of the steps to be executed; stores the original text information, the execution results, and the metadata in the source knowledge base; and, in response to receiving a source tracing request from a user for a task report on the target task, extracts the original text information, the execution results, and the metadata from the source knowledge base to generate a source tracing report. This achieves simultaneous storage of original information, intermediate execution results, and their logically related metadata during task generation, constructing a complete mapping system from factual drafts to final conclusions, and realizing logical traceability and evidence path restoration of AI-generated content.

[0052] Based on the above embodiments, modified embodiments of the above embodiments are proposed. It should be noted that, in order to keep the description brief, only the differences from the above embodiments are described in the modified embodiments.

[0053] In an optional embodiment of the present invention, the source tracing knowledge base includes an original data knowledge base, a subtask data knowledge base, and a metadata knowledge base. The step of storing the original text information, the execution result, and the metadata into the source tracing knowledge base includes: The original text information is stored in the original data knowledge base; The execution result is stored in the subtask data knowledge base; The metadata is stored in the metadata knowledge base.

[0054] In this embodiment of the invention, the original text information can be stored in the original data knowledge base to preserve the source evidence of data-driven decision-making and ensure that the system can retrieve the most original and unaltered draft materials when generating a traceability report.

[0055] The source material knowledge base is a subset of the source knowledge base, specifically used for storing, indexing, and managing the underlying source material.

[0056] In this embodiment of the invention, the execution result can be stored in the subtask data knowledge base to record the intermediate conclusions or values ​​produced by each specific execution step, thus materializing the result of the "processing procedure" for direct extraction later.

[0057] The subtask data knowledge base is a subset of the source knowledge base, specifically used to store the structured data, reasoning conclusions, and semantic features generated by each subtask.

[0058] In this embodiment of the invention, the metadata can be stored in the metadata knowledge base to establish and solidify the logical "index bridge" between the original material and the execution result, ensuring that the system knows which result was derived from which piece of material.

[0059] The metadata knowledge base is a subset of the source knowledge base, and is a central node specifically used to store and maintain data flow paths and logical mapping relationships.

[0060] This optional embodiment achieves the classification and solidification of original materials, intermediate conclusions and logical relationships through a distributed storage mechanism that separates the three databases, thereby ensuring the efficiency of data retrieval and the accuracy of logical reconstruction during the process of generating traceability reports.

[0061] Optionally, the original data knowledge base includes a vector database and a structured search index, and the step of storing the original text information into the original data knowledge base includes: Generate a timestamp for the original text information and obtain the source path of the original text information; The original text information is semantically sliced ​​to generate multiple original text blocks; The original text block is converted into a high-dimensional feature vector, and the high-dimensional feature vector is stored in the vector database; The original text block, the timestamp, and the source path are stored in the structured search index.

[0062] In this embodiment of the invention, a timestamp of the original text information can be generated, and the source path of the original text information can be obtained, so as to establish a definite time coordinate and source attribution for the data, ensuring that each piece of material has time validity proof and path traceability when restoring the evidence chain in the future.

[0063] A timestamp can be a precise record of the time when the original text was acquired or processed.

[0064] The source path can be the specific storage address or access link of the original text information in external storage or network environment.

[0065] In this embodiment of the invention, the original text information can be semantically sliced ​​to generate multiple original text blocks, thereby transforming long documents into fine-grained semantic units, improving the accuracy of subsequent retrieval, and ensuring that specific segments can be accurately located during source tracing.

[0066] Semantic slicing is an operation that uses natural language processing techniques to break down text into independent segments based on its logical structure or semantic meaning.

[0067] The original text block is the smallest text unit generated after slicing, which has independent semantic expression capabilities.

[0068] In this embodiment of the invention, the original text block can be converted into a high-dimensional feature vector and stored in the vector database to achieve semantic digitization of the text through mathematical modeling, and to support the subsequent rapid matching of relevant evidence based on semantic similarity using vector retrieval technology.

[0069] High-dimensional feature vectors are numerical arrays representing the semantic features of text, and are usually generated by pre-trained embedding models.

[0070] Vector databases are database systems specifically designed for the efficient storage, indexing, and retrieval of vector data.

[0071] In this embodiment of the invention, the original text block, the timestamp, and the source path can be stored in the structured search index to establish a hard association index between the text content and its attribute information (time, path), ensuring that the complete evidence entity can be quickly restored based on the metadata during the tracing stage.

[0072] A structured search index is a data structure that maps text content to its corresponding attribute fields (such as time and source) to support precise queries.

[0073] This optional embodiment achieves a precise closed loop from semantic matching to physical location of original materials through a dual mapping mechanism of vector features and structured indexes, which greatly improves the retrieval efficiency and information restoration completeness of the source tracing report in the evidence extraction stage.

[0074] Optionally, the step of storing the execution result in the subtask data knowledge base includes: Extract the execution logic description and preset output format corresponding to the step to be executed; The steps to be executed are performed according to the preset output format to obtain intermediate reasoning conclusions that are consistent with the execution logic description. Extract key values ​​and core fact fragments from the intermediate reasoning conclusions, and identify the key values ​​and core fact fragments as structured fact data; The intermediate reasoning conclusions are used to perform vectorization processing to generate conclusion feature vectors; The structured factual data and the conclusion feature vector are stored together in the subtask data knowledge base.

[0075] In this embodiment of the invention, the execution logic description and preset output format corresponding to the step to be executed can be extracted to provide the system with standardized processing procedures and output specifications, ensuring that the artificial intelligence is logically controlled and outputs are consistent when executing specific sub-tasks.

[0076] The execution logic description is a predefined set of instructions, algorithm formulas, or reasoning rules that specifies how to process data.

[0077] The default output format specifies the data field types, quantities, and organization (such as JSON format, numeric type, string type, etc.) that the execution result should include.

[0078] In this embodiment of the invention, the steps to be executed can be performed according to the preset output format to obtain intermediate reasoning conclusions that are consistent with the execution logic description, so as to transform the underlying instructions into specific analysis results and produce interim conclusions with practical business significance.

[0079] Intermediate reasoning conclusions are the original text outputs generated after the execution of the steps to be executed, which include the logical deduction process for the steps to be executed and the final answer.

[0080] In this embodiment of the invention, key numerical values ​​and core factual fragments can be extracted from the intermediate reasoning conclusions, and the key numerical values ​​and core factual fragments can be identified as structured factual data. This allows for the extraction of the most essential evidence from lengthy textual conclusions, transforming unstructured information into precise data that can be directly compared and verified.

[0081] Key figures are quantitative data such as amounts, proportions, and dates mentioned in the conclusion.

[0082] Core fact fragments are key qualitative descriptions or phrases that support the conclusion.

[0083] Structured fact data is a collection of core information that has been cleaned and formatted so that it can be directly parsed and retrieved by computers.

[0084] Taking the output of "Subtask 1.1: Obtain historical financial data for brand XXX" as an example: Intermediate inference conclusion (unstructured text): "By retrieving the 2023 annual report of the XXX brand, the system confirmed in the consolidated profit statement section that the brand's operating revenue reached 12.5 billion yuan in fiscal year 2023, an increase of 15% year-on-year, and net profit reached 1.2 billion yuan." Key figures extracted: 12.5 billion yuan, 15%, 1.2 billion yuan.

[0085] Extract key factual information: fiscal year 2023, operating revenue, year-on-year growth, and net profit.

[0086] The data is identified as structured factual data (internal system format): {"Year": "2023", "Indicator": "Operating Revenue", "Value": "125", "Unit": "100 Million Yuan", "Growth Rate": "15%"} In this embodiment of the invention, the intermediate reasoning conclusion can be used to perform vectorization processing to generate a conclusion feature vector, so as to transform the text conclusion into a vector representation in a mathematical space, supporting rapid matching and logical verification through semantic similarity in the subsequent source tracing stage.

[0087] Vectorization is the process of converting text into a numerical array using a deep learning model.

[0088] The conclusion feature vector represents a high-dimensional numerical vector of the semantic features of the intermediate reasoning conclusion.

[0089] In this embodiment of the invention, the structured factual data and the conclusion feature vector can be stored together in the subtask data knowledge base to complete the physical persistence of the execution results. Through the storage of "numerical + semantic" dual dimensions, refined data support is provided for generating traceability reports.

[0090] This optional embodiment achieves the transformation of execution results from unstructured text to high-precision digital evidence by performing structured fact extraction and semantic vectorization processing on intermediate inference conclusions, providing underlying data assurance for data comparison and consistency verification in subsequent source tracing reports.

[0091] Optionally, the step of storing the metadata in the metadata knowledge base includes: When performing the steps to be performed, the target tool used to perform the steps to be performed is determined; Record the tool identifier of the target tool; Record the call index between the step to be executed and the original text block corresponding to the step to be executed; Generate a tool execution timestamp that records the tool identifier invocation action; The tool identifier, the call index, and the tool execution timestamp are encapsulated into metadata that characterizes the path mapping relationship between the original text block and the execution result.

[0092] In this embodiment of the invention, when executing the step to be executed, the target tool used to execute the step can be determined to clarify the physical carrier of the task execution, and provide a clear attribution of responsibility for the output logic of the subsequent traceability conclusion (whether it comes from a search tool, a calculation tool or an analysis model).

[0093] Target tools: Software modules or application programming interfaces (APIs) within the system that are called to perform specific functions, such as "financial calculators", "text matching engines", or "industry analysis agents".

[0094] In this embodiment of the invention, the tool identifier of the target tool can be recorded to lock the execution tool through a unique identity code, ensuring that the traceability report can accurately show which specific processing module generated the conclusion.

[0095] A tool identifier is a unique identifier (ID) or name assigned to each target tool.

[0096] In this embodiment of the invention, the call index between the step to be executed and the original text block corresponding to the step to be executed can be recorded to establish a precise mapping between the execution action and the input material, thus solving the problem of locating "which part in a massive amount of drafts triggered the step".

[0097] A call index is a pointer that records the path of data flow, typically containing the storage address of the original text block in the database or its offset in the source document.

[0098] In this embodiment of the invention, a tool execution timestamp that records the tool identifier invocation action can be generated to provide temporal evidence of the execution process, prove the order and immediacy of the conclusion output, and enhance the rigor of the evidence chain.

[0099] Tool execution timestamps are precise records of the system time when a target tool begins execution or completes a calculation.

[0100] In this embodiment of the invention, the tool identifier, the call index, and the tool execution timestamp can be encapsulated into metadata that characterizes the path mapping relationship between the original text block and the execution result. For example, multiple data items can be integrated into a logical unit (such as a JSON object), thereby packaging fragmented process information into a complete logical record, which is stored in the metadata knowledge base as a bridge connecting "input (original material)" and "output (execution result)".

[0101] The path mapping relationship records the causal logic of the entire path of data from the input end, including which tools it passes through, when it is processed, and what results it produces.

[0102] This optional embodiment constructs a logical flow map of the entire conclusion generation process by synchronously recording and encapsulating the execution tool identifier, material call index, and execution timestamp, thereby achieving transparent traceability and precise mapping of the artificial intelligence decision-making path.

[0103] In practical applications, in order to ensure that the large model achieves full traceability and verifiability when generating research reports, the embodiments of the present invention simultaneously build a traceability knowledge base during the report generation process, and after the generation is completed, use the traceability engine to reconstruct the evidence chain and verify consistency, thereby ensuring the factual accuracy and source transparency of the report content.

[0104] refer to Figure 2 and Figure 3 , Figure 2 This is a schematic diagram of a knowledge base construction process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of another knowledge base construction process provided in an embodiment of the present invention; 1. Dynamic construction of the source knowledge base and three core knowledge structures: When a report generation task starts, the system synchronously creates a corresponding source knowledge base. During report generation, tools and sub-agents may be used to retrieve large amounts of information from various data sources, such as internal and external research report queries, announcement data queries, news searches, and internal meeting queries. The source knowledge base collects and stores information related to the execution of these sub-tasks after these tools and sub-agents have finished. The execution entity is the data access module, responsible for format unification, time standardization, deduplication, and context binding of data from different sources. This module performs preliminary validation and classification of the input while collecting data, ensuring the usability and consistency of the raw information for subsequent storage. Finally, the acquired knowledge content is divided into three core categories for corresponding storage: The raw data knowledge base stores raw information input from external sources, such as text content like announcements, research reports, news, and public opinion, raw data obtained from data interfaces, and related content like timestamps and source paths. Text content is semantically sliced ​​and vectorized before being written to a vector library to support semantic retrieval and similarity calculation; it is also simultaneously written to the Elasticsearch (ES) index for precise keyword matching, enabling dual-channel retrieval. Data content is not vectorized; it is only stored in the keyword search engine.

[0105] Subtask Data Knowledge Base: This base stores the structured results, data tables, and key conclusions generated by each sub-agent during the research process. This data is also stored in a dual-path manner: its numerical values ​​and key facts are written to Elasticsearch to ensure accurate and fast queries; its detailed textual descriptions or context are generated as vectors and stored in a vector library to supplement semantic understanding.

[0106] Metadata knowledge base: Records metadata structures such as subtask objectives, execution logs, tool call information, associated tags, timestamps, execution results, and session indexes. The three types of knowledge bases work together during task execution: the original data knowledge base provides factual evidence, the subtask data knowledge base records the generation logic and intermediate conclusions, and the metadata knowledge base maintains the task context and invocation path. Together, they constitute an independent knowledge indexing system for the current research task, providing a solid data foundation for subsequent tracing, verification, and retraining.

[0107] In an optional embodiment of the present invention, the tracing request includes: A request to trace the source of the target data in the task report; or, This is used to characterize the traceability request that automatically generates a traceability report when the task report is determined to be completed.

[0108] The step of generating a source tracing report by extracting the original text information, the execution result, and the metadata from the source tracing knowledge base in response to receiving a user's task report for the target task includes: In response to receiving a user's selection operation for target data in a task report, and receiving a confirmation instruction from the user for the selection operation, determine whether the selected target data is valid data; When the selected target data is determined to be invalid, a prompt message is displayed to indicate that the selected target data does not have associated tracing results; When the selected target data is determined to be valid data, the original text information, the execution result, and the metadata are extracted from the source tracing knowledge base to generate a source tracing report; or, In response to receiving an automatic execution tracing instruction sent by the user, which indicates that an automatic tracing operation will be performed on the overall task report after the task report is generated, the original text information, the execution result, and the metadata are extracted from the tracing knowledge base to generate a tracing report.

[0109] Optionally, in embodiments of the present invention, when receiving a user's selection operation for target data in a task report and receiving a confirmation instruction from the user for the selection operation, it can determine whether the selected target data is valid data, so as to perform pre-filtering in the scenario of manual tracing, and avoid the system performing invalid retrieval on non-critical information (such as titles and blank lines) by verifying whether the selected content has tracing attributes.

[0110] The target data can be specific text, numerical values, or chart conclusions selected by the user through highlighting on the front-end interface.

[0111] Valid data can refer to report content that has corresponding metadata records and original evidence support in the traceability knowledge base.

[0112] The prompt message can be an explanatory text displayed by the system when a user selects content that cannot be traced.

[0113] Optionally, in response to receiving an automatic tracing instruction sent by a user, which indicates that an automatic tracing operation will be performed on the entire task report after the task report is generated, the system can extract the original text information, the execution result, and the metadata from the tracing knowledge base to generate a tracing report, thereby providing a fully automated tracing solution. This allows the system to actively complete the mapping of the entire conclusion and evidence without human intervention after the report is generated, providing the user with a complete "causal audit working paper".

[0114] Automatic execution of traceability commands can be user-preset or real-time triggered control signals that require a full logical scan of the report.

[0115] In its specific implementation, the traceability knowledge base of this invention provides two different application methods to support different information verification query scenarios: refer to Figure 4 , Figure 4 This is a schematic diagram of a word-segmentation and source tracing process provided in an embodiment of the present invention; 1. Word-highlighting and source tracing mode (manually triggered): The triggering method is in response to the user selecting target data in the report front end, such as any text, data or conclusion, and then clicking the source tracing button. The system automatically extracts the highlighted text and generates a source tracing request, performs a search in the current dialogue knowledge base, and returns the original source of the information.

[0116] The output includes the returned evidence path tree, original text excerpts, and data sources, for users to manually verify and cite.

[0117] The word-by-word tracing mode enables precise location and instant evidence verification of specific content in the report through on-demand manual tracing, greatly reducing the time cost for users to perform secondary manual verification of generated content and providing a compliant evidence chain that can be directly cited.

[0118] refer to Figure 5 , Figure 5 This is a schematic diagram of an automatic traceability process provided in an embodiment of the present invention; 2. Automatic source tracing mode (system triggered): The user chooses whether to enable it. When enabled, the system will automatically execute it after the task report is generated.

[0119] Execution logic: The task report is divided into multiple semantic blocks by topic by the NLP model, and source retrieval is automatically performed on each block. Then the system summarizes the source retrieval results of each block, reconstructs the causal path of "raw data → tool call → semantic aggregation → report output" in combination with the task log, and draws a causal relationship graph.

[0120] The output includes a complete "Automatic Source Report," which details: the source file ID, citation type (external / internal), timestamp, credibility score (T-Score), and data version for each text segment. If the same conclusion comes from multiple source cross-validation, the system will automatically label the multiple sources.

[0121] The automatic tracing mode achieves causal transparency throughout the entire generation process through globally automated logical reconstruction and multi-source cross-validation, ensuring that each task report is accompanied by a complete digital audit working paper, significantly improving the overall authority and transparency of the system output.

[0122] For example, whether it's the word-selection mode or the automatic mode, their commonality lies in the fact that their core evidence-gathering path is completely consistent. Both must first obtain a "semantic block" (selected by the user or segmented by the system); both use this semantic block to match the corresponding logical relationship metadata in the metadata knowledge base, which is the only key to finding evidence. Both ultimately retrieve the source text block (draft) from the "original database" and the source execution result (processing logic) from the "subtask database"; both rely on the "three-database linkage" storage closed loop completed synchronously during the execution of the previous tasks.

[0123] The difference lies in: 1. Triggering mechanism and interaction logic: The word highlighting and source tracing mode is initiated by a human. The system is in a passive state, activated by responding to the user's selection operation (target data selection) on the front end, which is a "point-to-point" random sampling.

[0124] The automatic source tracing mode is triggered by system events. It typically runs automatically after the report generation task is completed, with the system in a proactive state, representing a routine audit that provides "full coverage".

[0125] 2. Granularity and scale of the objects being processed: The word-selection source tracing mode targets a single piece of data. The logical path is: select a segment → query a segment → generate a local report. That is, the word-selection mode operates on a "point-to-point" basis: input a semantic block → match a set of metadata → generate a local piece of evidence.

[0126] The automatic source tracing mode object is the overall task report. It includes a preliminary action—using an NLP model to divide the entire report into multiple semantic blocks according to the topic, and then performing source tracing in batches for each block. That is, the automatic mode is a "face-to-face" mode of inputting N semantic blocks → batch matching N sets of metadata → summarizing and generating a causal path graph covering the entire report.

[0127] 3. Depth and form of output results: The source tracing report generated by the word highlighting and source tracing mode is a local evidence tree. It focuses on showing "how this piece of data came about", making it convenient for users to quickly cite or verify specific points.

[0128] The source tracing report generated by the automatic source tracing mode is a complete causal path diagram. It not only presents evidence, but also aggregates the source tracing results of all semantic blocks, reconstructing the complete logical path from "raw data → tool invocation → semantic aggregation → report output".

[0129] 4. Future feature extensions: The word highlighting and source tracing mode usually stops at the display of information and mainly serves "human verification".

[0130] Automatic source tracing mode: This is the prerequisite for "rigorous mode". Only after completing full source tracing can the system further calculate the credibility score through semantic deviation comparison, thereby triggering automatic rollback and re-execution of logic.

[0131] In an optional embodiment of the present invention, after the step of extracting the original text information, the execution result, and the metadata from the source tracing knowledge base to generate a source tracing report in response to receiving a user's source tracing request for a task report on the target task, the method further includes: In response to receiving a verification request sent by a user, which indicates that a verification operation is performed on the source tracing report when the source tracing report is generated, a credibility score is calculated by comparing the semantic deviation between the source tracing text block and the source tracing execution result. When the credibility score is determined to be lower than a preset threshold, the abnormal task node corresponding to the source text block is determined based on the logical relationship metadata. Re-execute the pending steps that are associated with the abnormal task node.

[0132] In this embodiment of the invention, when an automatic tracing instruction sent by a user is received, which indicates that an automatic tracing operation will be performed on the overall task report after the task report is generated, a verification operation for the tracing report can be performed. This allows the system to perform a deep semantic review of these evidence chains after the tracing report has found all the "original evidence" and "execution process", thereby determining whether the conclusion is true and reliable and preventing model illusion from misleading the user.

[0133] Verification requests can be control signals sent by users through interface switches (such as enabling "strict mode") or system preset instructions. Their function is to instruct the system to perform consistency comparisons on the generated traceability content.

[0134] Verification operations can be specific audit actions performed by the system, including a series of calculation processes such as semantic discrepancies between calculation conclusions and evidence, determining whether numerical values ​​are accurate, and whether logic is consistent.

[0135] In this embodiment of the invention, a credibility score can be calculated by comparing the semantic deviation between the source tracing text block and the source tracing execution result. This allows for the quantitative detection of whether the conclusion deviates from the original basis, thereby identifying possible model "illusions" or logical errors.

[0136] Semantic bias is the degree of inconsistency between the source text block (fact source) and the execution result (output) in terms of semantic meaning, numerical logic, or factual tendency.

[0137] Credibility score is a value that represents the degree of consistency between the execution result and its source of evidence. The higher the score, the more reliable the conclusion.

[0138] In this embodiment of the invention, when the credibility score is determined to be lower than a preset threshold, the abnormal task node corresponding to the source text block can be determined based on the logical relationship metadata, so that when the conclusion is found to be unreliable, the source link that caused the error can be automatically locked using the logical path.

[0139] The preset threshold is the minimum credibility value set by the system.

[0140] An abnormal task node is a specific execution step that is determined to have an output logic error or a distorted data reference.

[0141] In this embodiment of the invention, pending steps that are associated with the abnormal task node can be re-executed to trigger the system rollback and error correction mechanism. The report content can be corrected by recalculating the damaged node to ensure the accuracy of the final output.

[0142] Re-execution means that the system automatically rolls back the erroneous step and runs the subtask again based on the corrected parameters or a more accurate matching algorithm.

[0143] refer to Figure 6 , Figure 6 This is a flowchart illustrating a method for verifying traceability results provided in an embodiment of the present invention; For example, the "rigorous mode" (the verification submodule for automatic traceability) is triggered by the user's choice of whether to enable it. When enabled, it is executed automatically by the system after the report is generated and the automatic traceability is completed.

[0144] Execution logic: After the automatic source tracing report is generated, the system performs consistency verification on the original report content. Specifically, it compares the report paragraphs with the evidence text for factual consistency, numerical deviations, and semantic relationships. If illusions or citation errors are detected in the report content, the corresponding task is marked and the rollback module is triggered. The system automatically rolls back and re-executes the corresponding sub-tasks and subsequent tasks to improve the accuracy and reliability of the information in the final report.

[0145] The rigorous mode introduces a closed-loop correction mechanism of "verification-discovery-recalculation" to automatically intercept and self-correct model "illusions" and factual errors, fundamentally ensuring the consistency and high reliability of the final delivery report at both the numerical and logical levels.

[0146] This optional embodiment achieves a technological upgrade from simple "information tracing" to "automatic logical quality inspection" by introducing a reliable quantitative evaluation and an abnormal node locking mechanism, ensuring accurate backtracking and automatic error correction when distorted report content is detected.

[0147] In an optional embodiment of the present invention, the step of extracting the original text information, the execution result, and the metadata from the source tracing knowledge base to generate a source tracing report includes: Using semantic blocks that characterize the target data or overall task report, the corresponding logical relation metadata is matched from the metadata knowledge base; Based on the logical relationship metadata, source tracing text blocks are extracted from the original data knowledge base, and source tracing execution results corresponding to the source tracing text blocks are extracted from the subtask data knowledge base; The tracing report is generated based on the tracing text block and the tracing execution result.

[0148] In this embodiment of the invention, semantic blocks representing the target data or overall task report can be used to match corresponding logical relation metadata from the metadata knowledge base, thereby enabling the logical location of the causal path involved in the generation process of the content that the user is interested in (word highlighting or full text) as an index.

[0149] A semantic block is a digital information unit with a specific meaning obtained by vectorizing target data (local) or task report (overall).

[0150] Logical relationship metadata is a record stored in the metadata knowledge base that describes the mapping relationship between raw materials, processing tools, and execution results.

[0151] In this embodiment of the invention, the source text block can be extracted from the original data knowledge base based on the logical relationship metadata, and the source execution result corresponding to the source text block can be extracted from the subtask data knowledge base. With the guidance of metadata, the two elements of the evidence chain supporting the current conclusion, namely the factual draft (input) and the processing product (output), can be accurately restored.

[0152] Source text blocks are specific original text fragments extracted from the original database that are directly related to the current task.

[0153] The tracing execution result is an intermediate conclusion extracted from the subtask database and generated based on the tracing text block through specific logical processing.

[0154] In this embodiment of the invention, the tracing report can be generated based on the tracing text block and the tracing execution result, so as to integrate the scattered logical connections, original evidence and intermediate products into a verification file that can be intuitively read by the user.

[0155] A source tracing report is an analytical document that shows the entire path to the conclusion, the related evidence, and the logical derivation process.

[0156] For example, assuming the task report states "Company XXX earned 10 billion this year," the source tracing report will display the following to the user: Original evidence (source text block): Directly post a screenshot or text from page 10 of the annual report: "Net profit attributable to the parent company in 2025 was 10.02 billion yuan".

[0157] Calculation process (tracing the execution result): The system shows that it rounded "100 billion" from "100.2 billion".

[0158] Reliability rating (credibility score): "99 points" indicates that the conclusion matches the original text and the data is reliable.

[0159] refer to Figure 7 , Figure 7 This is a flowchart illustrating a source tracing and query mechanism provided in an embodiment of the present invention; During the report generation or review phase, when performing source tracing, the source tracing engine is responsible for retrieval and evidence construction. The inputs to this process include: a fragment of the report content to be traced, and the dialogue ID of the current research task. The source tracing engine first extracts numerical content from the input for individual matching, and then comprehensively queries the source tracing knowledge base using the following three methods: Regular expression matching: precisely finds content that exactly matches the search term. Only applicable to precise targeting of numeric content.

[0160] Keyword retrieval: Similar to RAG, the large model first derives keywords from the input query content, and then performs keyword retrieval to obtain relevant content.

[0161] Vector similarity calculation: The content to be queried is converted into a semantic vector, and then a similar vector search is performed for semantic matching and contextual association of semantic text content.

[0162] After querying the content corresponding to the source of the content to be traced from the original data knowledge base and the subtask data knowledge base using the above search method, it is possible not only to locate the external source of the content to be traced, such as original data like announcements, research reports, and news sentiment, but also to determine whether the information comes from the original content of the data source or new content generated during the subtask summary. For example, opinions on specific events generally come from the original content, while comparisons of differences in opinions on events within the industry are generally new content generated during the subtask summary. These located locations can serve as "anchor points" for information, and combined with the metadata knowledge base, the entire process from external input to model inference output can be reconstructed. This process includes tool call records, data extraction paths, field mapping, and context merging information, thereby restoring the information flow path from data source to inference generation.

[0163] Ultimately, the tracing engine integrates two layers of evidence: original data and subtask data. Combined with task logs and tool call chains from the metadata knowledge base, it generates an evidence path tree. The final output includes evidence answers directly responding to queries, original text fragments, task nodes, tool sources, and a credibility score (T-Score), forming a traceable, verifiable, and citationable evidence structure.

[0164] To enable those skilled in the art to better understand the embodiments of the present invention, an example is used below to illustrate the embodiments of the present invention.

[0165] In their daily investment research work, researchers often need to analyze market hotspots and generate data-driven research reports within a short period of time. When researchers input a research topic (such as "generate an A-share industry rotation strength analysis report for September") into the "DeepResearch" platform, the system will first create an empty traceability knowledge base to carry the data flow in the subsequent generation process.

[0166] After DeepResearch begins report generation, it intelligently breaks down the research objectives into corresponding sub-task graphs and executes them. During sub-task execution, DeepResearch invokes various sub-agents, such as research report search, market data acquisition, internet search, and internal meeting search, to obtain a large amount of information. At this time, the source knowledge base writes information such as the input, output, data source, tool path, and time of each invocation into the original data knowledge base, sub-task data knowledge base, and metadata knowledge base to ensure data isolation and efficient retrieval between different research tasks. As report generation progresses, this initially empty knowledge base is continuously "filled," gradually forming a structured data system containing a complete chain of evidence. When DeepResearch research report generation ends, the source knowledge base has also completed its establishment process.

[0167] Once the report is generated, the system automatically triggers the source tracing and verification process. The source tracing engine analyzes and verifies the research report segment by segment. Using natural language processing technology, the system divides the report content into multiple semantic blocks and performs source tracing for each block. Based on the source tracing results for each semantic block, the system synthesizes and outputs an automatic source tracing report. This report clearly displays the source file, generation tool, timestamp, and credibility score (T-Score) for each paragraph and data indicator in the report, providing verifiable, traceable, and reproducible technical support for investment research. When the user enables "rigorous mode," the system automatically compares and verifies the consistency between the original research report and the source content after automatic source tracing. After summarizing any detected factual biases, citation errors, or misleading information, the system locates the corresponding subtasks and triggers a rollback mechanism, re-executing the corresponding subtasks and subsequent tasks, thereby improving the reliability and quality of the final research report.

[0168] In addition to automatic source tracing, if researchers have doubts about a certain section of content, they can directly select the report text to trigger the word-by-word source tracing function. For example, after selecting "the average standard deviation of the industry's September price change from 2015 to 2024 was 28.2%", the system will immediately retrieve the corresponding data source in the source tracing knowledge base and display the original research report, calculation process, and market data fragments cited in the conclusion.

[0169] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0170] Reference Figure 8The diagram illustrates a structural block diagram of a traceability report generation device provided in an embodiment of the present invention, which may specifically include the following modules: The target task determination module 801 is used to determine the target task for the research question based on the research question input by the user. The traceability knowledge base creation module 802 is used to create a traceability knowledge base that is related to the research question. The original text information acquisition module 803 is used to acquire the original text information of the steps to be executed for performing the target task; Metadata determination module 804 is used to determine metadata that characterizes the logical relationship between the original text information and the execution result of the step to be executed; The traceability knowledge base storage module 805 is used to store the original text information, the execution result and the metadata into the traceability knowledge base; The traceability report generation module 806 is used to extract the original text information, the execution result and the metadata from the traceability knowledge base to generate a traceability report in response to receiving a traceability request from a user for a task report on the target task.

[0171] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0172] In addition, embodiments of the present invention also provide an electronic device, such as... Figure 9 As shown, it includes a processor 901, a communication interface 902, a memory 903, and a communication bus 904, wherein the processor 901, the communication interface 902, and the memory 903 communicate with each other through the communication bus 904. Memory 903 is used to store computer programs; When the processor 901 executes the program stored in the memory 903, it implements any of the traceability report generation methods described in the above embodiments: The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0173] The communication interface is used for communication between the aforementioned terminal and other devices.

[0174] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0175] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0176] like Figure 10 As shown, in another embodiment of the present invention, a computer-readable storage medium 1001 is also provided, which stores instructions that, when run on a computer, cause the computer to execute the traceability report generation method described in the above embodiment.

[0177] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described traceability report generation method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0178] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0179] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0180] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0181] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for generating a source tracing report, characterized in that, include: Based on the research question input by the user, determine the target task for the research question; Create a source knowledge base that is related to the research question; Obtain the raw text information of the steps to be executed for performing the target task; Determine metadata used to characterize the logical relationship between the original text information and the execution result of the step to be executed; The original text information, the execution result, and the metadata are stored in the source tracing knowledge base; In response to receiving a user's request for tracing the source of a task report for the target task, the system extracts the original text information, the execution result, and the metadata from the tracing knowledge base to generate a tracing report.

2. The method according to claim 1, characterized in that, The source tracing knowledge base includes a raw data knowledge base, a subtask data knowledge base, and a metadata knowledge base. The step of storing the raw text information, the execution result, and the metadata into the source tracing knowledge base includes: The original text information is stored in the original data knowledge base; The execution result is stored in the subtask data knowledge base; The metadata is stored in the metadata knowledge base.

3. The method according to claim 2, characterized in that, The original data knowledge base includes a vector database and a structured search index. The step of storing the original text information into the original data knowledge base includes: Generate a timestamp for the original text information and obtain the source path of the original text information; The original text information is semantically sliced ​​to generate multiple original text blocks; The original text block is converted into a high-dimensional feature vector, and the high-dimensional feature vector is stored in the vector database; The original text block, the timestamp, and the source path are stored in the structured search index.

4. The method according to claim 3, characterized in that, The step of storing the execution result in the subtask data knowledge base includes: Extract the execution logic description and preset output format corresponding to the step to be executed; The steps to be executed are performed according to the preset output format to obtain intermediate reasoning conclusions that are consistent with the execution logic description. Extract key values ​​and core fact fragments from the intermediate reasoning conclusions, and identify the key values ​​and core fact fragments as structured fact data; The intermediate reasoning conclusions are used to perform vectorization processing to generate conclusion feature vectors; The structured factual data and the conclusion feature vector are stored together in the subtask data knowledge base.

5. The method according to claim 4, characterized in that, The step of storing the metadata in the metadata knowledge base includes: When executing the step to be executed, determine the target tool used to execute the step to be executed; Record the tool identifier of the target tool; Record the call index between the step to be executed and the original text block corresponding to the step to be executed; Generate a tool execution timestamp that records the tool identifier invocation action; The tool identifier, the call index, and the tool execution timestamp are encapsulated into metadata that characterizes the path mapping relationship between the original text block and the execution result.

6. The method according to claim 5, characterized in that, The step of generating a source tracing report by extracting the original text information, the execution result, and the metadata from the source tracing knowledge base in response to receiving a user's task report for the target task includes: In response to receiving a user's selection operation for target data in a task report, and receiving a confirmation instruction from the user for the selection operation, determine whether the selected target data is valid data; When the selected target data is determined to be invalid, a prompt message is displayed to indicate that the selected target data does not have associated tracing results; When the selected target data is determined to be valid data, the original text information, the execution result, and the metadata are extracted from the traceability knowledge base to generate a traceability report.

7. The method according to claim 5, characterized in that, The step of generating a source tracing report by extracting the original text information, the execution result, and the metadata from the source tracing knowledge base in response to receiving a user's task report for the target task includes: In response to receiving an automatic execution tracing instruction sent by the user, which indicates that an automatic tracing operation will be performed on the overall task report after the task report is generated, the original text information, the execution result, and the metadata are extracted from the tracing knowledge base to generate a tracing report.

8. The method according to claim 6 or 7, characterized in that, The step of extracting the original text information, the execution result, and the metadata from the source tracing knowledge base to generate a source tracing report includes: Using semantic blocks that characterize the target data or the overall task report, the corresponding logical relation metadata is matched from the metadata knowledge base; Based on the logical relationship metadata, source tracing text blocks are extracted from the original data knowledge base, and source tracing execution results corresponding to the source tracing text blocks are extracted from the subtask data knowledge base; The tracing report is generated based on the tracing text block and the tracing execution result.

9. The method according to claim 8, characterized in that, After the step of extracting the original text information, the execution result, and the metadata from the source tracing knowledge base to generate a source tracing report in response to receiving a user's task report request for the target task, the method further includes: In response to receiving a verification request sent by a user, which indicates that a verification operation is performed on the source tracing report when the source tracing report is generated, a credibility score is calculated by comparing the semantic deviation between the source tracing text block and the source tracing execution result. When the credibility score is determined to be lower than a preset threshold, the abnormal task node corresponding to the source text block is determined based on the logical relationship metadata. Re-execute the pending steps that are associated with the abnormal task node.

10. A traceability report generation device, characterized in that, include: The target task determination module is used to determine the target task for the research question based on the research question input by the user. The knowledge base creation module is used to create a knowledge base that is related to the research question. The raw text information acquisition module is used to acquire the raw text information of the steps to be executed for performing the target task; Metadata determination module, used to determine metadata that characterizes the logical relationship between the original text information and the execution result of the step to be executed; The source tracing knowledge base storage module is used to store the original text information, the execution result, and the metadata into the source tracing knowledge base; The traceability report generation module is used to extract the original text information, the execution result, and the metadata from the traceability knowledge base to generate a traceability report in response to receiving a user's traceability request for a task report on the target task.

11. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in claims 1-7.

12. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the method as described in claims 1-7.

Images