An evidence chain-based intelligent report generation method and system

Abstract

The embodiment of the specification discloses a kind of evidence chain-based intelligent report generation method and system.Therein, method includes the structural modeling of report template;Determine the task unit corresponding to target report generation task, associate material data to task unit;Evidence extraction is carried out to the material data associated with task unit;At least one filling candidate value is detected based on source evidence feature Conflicts to obtain conflict detection data, and the confidence of filling candidate value is determined based on pre-set reliability model;Fill target filling value to report template, generate target report;And slot is rendered as traceable data node, to complete the structured rendering and traceable management of slot.This embodiment of the specification can carry out evidence extraction and automatic conflict resolution, and realizes data fine traceability.

Description

Technical Field

[0001] This application relates to the field of automatic report generation technology, specifically to an intelligent report generation method and system based on a chain of evidence. Background Technology

[0002] With the deepening of enterprise digital transformation, the demand for automated generation of various task reports such as investigation reports, audit reports, and legal opinions is becoming increasingly urgent. Currently, automatic report generation often employs placeholder replacement technology based on Word templates or large model generation technology. However, placeholder replacement technology typically obtains field values ​​from a single, pre-defined structured data source, failing to handle multi-document scenarios. That is, when the data required for the report is scattered across multiple unstructured documents, it cannot automatically extract and link to specific sources, resulting in a lack of traceability in the generated results, making it difficult for reviewers to verify the data's authenticity. While large model generation technology can handle multi-source documents, the generated content is a black-box output; users cannot trace the original source of a specific field value, only obtaining macro-level information about the parameter library, which also fails to meet the field-by-field traceability requirements of rigorous audit scenarios. Furthermore, in real-world scenarios, the same field may have multiple candidate values ​​in different documents; for example, a company's registered capital may differ in the business registration form, articles of association, and capital verification report. Placeholder replacement techniques lack mechanisms for conflict detection, classification, and disambiguation among multiple candidate values. While some large model generation techniques propose cross-document verification mechanisms, their conflict resolution primarily relies on rule matching and authoritative source priority strategies, making it difficult to handle subtle differences in numerical fields. Furthermore, current technologies generate static reports, resulting in a poor editing and review experience and a lack of synchronization and traceability mechanisms. Therefore, there is an urgent need for a report generation method with capabilities for evidence extraction, automated conflict disambiguation, and traceable editing. Summary of the Invention

[0003] This specification provides an intelligent report generation method and system based on a chain of evidence, the technical solution of which is as follows:

[0004] In a first aspect, embodiments of this specification provide an intelligent report generation method based on a chain of evidence, comprising: acquiring a report template; performing structured modeling on the report template, the structured modeling including determining slots and slot attributes; a slot being a data object corresponding to a field to be filled in the report template; acquiring material data; determining a task unit corresponding to a target report generation task; associating the material data with the task unit; based on slot attributes, performing evidence extraction on the material data associated with the task unit, the evidence extraction including retrieval and recall and structured extraction, to obtain at least one candidate value for filling the slot and source evidence features corresponding to the candidate value; based on the source evidence features, performing conflict detection on at least one candidate value to obtain conflict detection data, and determining the confidence level of the candidate value based on a pre-set confidence model, the conflict detection data and confidence level being used to determine the target value for filling the slot; filling the target value into the report template to generate a target report; and rendering the slot as a traceable data node to complete the structured rendering and traceable management of the slot.

[0005] Secondly, embodiments of this specification provide an intelligent report generation system based on a chain of evidence, comprising: a slot module for acquiring a report template and performing structured modeling on the report template, the structured modeling including determining slots and slot attributes; a slot is a data object corresponding to a field to be filled in the report template; a task determination module for acquiring material data, determining a task unit corresponding to a target report generation task, and associating the material data with the task unit; an evidence extraction module for performing evidence extraction on the material data associated with the task unit based on slot attributes, the evidence extraction including retrieval and recall and structured extraction, to obtain at least one candidate value for filling the slot and source evidence features corresponding to the candidate value; a filling determination module for performing conflict detection on at least one candidate value for filling based on the source evidence features to obtain conflict detection data, and determining the confidence level of the candidate value for filling based on a pre-set confidence model, the conflict detection data and confidence level being used to determine the target filling value for the slot; and a report generation module for filling the target filling value into the report template to generate a target report; and rendering the slot as a traceable data node to complete the structured rendering and traceable management of the slot.

[0006] The beneficial effects of the technical solutions provided in some embodiments of this specification include at least the following:

[0007] The embodiments in this specification can perform structured modeling of report templates and define slots, which can upgrade traditional placeholders to intelligent field definitions, providing a solid foundation for subsequent evidence extraction, conflict resolution, and traceable editing, and enabling the reusability, maintainability, and intelligence of report templates;

[0008] Furthermore, the embodiments of this specification can also associate material data with the task unit corresponding to the target report generation task, that is, an independent session can be created for each target report generation task, and the data of different target report generation tasks are completely isolated and do not interfere with each other. This supports multiple users to process multiple tasks at the same time, and the embodiments of this specification significantly improve the parallel processing capability of report generation.

[0009] Furthermore, the embodiments of this specification can also extract evidence of material data associated with task units based on slot attributes. That is, the embodiments of this specification use a two-stage evidence extraction mechanism including retrieval and structured extraction to associate each filling candidate value with the corresponding source evidence. Reviewers can view the original source of any field with one click without manual backtracking of documents, thus realizing fine-grained traceability of report generation results and process data at the field level.

[0010] Furthermore, the embodiments of this specification can also perform conflict detection on candidate filling values ​​based on source evidence features, and determine the confidence level of candidate filling values ​​based on a pre-set confidence model. Conflict detection data and confidence levels are used to determine the target filling value for the slot. The embodiments of this specification can automatically process multiple candidate values ​​from multiple material documents through conflict detection and confidence level determination, intelligently integrate multi-source data, automatically identify and disambiguate conflicts, reduce manual review costs, and improve the accuracy of target filling value identification. Moreover, the embodiments of this specification can render slots as traceable data nodes, and implement a synchronization mechanism through data nodes to ensure document consistency and traceability of document editing. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram illustrating an application scenario of the intelligent report generation method based on the chain of evidence provided in this manual.

[0013] Figure 2 This is a flowchart illustrating the intelligent report generation method based on the chain of evidence provided in this manual.

[0014] Figure 3 This is a flowchart illustrating the process of extracting evidence from material data, as provided in this instruction manual.

[0015] Figure 4 This is a flowchart illustrating the process for determining the confidence level of candidate values ​​to be filled, as provided in this manual.

[0016] Figure 5 This is a flowchart illustrating the process for determining the target fill value of a slot, as provided in this manual.

[0017] Figure 6 This is a flowchart illustrating the process of rendering slots as traceable data nodes, as provided in this manual.

[0018] Figure 7 This is a schematic diagram of the structure of the intelligent report generation system based on the chain of evidence provided in this specification.

[0019] Figure 8 This is a schematic diagram of the structure of an electronic device provided in this specification. Detailed Implementation

[0020] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings.

[0021] The terms "first," "second," etc., in the description, claims, and accompanying drawings are used to distinguish different objects and not to describe a particular order. Furthermore, the term "comprising" and any variations thereof are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0022] This specification provides a method for generating intelligent reports based on a chain of evidence through several embodiments. The executing entity of this method can be the intelligent report generation system based on a chain of evidence provided in the embodiments of this invention.

[0023] Before this specification elaborates on the intelligent report generation method based on the chain of evidence in conjunction with one or more embodiments, it first introduces the application scenarios of this intelligent report generation method based on the chain of evidence.

[0024] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of an intelligent report generation method based on a chain of evidence provided in an embodiment of the present invention. In this embodiment, the intelligent report generation system 100 based on a chain of evidence can be integrated into an electronic device, such as a terminal or a server. The terminal can be a mobile phone, tablet computer, smart Bluetooth device, laptop computer, or personal computer (PC); the server can be a single server or a server cluster composed of multiple servers.

[0025] In some embodiments, the evidence chain-based intelligent report generation system 100 can also be integrated into multiple electronic devices. For example, the evidence chain-based intelligent report generation system 100 can be integrated into multiple servers, and the evidence chain-based intelligent report generation method of this application can be implemented by multiple servers.

[0026] In some embodiments, the server may also be implemented as a terminal. The terminal may be a mobile phone, tablet computer, smart Bluetooth device, laptop computer, or personal computer (PC), etc. The terminal includes a central processing unit (CPU), a graphics processing unit (GPU), memory, storage devices, a network communication module, sensors, a display screen, a battery and power management module, etc.

[0027] For example, refer to Figure 1 The electronic device may include a server 110, a storage terminal 120, etc. The storage terminal 120 stores different report templates and material data, etc. The server 110 and the storage terminal 120 communicate with each other, which will not be described in detail here.

[0028] The server 110 may include a processor and memory. The server 110 can acquire a report template, perform structured modeling on the report template (including determining slots and slot attributes), where a slot is the data object corresponding to the field to be filled in the report template; acquire material data, determine the task unit corresponding to the target report generation task, and associate the material data with the task unit; based on slot attributes, perform evidence extraction on the material data associated with the task unit (including retrieval and structured extraction) to obtain at least one candidate value for filling the slot and the source evidence features corresponding to the candidate value; based on the source evidence features, perform conflict detection on at least one candidate value to obtain conflict detection data, and determine the confidence level of the candidate value based on a pre-set confidence model; the conflict detection data and confidence level are used to determine the target value for filling the slot; fill the report template with the target value to generate the target report; and render the slot as a traceable data node to complete the structured rendering and traceable management of the slot.

[0029] It should be noted that, Figure 1The schematic diagram of the intelligent report generation system based on the chain of evidence shown is merely an example. The intelligent report generation system and scenario based on the chain of evidence described in this embodiment are for the purpose of more clearly illustrating the technical solutions of this embodiment and do not constitute a limitation on the technical solutions provided by this embodiment. As those skilled in the art will know, with the evolution of the intelligent report generation system based on the chain of evidence and the emergence of new scenarios, the technical solutions provided by this embodiment are also applicable to similar technical problems.

[0030] Please see Figure 2 , Figure 2 This is a flowchart illustrating a smart report generation method based on a chain of evidence provided in an embodiment of the present invention. This smart report generation method based on a chain of evidence can be... Figure 1 The evidence chain-based intelligent report generation system 100 shown is executed. This evidence chain-based intelligent report generation method may include at least the following steps:

[0031] 200. Obtain the report template and perform structured modeling on the report template. Structured modeling includes determining the slots and slot attributes.

[0032] In this embodiment, the report template can be a document frame containing placeholders for fields to be filled, used to define the structure, format, and filling positions of the target report. Each placeholder corresponds to a slot, and the placeholders can be used for data extraction, conflict checking, and filling operations of the fields to be filled in subsequent steps. The slot can be the data object corresponding to the field to be filled in the report template.

[0033] In some embodiments, the report template is structured and modeled, including: determining the metadata corresponding to the report template, the metadata including at least the template name, applicable scenarios and version number; determining the fields to be filled in the report template, defining slots for the fields to be filled, and the slots corresponding to several slot attributes; determining the slot group, the slot group including several semantically corresponding slots, the slot group being used for batch extraction and logical verification.

[0034] This embodiment can define slots for each field to be filled in the report template. Each slot can include several structured slot attributes, which may include a unique slot identifier, slot data type (text, numeric, date, enumeration, table, etc.), whether it is required, validation rules (regular expressions, numerical range, logical constraints, etc.), extraction configuration data (search keywords, extraction strategy, maximum number of candidates, etc.), front-end display configuration (input control type, prompt text, unit, etc.), and the chapter identifier.

[0035] The embodiments in this specification can perform structured modeling of report templates and define slots, upgrading traditional placeholders to intelligent field definitions, providing a solid foundation for subsequent evidence extraction, conflict resolution, and traceable editing, and enabling reusability, maintainability, and intelligence of report templates.

[0036] 210. Obtain material data, determine the task unit corresponding to the target report generation task, and associate the material data with the task unit.

[0037] In this embodiment, the material data can be the set of original documents received by the system corresponding to the target report generation task. The task unit can be an independent storage space created by the system for a single target report generation task. Each time a new target report generation task is created in this embodiment, the system can generate a corresponding task identifier, record the creation time, initialize the task status to in progress, and associate the material data, the report template selected by the user, and other information with the task unit.

[0038] This system can generate different target reports based on different user needs, with each target report corresponding to a different target report generation task. Users can upload corresponding material data based on the current target report generation task A. For example, based on the current target report generation task A, users can upload material data, which may include multiple documents, such as PDF business registration documents, Word articles of association, Excel financial statements, and scanned copies. The uploaded material data can be bound to the task unit corresponding to the current target report generation task A and stored in the independent storage space corresponding to that task unit, ensuring that documents from different tasks are not mixed up.

[0039] In some embodiments, acquiring material data, determining the task unit corresponding to the target report generation task, and associating the material data with the task unit includes: acquiring material data corresponding to the target report generation task; establishing a session corresponding to the target report generation task; the session is used to isolate and store the material data, record file metadata corresponding to the material data; and is also used to manage the task data corresponding to the target report generation task, so as to associate the task data with the session.

[0040] In this embodiment, the session corresponding to the target report generation task may include session ID, creation time, task status (in progress, completed, archived), associated report template ID, etc.; the file metadata corresponding to the material data may include file name, type, size, upload time, document ID, etc.

[0041] This embodiment associates material data with the task unit corresponding to the target report generation task. That is, it can create an independent session for each target report generation task. User-uploaded material files (PDF, Word, Excel, images, etc.) are stored in isolation by session, and file metadata is recorded. Intermediate data within the session, such as extraction results, draft versions, and chart screenshots, are all associated with the session, facilitating auditing and rollback. In this embodiment, data from different target report generation tasks are completely isolated and do not interfere with each other. It supports multiple users processing multiple tasks simultaneously, significantly improving the parallel processing capability of report generation.

[0042] 220. Based on slot attributes, perform evidence extraction on the material data associated with the task unit. Evidence extraction includes retrieval and recall as well as structured extraction, to obtain at least one candidate value for filling the slot and the source evidence features corresponding to the candidate value.

[0043] In some embodiments, slot attributes may include at least the extraction of configuration data. See also... Figure 3 , Figure 3 This is a schematic diagram of the process for evidence extraction from material data provided in an embodiment of the present invention. Based on slot attributes, evidence extraction is performed on the material data associated with the task unit. Evidence extraction includes retrieval and recall as well as structured extraction, to obtain at least one candidate value for filling the slot and the source evidence features corresponding to the candidate value, including:

[0044] 300. Determine the retrieval query corresponding to the slot based on the extracted configuration data;

[0045] 310. Input the search query into the search enhancement service, which is used to retrieve several paragraphs corresponding to the search query from the material data associated with the task unit;

[0046] 320. Based on the retrieval query, perform similarity retrieval in the material data through retrieval enhancement services to obtain several paragraphs similar to the retrieval query; generate an evidence field for each paragraph;

[0047] 330. Input several paragraphs and slot attributes into the feature extraction service. The feature extraction service is used to perform field recognition on each paragraph in order to extract the field values ​​corresponding to the slots and obtain a list of candidate values ​​corresponding to the slots.

[0048] In this embodiment, the evidence field includes at least the source document identifier of the paragraph, the source document name, the source document type, the paragraph identifier, the paragraph content, the position information of the paragraph in the source document, and the similarity score between the paragraph and the search query.

[0049] In this embodiment, the candidate value list includes at least one filler candidate value. The filler candidate value has source evidence features, which include at least field value, source evidence, field extraction method, and confidence level. The source evidence includes at least an evidence field.

[0050] In this embodiment, the retrieval enhancement service can be the core module for the system to perform the first-stage retrieval recall. The retrieval enhancement service receives the retrieval query determined based on the extracted configuration data and performs similarity calculations in the material data associated with the task unit. The similarity calculation can use a vector retrieval model to calculate semantic similarity, a full-text retrieval engine to calculate keyword matching, or a hybrid retrieval strategy to fuse the results of both. The retrieval enhancement service can sort document paragraphs in descending order according to the similarity score, select the top K paragraphs with the highest scores as the recall results, and generate corresponding evidence fields for each of the top K paragraphs.

[0051] In this embodiment, the feature extraction service can be the core module for the system to perform the second stage of structured extraction. The feature extraction service first receives several paragraphs and slot attributes output from the retrieval and recall stage. Then, based on the slot data type and extraction configuration data, it uses a corresponding extraction strategy to identify and extract field values ​​from each paragraph. For example, for fields with fixed formats, the feature extraction service can use regular expression matching; for named entity fields, it can use a pre-trained NER model; for complex semantic scenarios, it can use a large language model for prompt word extraction. The feature extraction service generates a candidate value object for each extraction result, i.e., a candidate value list corresponding to the slot. The candidate values ​​in the candidate value list have source evidence features, which at least include field values, source evidence, field extraction method, and confidence level. Source evidence at least includes evidence fields. This embodiment can also count the number of preset keywords matched in the paragraph, providing a basis for subsequent confidence level calculations.

[0052] In this embodiment, the evidence extraction of material data associated with task units can include a first-stage retrieval and recall and a second-stage structured extraction. The first-stage retrieval and recall corresponds to steps 300 to 320, and the second-stage structured extraction corresponds to step 330. In this embodiment, the first-stage retrieval and recall can quickly filter using lightweight indexing technology, generating evidence fields for each paragraph similar to the retrieval query to record information such as the paragraph's source document identifier, source document name, source document type, paragraph identifier, and paragraph content. In this embodiment, the second-stage structured extraction can extract the field values ​​corresponding to the slots using an element extraction service, and the extracted field values ​​can be associated with the evidence fields. This embodiment significantly reduces computational resource consumption through a two-stage evidence extraction process and can adapt to different types of documents, achieving unified processing of multi-source heterogeneous data. For example, the first stage can uniformly process various documents such as PDF, Word, Excel, and scanned documents; the second stage can select different extraction strategies according to the document type.

[0053] This embodiment employs a two-stage evidence extraction mechanism, including retrieval and structured extraction, to associate each candidate value with corresponding source evidence. Reviewers can view the original source of any field with a single click, eliminating the need for manual document backtracking. This achieves refined traceability of report generation results and process data at the field level.

[0054] 230. Based on source evidence features, conflict detection is performed on at least one filling candidate value to obtain conflict detection data, and the confidence of the filling candidate value is determined based on a pre-set confidence model. The conflict detection data and confidence are used to determine the target filling value of the slot.

[0055] In some embodiments, conflict detection is performed on at least one padding candidate value based on source evidence features to obtain conflict detection data, including: determining the conflict type corresponding to the padding candidate value according to the source evidence features corresponding to the padding candidate value, wherein the conflict type is a conflict-free type, a value conflict type, a logical conflict type, a format conflict type, or a data inconsistency type.

[0056] In this embodiment, value conflict can occur when multiple different candidate values ​​exist for the same slot. Logical conflict can occur when a candidate value differs from a preset logical constraint (e.g., "Total Assets ≠ Liabilities + Owner's Equity"). Format conflict can occur when candidate values ​​have inconsistent formats (e.g., dates "2024-01-01" vs. "January 1, 2024"). Data inconsistency can occur when the description of the same entity differs in different documents (e.g., a company name abbreviation vs. full name). No conflict can occur when the system determines that the candidate values ​​for a slot do not have value conflicts, logical conflicts, format conflicts, or data inconsistencies.

[0057] In some embodiments. See Figure 4 , Figure 4 This is a flowchart illustrating the process of determining the confidence level of candidate values ​​for filling in according to an embodiment of the present invention. Determining the confidence level of candidate values ​​for filling in according to a pre-set confidence model includes:

[0058] 400. Obtain the similarity score from the source evidence corresponding to the candidate value to be filled, and use the similarity score as the first feature;

[0059] 410. Obtain the source document type from the source evidence corresponding to the candidate value, determine the credibility weight based on the source document type, and obtain the second feature;

[0060] 420. Determine the slot corresponding to the candidate filling value and obtain the total number of candidate filling values ​​for the slot. Determine the third feature based on the reciprocal of the total number of candidate filling values.

[0061] 430. Obtain the number of preset keywords appearing in the paragraph corresponding to the candidate values ​​for filling, and the total number of preset keywords. The ratio between the number of preset keywords and the total number of preset keywords is the fourth feature.

[0062] 440. When the slot type of the slot corresponding to the candidate value to be filled is numerical, the fifth feature is determined based on the difference between the candidate value to be filled and the preset reference value.

[0063] 450. Obtain the logical verification pass flag corresponding to the candidate value to be filled. The logical verification pass flag is the sixth feature.

[0064] 460. The first to sixth features are weighted and summed to obtain the confidence level of the candidate values.

[0065] In this embodiment, the logical verification pass flag can be 0 or 1, where 0 indicates that there is no logical conflict in the candidate values ​​to be filled, and 1 indicates that there is a logical conflict in the candidate values ​​to be filled. This embodiment can determine the credibility weight by querying the credibility weight list based on the source document type; for example, the credibility weight of a source document of type "article" is greater than the credibility weight of a source document of type "internal report". The confidence level in this embodiment integrates multiple dimensions such as similarity score, credibility weight, and keyword coverage to comprehensively reflect the credibility of the candidate values ​​to be filled.

[0066] This embodiment can perform conflict detection on candidate filling values ​​based on source evidence features, and determine the confidence level of candidate filling values ​​based on a pre-set confidence model. Conflict detection data and confidence levels are used to determine the target filling value of the slot. This embodiment can automatically process multiple candidate values ​​from multiple material documents by detecting conflicts and determining confidence levels. It can intelligently integrate multi-source data, automatically identify and disambiguate conflicts, reduce the cost of manual review, and improve the accuracy of target filling value identification.

[0067] In some embodiments, please refer to Figure 5, Figure 5 This is a schematic flowchart illustrating the process of determining the target fill value of a slot according to an embodiment of the present invention. Determining the target fill value of a slot includes:

[0068] 500. When the confidence level of the candidate value to be filled is not less than the preset high threshold and the conflict type is a non-conflict type, the candidate value to be filled is determined as the target value to be filled in the slot, and the status of the slot is determined to be confirmed.

[0069] 510. When the confidence level of the candidate value to be filled is less than the preset high threshold and not less than the preset low threshold, the candidate value to be filled is determined as the suggested value to be filled, and the status of the slot is marked as pending confirmation.

[0070] 520. When the confidence level of the candidate value to be filled is less than the preset low threshold, or the conflict type is a logical conflict type, the status of the slot is determined to be pending manual confirmation.

[0071] This embodiment can trigger different decision actions based on the confidence level of the candidate values ​​to be filled and the type of conflict. The decision actions can be divided into three levels: automatic filling, automatic suggestion, and forced manual confirmation, so as to realize differentiated automated processing.

[0072] 240. Fill the target values ​​into the report template to generate the target report; and render the slots as traceable data nodes to complete the structured rendering and traceable management of the slots.

[0073] In some embodiments, filling a report template with target values ​​to generate a target report includes: determining slot placeholders in the report template and traversing the slot placeholders to fill the corresponding slot placeholders with target values ​​for different slots; generating a target report in response to an export command and attaching an evidence tracing appendix according to configuration options.

[0074] In some embodiments, filling the target fill value of different slots into the corresponding slot placeholder position includes: when the slot data type of any slot is a chart type, calling the front-end rendering service to generate a visual chart based on the target fill value of any slot; determining the rendering area based on the visual chart, taking a screenshot of the rendering area to obtain an image file, and inserting the image file into the position of the corresponding slot placeholder in any slot in the report template.

[0075] In some embodiments, a data node can be an atomic node. See also Figure 6 , Figure 6 This is a schematic diagram illustrating the process of rendering slots as traceable data nodes according to an embodiment of the present invention. Rendering slots as traceable data nodes to complete the structured rendering and traceable management of slots includes:

[0076] 600. Based on the target fill value of any slot, render the target fill value of any slot as an atomic node using a text editor;

[0077] 610. When the target report includes several reference locations, and the several reference locations point to the same slot, the several reference locations correspond to several atomic nodes, and the several atomic nodes share the data source corresponding to the same slot.

[0078] 620. When it is detected that a user has modified the target fill value at any reference location, the target fill value corresponding to all reference locations is updated synchronously; the synchronous update adopts a transaction mechanism; and the operation history data of the target fill value update is recorded.

[0079] In this embodiment, the system can render each slot as an indivisible atomic node. Each atomic node carries at least the slot identifier, target fill value, status identifier, and corresponding source evidence for any given slot. The status identifier can include confirmed status, pending confirmation status, manual confirmation required status, conflict status, etc.

[0080] In this embodiment, when the target report includes several reference locations, and these reference locations all point to the same slot, that is, when multiple places in the document reference the same slot, all reference nodes share the same slot data source. After the user modifies the slot value at any location, the system automatically updates all reference locations synchronously. The synchronous update adopts a transaction mechanism to ensure consistency.

[0081] In this embodiment, the system synchronous update adopts a transaction mechanism, including starting a transaction in response to the value modification operation of atomic nodes; performing updates to the shared data source, refreshing all associated atomic nodes, and writing modification history records within the transaction; if all the above operations are successful, the transaction is committed to make the modification permanent; if any operation fails, the transaction is rolled back to restore the shared data source, atomic node display, and history records to their state before the modification.

[0082] In this embodiment, the operation history data includes at least the target fill value before modification, the target fill value after modification, the source evidence characteristics before modification, the source evidence characteristics after modification, the operator identifier, the timestamp, and the version number; the operation history data is used to support the target report to be rolled back to any historical state by version.

[0083] The embodiments in this specification can perform structured modeling of report templates and define slots, which can upgrade traditional placeholders to intelligent field definitions, providing a solid foundation for subsequent evidence extraction, conflict resolution, and traceable editing, and enabling the reusability, maintainability, and intelligence of report templates;

[0084] Furthermore, the embodiments of this specification can also associate material data with the task unit corresponding to the target report generation task, that is, an independent session can be created for each target report generation task, and the data of different target report generation tasks are completely isolated and do not interfere with each other. This supports multiple users to process multiple tasks at the same time, and the embodiments of this specification significantly improve the parallel processing capability of report generation.

[0085] Furthermore, the embodiments of this specification can also extract evidence of material data associated with task units based on slot attributes. That is, the embodiments of this specification use a two-stage evidence extraction mechanism including retrieval and structured extraction to associate each filling candidate value with the corresponding source evidence. Reviewers can view the original source of any field with one click without manual backtracking of documents, thus realizing fine-grained traceability of report generation results and process data at the field level.

[0086] Furthermore, the embodiments of this specification can also perform conflict detection on candidate filling values ​​based on source evidence features, and determine the confidence level of candidate filling values ​​based on a pre-set confidence model. Conflict detection data and confidence levels are used to determine the target filling value for the slot. The embodiments of this specification can automatically process multiple candidate values ​​from multiple material documents through conflict detection and confidence level determination, intelligently integrate multi-source data, automatically identify and disambiguate conflicts, reduce manual review costs, and improve the accuracy of target filling value identification. Moreover, the embodiments of this specification can render slots as traceable data nodes, and implement a synchronization mechanism through data nodes to ensure document consistency and traceability of document editing.

[0087] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0088] Please see Figure 7 , Figure 7 This is a schematic diagram of the structure of the intelligent report generation system based on the chain of evidence provided in the embodiments of this specification.

[0089] like Figure 7 As shown, the intelligent report generation system based on the chain of evidence may include at least a slot module 700, a task determination module 710, an evidence extraction module 720, a filling determination module 730, and a report generation module 740, wherein:

[0090] The slot module 700 is used to obtain the report template and perform structured modeling on the report template. The structured modeling includes determining the slots and slot attributes; the slot is the data object corresponding to the field to be populated in the report template.

[0091] The task determination module 710 is used to acquire material data, determine the task unit corresponding to the target report generation task, and associate the material data with the task unit.

[0092] The evidence extraction module 720 is used to extract evidence from the material data associated with the task unit based on the slot attributes. The evidence extraction includes retrieval and recall as well as structured extraction, to obtain at least one candidate value for filling the slot and the source evidence features corresponding to the candidate value.

[0093] The filling determination module 730 is used to perform conflict detection on at least one filling candidate value based on source evidence features to obtain conflict detection data, and to determine the confidence of the filling candidate value based on a pre-set confidence model. The conflict detection data and the confidence are used to determine the target filling value of the slot.

[0094] The report generation module 740 is used to fill the target values ​​into the report template to generate the target report; and to render the slots as traceable data nodes to complete the structured rendering and traceable management of the slots.

[0095] In some embodiments, the slot module 700 includes a structured modeling module, which is used to: determine the metadata corresponding to the report template, the metadata including at least the template name, applicable scenario and version number; determine the fields to be filled in the report template, define slots for the fields to be filled, and the slots correspond to several slot attributes; determine the slot group, the slot group including several semantically corresponding slots, and the slot group is used for batch extraction and logical verification.

[0096] In some embodiments, the task determination module 710 includes a task determination submodule, which is used to: acquire material data corresponding to the target report generation task; establish a session corresponding to the target report generation task; the session is used to isolate and store the material data and record the file metadata corresponding to the material data; and is also used to manage the task data corresponding to the target report generation task so as to associate the task data with the session.

[0097] In some embodiments, slot attributes include at least extraction configuration data. The evidence extraction module 720 includes an extraction submodule, which is used to: determine the retrieval query corresponding to the slot based on the extraction configuration data; input the retrieval query to a retrieval enhancement service, which is used to recall several paragraphs corresponding to the retrieval query from the material data associated with the task unit; based on the retrieval query, perform a similarity retrieval in the material data through the retrieval enhancement service to obtain several paragraphs similar to the retrieval query; generate an evidence field for each paragraph; the evidence field includes at least the source document identifier, source document name, source document type, paragraph identifier, paragraph content, position information of the paragraph in the source document, and similarity score between the paragraph and the retrieval query; input several paragraphs and slot attributes to an element extraction service, which is used to perform field identification on each paragraph to extract the field value corresponding to the slot and obtain a candidate value list corresponding to the slot; the candidate value list includes at least one filling candidate value, which has source evidence features, which include at least the field value, source evidence, field extraction method, and confidence level; the source evidence includes at least the evidence field.

[0098] In some embodiments, the filling determination module 730 includes a conflict detection module, which is used to: determine the conflict type corresponding to the filling candidate value based on the source evidence features corresponding to the filling candidate value, wherein the conflict type is a conflict-free type, a value conflict type, a logical conflict type, a format conflict type, or a data inconsistency type.

[0099] In some embodiments, the filling determination module 730 includes a confidence determination module, which is used to: obtain a similarity score in the source evidence corresponding to the filling candidate value, the similarity score being a first feature; obtain the source document type in the source evidence corresponding to the filling candidate value, determine a confidence weight based on the source document type, and obtain a second feature; determine the slot corresponding to the filling candidate value and obtain the total number of filling candidate values ​​for the slot, and determine a third feature based on the reciprocal of the total number of filling candidate values; obtain the number of preset keywords appearing in the paragraph corresponding to the filling candidate value and the total number of preset keywords, the ratio between the number of preset keywords and the total number of preset keywords being a fourth feature; when the slot type of the slot corresponding to the filling candidate value is numerical, determine a fifth feature based on the difference between the filling candidate value and a preset reference value; obtain a logical verification pass flag corresponding to the filling candidate value, the logical verification pass flag being a sixth feature; and perform a weighted summation of the first to sixth features to obtain the confidence of the filling candidate value.

[0100] In some embodiments, the filling determination module 730 includes a candidate value determination module, which is configured to: determine the filling candidate value as the target filling value for the slot when the confidence level of the filling candidate value is not less than a preset high threshold and the conflict type is a non-conflict type, so as to fill the slot and determine the status of the slot as a confirmed status; determine the filling candidate value as a filling suggestion value when the confidence level of the filling candidate value is less than a preset high threshold and not less than a preset low threshold, and determine the status of the slot as a pending confirmation status; and determine the status of the slot as a pending manual confirmation status when the confidence level of the filling candidate value is less than a preset low threshold or the conflict type is a logical conflict type.

[0101] In some embodiments, the report generation module 740 includes a report generation submodule, which is configured to: determine slot placeholders in the report template and traverse the slot placeholders to fill the corresponding slot placeholders with target fill values ​​for different slots; generate a target report in response to an export instruction and attach an evidence tracing appendix according to configuration options; fill the corresponding slot placeholders with target fill values ​​for different slots, including: when the slot data type of any slot is a chart type, calling a front-end rendering service to generate a visual chart based on the target fill value of any slot; determining a rendering area based on the visual chart, taking a screenshot of the rendering area to obtain an image file, and inserting the image file into the position of the corresponding slot placeholder for any slot in the report template.

[0102] In some embodiments, the data nodes are atomic nodes. The report generation module 740 includes a node management module, which is used to: render the target fill value of any slot into an atomic node using a text editor based on the target fill value of any slot. The atomic node carries at least the slot identifier, target fill value, status identifier, and corresponding source evidence of any slot. When the target report includes several reference positions, and the several reference positions point to the same slot, the several reference positions correspond to several atomic nodes, and the several atomic nodes share the data source corresponding to the same slot. When it is detected that a user modifies the target fill value at any reference position, the target fill value corresponding to all reference positions is updated synchronously. The synchronous update adopts a transaction mechanism. The operation history data of the target fill value update is recorded. The operation history data includes at least the target fill value before modification, the target fill value after modification, the source evidence characteristics before modification, the source evidence characteristics after modification, the operator identifier, the timestamp, and the version number. The operation history data is used to support the target report to roll back to any historical state by version.

[0103] Based on the content of the intelligent report generation system based on the chain of evidence in multiple embodiments of this specification, it can be seen that the embodiments of this specification can perform structured modeling of report templates and define slots, and can upgrade traditional placeholders to intelligent field definitions, providing a solid foundation for subsequent evidence extraction, conflict resolution, and traceable editing, and can realize the reusability, maintainability and intelligence of report templates;

[0104] Furthermore, the embodiments of this specification can also associate material data with the task unit corresponding to the target report generation task, that is, an independent session can be created for each target report generation task, and the data of different target report generation tasks are completely isolated and do not interfere with each other. This supports multiple users to process multiple tasks at the same time, and the embodiments of this specification significantly improve the parallel processing capability of report generation.

[0105] Furthermore, the embodiments of this specification can also extract evidence of material data associated with task units based on slot attributes. That is, the embodiments of this specification use a two-stage evidence extraction mechanism including retrieval and structured extraction to associate each filling candidate value with the corresponding source evidence. Reviewers can view the original source of any field with one click without manual backtracking of documents, thus realizing fine-grained traceability of report generation results and process data at the field level.

[0106] Furthermore, the embodiments of this specification can also perform conflict detection on candidate filling values ​​based on source evidence features, and determine the confidence level of candidate filling values ​​based on a pre-set confidence model. Conflict detection data and confidence levels are used to determine the target filling value for the slot. The embodiments of this specification can automatically process multiple candidate values ​​from multiple material documents through conflict detection and confidence level determination, intelligently integrate multi-source data, automatically identify and disambiguate conflicts, reduce manual review costs, and improve the accuracy of target filling value identification. Moreover, the embodiments of this specification can render slots as traceable data nodes, and implement a synchronization mechanism through data nodes to ensure document consistency and traceability of document editing.

[0107] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the embodiment of the intelligent report generation system based on the chain of evidence is relatively simple in description because it is fundamentally similar to the embodiment of the intelligent report generation method based on the chain of evidence; relevant parts can be referred to the description of the method embodiment.

[0108] Please see Figure 8 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this specification.

[0109] like Figure 8As shown, the electronic device 800 may include at least one processor 810, at least one network interface 840, a user interface 830, a memory 850, and at least one communication bus 820.

[0110] The communication bus 820 can be used to realize the connection and communication of the above components.

[0111] The user interface 830 may include buttons, and the optional user interface may also include a standard wired interface or a wireless interface.

[0112] The network interface 840 may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0113] The processor 810 may include one or more processing cores. The processor 810 connects to various parts within the electronic device 800 using various interfaces and lines. It executes various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 850, and by calling data stored in the memory 850. Optionally, the processor 810 may be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor 810 may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 810 and may be implemented as a separate chip.

[0114] The memory 850 may include RAM or ROM. Optionally, the memory 850 may include a non-transitory computer-readable medium. The memory 850 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 850 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 850 may also be at least one storage device located remotely from the aforementioned processor 810. As a computer storage medium, the memory 850 may include an operating system, a network communication module, a user interface module, and an evidence chain-based intelligent report generation application. The processor 810 may be used to call the evidence chain-based intelligent report generation application stored in the memory 850 and execute the steps of the evidence chain-based intelligent report generation method mentioned in the foregoing embodiments.

[0115] This specification also provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figures 2-6 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0116] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this specification is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0117] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and its implementation can be combined arbitrarily.

[0118] The above embodiments are merely preferred embodiments described in this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for generating intelligent reports based on a chain of evidence, characterized in that, include: Obtain a report template and perform structured modeling on the report template. The structured modeling includes determining the slots and slot attributes. The slot is the data object corresponding to the field to be populated in the report template; Acquire material data, determine the task unit corresponding to the target report generation task, and associate the material data with the task unit; Based on the slot attributes, evidence extraction is performed on the material data associated with the task unit. The evidence extraction includes retrieval and recall as well as structured extraction, to obtain at least one candidate value for filling the slot and the source evidence features corresponding to the candidate value. Based on the source evidence features, conflict detection is performed on the at least one filling candidate value to obtain conflict detection data, and the confidence level of the filling candidate value is determined based on a pre-set confidence model. The conflict detection data and the confidence level are used to determine the target filling value of the slot. The target value is filled into the report template to generate the target report; The slot is then rendered as a traceable data node to complete the structured rendering and traceable management of the slot.

2. The method according to claim 1, characterized in that, The structured modeling of the report template includes: Determine the metadata corresponding to the report template, which includes at least the template name, applicable scenarios, and version number; Determine the fields to be populated in the report template, define slots for the fields to be populated, and each slot corresponds to several slot attributes; Determine a slot group, which includes several semantically corresponding slots. The slot group is used for batch extraction and logical verification.

3. The method according to claim 1, characterized in that, The steps of acquiring material data, determining the task unit corresponding to the target report generation task, and associating the material data with the task unit include: Obtain the material data corresponding to the target report generation task; Establish a session corresponding to the target report generation task; The session is used to isolate and store the material data, record the file metadata corresponding to the material data, and also to manage the task data corresponding to the target report generation task, so as to associate the task data with the session.

4. The method according to claim 1, characterized in that, The slot attributes include at least extracted configuration data. Based on the slot attributes, evidence-based extraction is performed on the material data associated with the task unit. This evidence-based extraction includes retrieval recall and structured extraction to obtain at least one candidate filling value for the slot and source evidence features corresponding to the candidate filling value, including: The retrieval query corresponding to the slot is determined based on the extracted configuration data; The search query is input into the search enhancement service, which is used to retrieve several paragraphs corresponding to the search query from the material data associated with the task unit. Based on the search query, a similarity search is performed on the material data using a search enhancement service to obtain several paragraphs similar to the search query; an evidence field is generated for each paragraph. The evidence field includes at least the source document identifier, source document name, source document type, paragraph identifier, paragraph content, the position information of the paragraph in the source document, and the similarity score between the paragraph and the search query; The paragraphs and slot attributes are input into the feature extraction service. The feature extraction service is used to perform field recognition on each paragraph to extract the field values ​​corresponding to the slot and obtain a candidate value list corresponding to the slot. The candidate value list includes at least one filler candidate value, which has source evidence features. The source evidence features include at least field value, source evidence, field extraction method, and confidence level; the source evidence includes at least the evidence field.

5. The method according to claim 4, characterized in that, The step of performing conflict detection on the at least one padding candidate value based on the source evidence features to obtain conflict detection data includes: The conflict type corresponding to the filling candidate value is determined based on the source evidence features corresponding to the filling candidate value. The conflict type is a conflict-free type, a value conflict type, a logical conflict type, a format conflict type, or a data inconsistency type.

6. The method according to claim 5, characterized in that, The process of determining the confidence level of the imputed candidate values ​​based on a pre-set confidence model includes: Obtain the similarity score from the source evidence corresponding to the candidate filling value, where the similarity score is the first feature; Obtain the source document type from the source evidence corresponding to the filling candidate value, determine the credibility weight based on the source document type, and obtain the second feature; Determine the slot corresponding to the filling candidate value, obtain the total number of filling candidate values ​​for the slot, and determine the third feature based on the reciprocal of the total number of filling candidate values; The number of preset keywords appearing in the paragraph corresponding to the candidate filling value and the total number of preset keywords are obtained, and the ratio of the number of preset keywords to the total number of preset keywords is the fourth feature; When the slot type of the slot corresponding to the candidate filling value is numerical, the fifth feature is determined based on the difference between the candidate filling value and the preset reference value. Obtain the logical verification pass flag corresponding to the filling candidate value, where the logical verification pass flag is the sixth feature; The confidence level of the filling candidate value is obtained by weighted summation of the first feature to the sixth feature.

7. The method according to claim 1, characterized in that, Determining the target fill value of the slot includes: When the confidence level of the filling candidate value is not less than a preset high threshold and the conflict type is a non-conflict type, the filling candidate value is determined as the target filling value of the slot to fill the slot, and the status of the slot is determined to be a confirmed status. When the confidence level of the filling candidate value is less than a preset high threshold and not less than a preset low threshold, the filling candidate value is determined as a filling suggestion value, and the status of the slot is determined to be pending confirmation. When the confidence level of the candidate filling value is less than a preset low threshold, or when the conflict type is a logical conflict type, the status of the slot is determined to be pending manual confirmation.

8. The method according to claim 1, characterized in that, The step of filling the target value into the report template to generate the target report includes: Determine the slot placeholders in the report template and iterate through the slot placeholders to fill the target fill values ​​of different slots into the positions of the corresponding slot placeholders; The system generates a target report in response to export commands and attaches an evidence tracing appendix based on configuration options. The step of filling the target fill values ​​of different slots into the positions of the corresponding slot placeholders includes: When the data type of any slot is a chart type, the front-end rendering service is called to generate a visual chart based on the target fill value of the slot. The rendering area is determined based on the visualization chart, and an image file is obtained by taking a screenshot of the rendering area. The image file is then inserted into the position of the placeholder corresponding to any slot in the report template.

9. The method according to claim 1, characterized in that, The data node is an atomic node. Rendering the slot as a traceable data node to complete the structured rendering and traceable management of the slot includes: Based on the target fill value of any slot, the target fill value of the arbitrary slot is rendered into an atomic node using a text editor. The atomic node carries at least the slot identifier, target fill value, status identifier, and corresponding source evidence of the arbitrary slot. When the target report includes several reference locations, and the several reference locations point to the same slot, the several reference locations correspond to several atomic nodes, and the several atomic nodes share the data source corresponding to the same slot. When a user is detected modifying the target fill value at any reference location, the target fill value corresponding to all reference locations is synchronously updated; the synchronous update adopts a transaction mechanism; and the operation history data of the target fill value update is recorded. The operation history data includes at least the target fill value before modification, the target fill value after modification, the source evidence characteristics before modification, the source evidence characteristics after modification, the operator identifier, the timestamp, and the version number; the operation history data is used to support the target report to be rolled back to any historical state by version.

10. A smart report generation system based on a chain of evidence, characterized in that, include: The slot module is used to obtain a report template and perform structured modeling on the report template. The structured modeling includes determining the slot and slot attributes. The slot is the data object corresponding to the field to be populated in the report template; The task determination module is used to acquire material data, determine the task unit corresponding to the target report generation task, and associate the material data with the task unit. The evidence extraction module is used to extract evidence from the material data associated with the task unit based on the slot attributes. The evidence extraction includes retrieval and recall as well as structured extraction, to obtain at least one filling candidate value for the slot and the source evidence features corresponding to the filling candidate value. A filling determination module is used to perform conflict detection on the at least one filling candidate value based on the source evidence features to obtain conflict detection data, and to determine the confidence level of the filling candidate value based on a preset confidence model. The conflict detection data and the confidence level are used to determine the target filling value of the slot. The report generation module is used to fill the target value into the report template to generate the target report; The slot is then rendered as a traceable data node to complete the structured rendering and traceable management of the slot.

Images