Iterative event context construction method based on semantic integrity driving and computer program product

Abstract

The invention belongs to the technical field of natural language processing and information summarization, and particularly relates to a semantic integrity driven iterative event venation construction method and a computer program product, the method comprises the following steps: 1, preliminary event venation construction: screening a document related to a target event from a multi-source web text, and constructing an initial event development sequence; step 2, semantic integrity detection and reflection analysis: performing semantic integrity detection on each event in the initial event development sequence, identifying missing basic semantic elements, and generating a list of semantic elements to be complemented; step 3, iterative completion and final event venation output: based on the to-be-completed semantic element list, generating element-oriented query to perform directional retrieval, and integrating new information obtained by retrieval into the event venation; and 4, repeatedly executing the steps 2 to 3 to form an iteration closed loop of generation-reflection-complementation until an iteration termination condition is reached, and outputting a final event context.

Description

Technical Field

[0001] The invention belongs to the field of natural language processing and information summarization technology, specifically involving an iterative event context construction method and computer program product based on semantic integrity, which is particularly suitable for application scenarios in open domains that complete key elements and structure complex events with fragmented and incomplete information. Background Technology

[0002] With the explosive growth of internet information, users' need to understand the key facts of complex social events, public issues, and emergencies is increasing. Against this backdrop, event tracing methods, as a technique that can automatically extract key nodes from scattered and heterogeneous text data (such as news reports, social media updates, and government announcements) and organize them into a structured narrative, have become an important research direction in the field of natural language processing. The core objective of these methods is to generate a complete, clearly expressed, and interpretable event development path, helping users quickly grasp the core elements and development outline of an event.

[0003] Existing methods for tracing the context of events primarily rely on information extraction and ranking models, or employ end-to-end sequence generation frameworks. However, these methods generally suffer from incomplete event descriptions and missing key semantic elements when dealing with complex events in open domains and involving multiple sources. For example, a report might only mention "an organization issued a policy" without specifying the policy's target audience, scope of implementation, or implementing body; another might describe "an emergency meeting" but fail to identify the chairperson, participants, or decision outcome. Such omissions are extremely common in real-world texts, resulting in event entries generated by traditional methods that, while formally valid, lack analytical value.

[0004] In recent years, large language models (LLMs) have been widely used in event extraction tasks due to their powerful semantic understanding capabilities. While LLMs can generate fluent summaries, they are prone to "illusions" in the absence of external validation, fabricating unmentioned details. To address this, retrieval-augmented generation (RAG) has become a mainstream solution, enhancing factual accuracy by incorporating external documents. However, existing RAG methods often remain at a coarse-grained level of content assembly, lacking a proactive evaluation mechanism for the completeness of event descriptions. They cannot determine whether an event is missing necessary components (such as the object of the action, the scope of influence, etc.), leading to a blind and inefficient completion process and difficulty in forming structured, verifiable factual units.

[0005] Furthermore, current methods generally lack the ability to systematically verify the semantic structure of events, making it difficult to identify common defects such as "actions without objects" or "decision-making without implementing entities," and also unable to proactively initiate targeted searches to fill gaps. This results in the generated event timelines often remaining at a superficial level, failing to meet the requirements of fact density and semantic completeness for in-depth analysis scenarios.

[0006] To address these challenges, there is an urgent need for a novel event context construction mechanism that can integrate external evidence with structured semantic analysis. Summary of the Invention

[0007] To address the issues of missing key semantic elements and incomplete descriptions in existing open-domain event contexts, this invention aims to overcome these shortcomings by proposing an iterative event context construction method and computer program product driven by semantic integrity.

[0008] In view of this, the present invention proposes an iterative event context construction method based on semantic integrity, comprising: Step 1: Preliminary event timeline construction: From multi-source network texts, filter documents related to the target event to construct an initial event development sequence; Step 2: Semantic integrity detection and reflective analysis: Perform semantic integrity detection on each event in the initial event development sequence, identify missing basic semantic elements, and generate a list of semantic elements to be completed; Step 3: Iterative completion and final event context output: Based on the list of semantic elements to be completed, generate element-oriented queries for targeted retrieval, and integrate the new information obtained from the retrieval into the event context; Step 4: Repeat steps 2 to 3 to form an iterative closed loop of "generation-reflection-completion" until the iteration termination condition is met, and output the final event sequence.

[0009] As an improvement to the above method, step 1 includes: Step 101: Topic-driven initial retrieval: Obtain a set of candidate documents related to the target event topic; Step 102: Webpage content parsing and cleaning: Extract structured text content and remove noise information; Step 103: Progress-based document screening: Identify and retain information containing key developments of the event from candidate documents to form a high-information-density input corpus; Step 104: Initial event context generation: Extract key events from the input corpus and organize them into an ordered sequence to form the initial event development sequence.

[0010] As an improvement to the above method, step 2 includes: Step 201: Semantic Role Coverage Detection: Identify missing basic semantic roles in the event using a large language model. The basic semantic roles include at least one of the following: subject, object, time, and location. Step 202: Domain Framework Matching Verification: Match the event with the predefined semantic framework, check whether the core arguments are complete, and determine whether it conforms to the expression specifications of typical event types; Step 203: Integrate the results of steps 201 and 202 through a semantic confidence scoring mechanism to generate a list of semantic elements to be completed.

[0011] As an improvement to the above method, the predefined semantic framework is built on the FrameNet structured semantic framework library and includes standardized description frameworks for different event types and the necessary set of core arguments.

[0012] As an improvement to the above method, step 3 includes: Step 301: Generate semantic element-oriented queries: Transform each missing item in the list of semantic elements to be completed into a specific retrieval question; Step 302: Targeted retrieval and information extraction: Obtain new documents related to the query through a search engine, and parse and extract information fragments that can be used to complete the event; Step 303: Information Fusion and Iterative Convergence: Integrate the newly acquired information into the existing event context and update the missing fields; proceed to Step 302 to perform a new round of detection and completion, forming iterative optimization and updating the event context.

[0013] As an improvement to the above method, step 301 further includes: A problem rewriting mechanism is introduced to decouple and simplify the structure of complex semantic retrieval by breaking it down into multiple retrieval sub-problems.

[0014] As an improvement to the above method, the iteration termination condition in step 4 includes any of the following: All missing items were successfully filled in; Reach the preset maximum number of iterations; The semantic gain from the newly added information is less than the preset threshold.

[0015] As an improvement to the above method, the method further includes: sorting all event nodes by global timestamps, merging semantically duplicate entries, and eliminating conflicts before outputting the final event timeline.

[0016] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0017] Compared with the prior art, the advantages of the present invention are: 1. Focusing on a semantic integrity-driven precision completion mechanism, this invention effectively enhances the factual density of event descriptions. Addressing the issues of fragmented event information and the frequent omission of key semantic elements (such as subject, object, scope, and basis) in open-domain texts, this invention introduces a joint detection mechanism based on a large language model and a structured semantic framework (such as FrameNet) after generating a preliminary framework. By identifying explicit information through semantic role labeling and combining it with domain-adapted event framework templates, the core argument filling status of each event is systematically verified, proactively identifying typical information gaps such as "missing policy object" and "unclear implementing entity." All missing items are transformed into element-oriented queries, which, after question rewriting and optimization, trigger targeted retrieval, achieving precise completion of key details and significantly enhancing the information integrity and interpretability of the event framework.

[0018] 2. Constructing a lightweight closed-loop optimization process to achieve self-iterative improvement of the event context. Unlike traditional unidirectional pipeline methods that are easily constrained by the quality of initial input, this invention designs an iterative closed-loop architecture of "generation—detection—completion—re-detection". After each information completion, the system re-executes semantic integrity analysis, continuously exposing deeper descriptive defects and gradually approaching a state of complete information. This process does not rely on large-scale labeled data or dedicated temporal reasoning models, requiring only a general-purpose large language model and a standard search engine interface. It possesses good cross-domain adaptability and deployment flexibility, and is particularly suitable for complex event analysis scenarios with sparse information and unclear development contexts, significantly reducing system complexity and computational overhead while ensuring output quality.

[0019] 3. The output event context has a standardized semantic structure, which facilitates automated parsing and reasoning for downstream tasks (such as public opinion analysis and decision support), significantly improving the interpretability and application compatibility of the system. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the semantic integrity-driven iterative event context construction method of the present invention; Figure 2 This is a detailed flowchart of the semantic integrity-driven iterative event context construction method of the present invention. Detailed Implementation

[0021] This invention proposes an iterative event context construction method driven by semantic integrity. By jointly detecting missing event elements through semantic role labeling and domain framework matching, it generates accurate queries and drives multiple rounds of retrieval and completion, realizing the dynamic evolution from "fragmented records" to "complete event units". This significantly improves the factual density and interpretability of event descriptions, and is especially suitable for open-domain complex event analysis tasks with sparse information and concise descriptions.

[0022] This invention aims to automatically construct complete, well-defined, and factually credible narratives of complex events from multi-source heterogeneous texts. By building a closed-loop framework of "generation-reflection-completion," and utilizing a large language model combined with semantic role labeling and a structured knowledge base, it identifies missing core information such as the subject, object, and scope of the event, and generates targeted queries for multi-round retrieval and completion. This effectively improves the completeness and factual density of event descriptions, enhances the readability and analytical value of the narrative, and is suitable for tasks involving the organization of complex events with fragmented information.

[0023] This method mainly includes the following three core steps: Step 1) Initial construction of the event timeline; This stage aims to filter documents from multi-source online text that are highly relevant to the target event and have high information density, and to construct a clear and focused initial event development sequence to lay the foundation for subsequent optimization. Specifically, it includes the following sub-steps: Step 101) Topic-driven initial retrieval: Obtain a set of candidate documents related to the target event topic to provide an initial data source for subsequent information extraction and context construction.

[0024] Step 102) Web page content parsing and cleaning: Extract structured text content from the original web page, remove noise information, and retain key metadata to provide high-quality, low-interference text input for analysis.

[0025] Step 103) Progress-based document screening: Identify and retain information containing key developments of the event from candidate documents, filter redundant or duplicate content, and form a high-information-density input corpus to support the accurate construction of subsequent event context.

[0026] Step 104) Initial event context generation: Extract key events from the filtered text and organize them into an ordered sequence to form a preliminary event development context, which serves as the basis for subsequent integrity checks and iterative optimization.

[0027] Step 2) Semantic integrity detection and reflective analysis: After obtaining the initial event outline, the system enters the core reflection phase, focusing on the semantic completeness of the event descriptions themselves. It proactively identifies key missing information elements in each event, achieving a leap from "passive generation" to "proactive review." This specifically includes the following sub-steps: Step 201) Semantic role coverage detection: Use a large language model to identify missing basic semantic roles such as subject, object, time, and location in the event, and discover event nodes with incomplete descriptions.

[0028] Step 202) Domain framework matching verification: Match the event with the predefined semantic framework, check whether the core arguments are complete, and determine whether it conforms to the expression specifications of typical event types.

[0029] Step 3) Iterative completion and final event context output: To fill the semantic information gaps discovered in the preceding stages, the system initiates a closed-loop dynamic optimization process. Through multiple rounds of iterative mechanisms—"problem discovery—query generation—information acquisition—fusion and update"—it continuously improves the event description until convergence is achieved. Specifically, this includes the following sub-steps: Step 301) Generate semantic element-guided queries: Transform the identified semantic missing information into specific questions to guide subsequent retrieval and improve the relevance and effectiveness of information completion.

[0030] Step 302) Targeted retrieval and information extraction: Obtain new documents related to the query through a search engine, and parse and extract information fragments that can be used to complete the event.

[0031] Step 303) Information Fusion and Iterative Convergence: Integrate the newly acquired information into the existing event context and update missing fields. Step 4) Repeat the detection and completion process from Step 2) to Step 3) to form an iterative closed loop of "generation-reflection-completion" until the iteration termination condition is met and the final event context is output.

[0032] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0033] Example 1 like Figure 1As shown, embodiments of the present invention provide an iterative event context construction method driven by semantic integrity. First, relevant articles are retrieved based on the event topic. After data cleaning and document filtering, a high-quality text corpus is constructed. Then, a preliminary event context is generated based on the corpus. Next, a core reflection stage is entered. Through a semantic integrity detection mechanism, the semantic integrity of the event description is focused on. A large language model is used to label the event with semantic roles, identifying missing key elements such as subject, action, object, time, and location. Simultaneously, combined with structured semantic framework libraries such as FrameNet, a domain-adaptive event description template is constructed. Event types are dynamically matched, and the completeness of core arguments is verified. Events that do not conform to the specifications are marked as "incomplete elements." For detected semantic missing items, the system generates element-oriented query questions and introduces a question rewriting mechanism to decouple and simplify complex semantic structures, improving retrieval accuracy. The questions are then submitted to a search engine to obtain supplementary documents and extract new information, completing the dynamic completion of event details. After completion, integrity detection is performed again, forming an iterative closed loop of "detection—completion—re-detection." Finally, the system sorts all event nodes by time, removes duplicates, and resolves conflicts, outputting a clear, complete, and logically coherent final event timeline.

[0034] The detailed process of the event tracing method used in this invention is as follows: Figure 2 As shown, its overall architecture consists of three core stages: initial event timeline construction, single-channel missing data detection and reflective analysis, and iterative completion and final event timeline output. The entire process starts with the user-inputted "event topic," obtains initial text resources through a search engine, and leverages Large Language Model (LLM), semantic framework parsing tools, and a structured knowledge base to automatically extract, fuse, verify, and refine the event timeline from massive amounts of unstructured text. Compared to the original solution, this version removes the "time point missing detection" branch, focusing on in-depth optimization of semantic integrity, thereby improving system efficiency and the accuracy of information completion.

[0035] Step 1: Initial Event Context Construction From multi-source network text, documents related to the target event are filtered out to construct an initial event development sequence.

[0036] Specifically, the system first uses the user-specified "event topic" as the query keyword and calls a common search engine (such as Google, Bing, or Baidu) to obtain a list of relevant web articles. These search results constitute the information pool for subsequent processing. To improve the quality and usability of the input text, the system further utilizes web content parsing tools (such as Jina) to perform structured parsing of the original web pages, automatically extracting the main text content, removing irrelevant noise such as advertisements, navigation bars, comment sections, and recommended links, and retaining key metadata such as publication time and publishing organization. The cleaned, high-quality text set is used for subsequent document screening and event extraction, effectively improving the information signal-to-noise ratio and processing efficiency. The system then proceeds to the initial timeline generation stage: In one embodiment, the top three ranked articles (i.e., the "top three articles") are first extracted, and key event nodes are extracted from them using a large language model to form an initial event chain. This step aims to quickly establish a basic outline of the event development, typically including the earliest key actions, subjects, and time points.

[0037] Next, the system performs topic relevance assessments on the remaining articles in turn, guiding the LLM to evaluate whether each article introduces new and valuable event information through structured prompt templates. For example, if an article mentions a new institution, policy, or scope of influence that has not been mentioned before, it is judged as "containing useful information."

[0038] For articles deemed "relevant to the topic and of supplementary value", the system transforms their content into structured event fragments, including elements such as timestamps, subjects, actions, objects, and locations. Then, it merges these fragments into the current event context using an event fusion algorithm, completing dynamic splicing and forming a preliminary event context framework.

[0039] Step 2: Semantic integrity detection and reflective analysis The system performs semantic integrity checks on each event in the initial event sequence, identifying missing basic semantic elements and generating a list of semantic elements to be completed. Specifically, after obtaining the initial event context, the system enters the core optimization phase—semantic integrity-based missing element detection and reflective analysis. Since original reports often suffer from incomplete event descriptions due to length limitations or incomplete information, direct output may lead to misunderstandings. Therefore, the system focuses on "horizontal information density," that is, the completeness of key semantic elements within a single event, constructing a hierarchical semantic parsing architecture. This architecture combines general semantic understanding capabilities with a structured knowledge guidance mechanism to achieve fine-grained, high-coverage integrity assessment.

[0040] General semantic role completion: Using a large language model, semantic role labeling (SRL) is performed on each extracted event description to identify core components such as explicitly expressed agents, actions, subjects / themes, time, location, and purpose. By designing role-filling prompts, the model is guided to actively infer whether there are common but unmentioned key components.

[0041] For example, for events such as "an organization releases a policy", if the original text does not clearly specify the policy target (e.g., "targeting high energy-consuming enterprises") or the scope of implementation (e.g., "implemented nationwide"), the system will mark the event as "incomplete elements" and generate role items to be completed, such as "missing: policy target" or "missing: applicable area".

[0042] In addition, the model can make preliminary inferences based on contextual common sense, such as inferring from the term "carbon tax" that the policy subject is likely the "Ministry of Finance" or the "Ministry of Ecology and Environment", providing directional guidance for subsequent supplementation.

[0043] Domain-enhanced framework matching: To further enhance the standardization and professionalism of semantic analysis, the system introduces a domain-adaptive framework library based on structured semantic knowledge bases such as FrameNet. This library predefines standardized semantic frameworks and essential core argument sets for different event types. For example: Event type: "Policy Release"; Required arguments: Subject, Policy Name, Target Audience, Effective Date, Legal Basis, Scope of Impact. The system dynamically matches extracted event verbs (such as "release," "sign," "notify") with semantic frameworks in the library and checks the fill status of each argument item by item according to the framework requirements. For events that do not conform to typical domain description patterns (such as "reaching a cooperation agreement" without specifying the cooperating parties or agreement terms), the system triggers a "semantic integrity alert," indicating the omission of key information.

[0044] Ultimately, the two types of detection results are integrated through a semantic confidence scoring mechanism: providing broad coverage and domain modules providing in-depth verification. The system generates a highly reliable "list of semantic elements to be completed," which includes the missing role type, associated event identifier, and completion priority (based on information importance and contextual dependency) for each item, providing precise guidance for subsequent detailed completion.

[0045] Step 3: Iterative completion and final event context output After completing the semantic integrity check, a closed-loop semantic completion mechanism is initiated based on the generated "list of semantic elements to be completed", realizing an automated cycle of "discovering problems → generating problems → searching information → completing the context → checking again".

[0046] For each missing element, the system automatically generates element-oriented query questions. For example: "Which industries are eligible for this carbon tax policy?"; "Which two countries signed this agreement?"; "Who chaired the meeting?" It is worth noting that when the generated query is too complex or contains multiple semantic structures, it may exceed the understanding capacity of large language models or the matching accuracy of search engines, leading to a decline in retrieval results. To address this, this invention introduces a question rewriting mechanism to semantically decouple and structurally simplify complex queries. This mechanism can automatically decompose a complex question (such as "After a country implemented a carbon tax, did any companies raise objections or apply for exemptions?") into multiple semantically clear and goal-oriented sub-questions: "Which companies raised objections after the implementation of the carbon tax?"; "Did any companies apply for carbon tax exemptions?"; "What are the names of the relevant companies and their reasons?".

[0047] By reducing the cognitive load and semantic complexity of individual queries, the accuracy of model understanding and the response efficiency of the retrieval system are significantly improved. These simplified questions will trigger deeper and more targeted web searches, thereby ensuring the continuity and effectiveness of the completion process.

[0048] These questions are submitted to a search engine to retrieve new relevant documents. The system parses, denoises, and extracts events from the new documents, extracting information fragments that can be used to complete the information and integrating them into the existing context. For example, if a new document states that "the steel and cement industries can apply for phased tax reductions and exemptions," the system will add this information to the "Scope of Policy Impact" field.

[0049] After completion, the system re-executes the second stage of semantic integrity detection to check for any remaining unaddressed semantic gaps, forming an iterative closed loop of "generation-reflection-completion". This process can continue for multiple rounds until any of the following termination conditions are met: all missing items are effectively completed; the preset maximum number of iterations is reached (3 rounds in one embodiment); the newly added information no longer brings significant semantic gains (in one embodiment, the overlap of completed information in two consecutive rounds is >90%).

[0050] Finally, all event nodes are sorted by global timestamp, semantically duplicated entries are merged, potential contradictions are resolved (such as different sources of different expressions of "policy release time", taking the release time of authoritative media as the standard), and a complete event development timeline with clear structure, rigorous logic and reliable facts is output.

[0051] Example 2 Embodiment 2 of the present invention provides a computer program product, including a computer program. When the computer program is executed by a processor, it can implement the various steps in the above method embodiments.

[0052] The main innovative points of this invention include: The innovation of this invention lies in proposing an iterative event context construction method driven by semantic integrity. Unlike existing technologies that are limited to static information extraction or single-round generation, this invention is the first to construct an automated closed-loop framework of "generation-reflection-completion," which actively and structurally diagnoses the semantic integrity of the initial event context through a large language model. This method abandons complex temporal logic modeling and focuses on the information density and semantic completeness of the event description itself. It uses semantic role labeling technology to identify explicitly expressed elements such as subjects, actions, objects, time, and location in the event, and combines it with structured semantic knowledge bases such as FrameNet to construct standardized description frameworks for typical event types (such as "policy release," "agreement signing," and "crisis notification"), dynamically matching event predicates and systematically checking the filling status of core arguments. For events that do not meet the framework requirements (such as "policy release" but without specifying the scope of implementation or target group), the system automatically marks them as "semantically incomplete" and generates high-precision element-oriented query questions (such as "Which industries are applicable to this policy?" or "Who chaired the meeting?"). Furthermore, this invention introduces a problem rewriting and semantic decoupling mechanism, breaking down complex queries into semantically clear and highly searchable sub-problems, thereby improving the search engine's response efficiency and information retrieval quality. After the newly obtained information is integrated into the existing context, the system restarts the integrity check, forming a multi-round iterative optimization loop to continuously expose and repair deeper-level information omissions. This process does not rely on domain-specific time-series models or large-scale labeled data, and has good generalization ability and engineering deployability. It is worth noting that in the above system embodiments, the various modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional module are only for mutual distinction and are not used to limit the scope of protection of this invention.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A semantically complete, iterative event context construction method, comprising: Step 1: Preliminary event timeline construction: From multi-source network texts, filter documents related to the target event to construct an initial event development sequence; Step 2: Semantic integrity detection and reflective analysis: Perform semantic integrity detection on each event in the initial event development sequence, identify missing basic semantic elements, and generate a list of semantic elements to be completed; Step 3: Iterative completion and final event context output: Based on the list of semantic elements to be completed, generate element-oriented queries for targeted retrieval, and integrate the new information obtained from the retrieval into the event context; Step 4: Repeat steps 2 to 3 to form an iterative closed loop of "generation-reflection-completion" until the iteration termination condition is met, and output the final event sequence.

2. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, Step 1 includes: Step 101: Topic-driven initial retrieval: Obtain a set of candidate documents related to the target event topic; Step 102: Webpage content parsing and cleaning: Extract structured text content and remove noise information; Step 103: Progress-based document screening: Identify and retain information containing key developments of the event from candidate documents to form a high-information-density input corpus; Step 104: Initial event context generation: Extract key events from the input corpus and organize them into an ordered sequence to form the initial event development sequence.

3. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, Step 2 includes: Step 201: Semantic Role Coverage Detection: Identify missing basic semantic roles in the event using a large language model. The basic semantic roles include at least one of the following: subject, object, time, and location. Step 202: Domain Framework Matching Verification: Match the event with the predefined semantic framework, check whether the core arguments are complete, and determine whether it conforms to the expression specifications of typical event types; Step 203: Integrate the results of steps 201 and 202 through a semantic confidence scoring mechanism to generate a list of semantic elements to be completed.

4. The iterative event context construction method based on semantic integrity as described in claim 3, characterized in that, The predefined semantic framework is built on the FrameNet structured semantic framework library and includes standardized description frameworks for different event types and the necessary set of core arguments.

5. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, Step 3 includes: Step 301: Generate semantic element-oriented queries: Transform each missing item in the list of semantic elements to be completed into a specific retrieval question; Step 302: Targeted retrieval and information extraction: Obtain new documents related to the query through a search engine, and parse and extract information fragments that can be used to complete the event; Step 303: Information Fusion and Iterative Convergence: Integrate the newly acquired information into the existing event context and update the missing fields; proceed to Step 302 to perform a new round of detection and completion, forming iterative optimization and updating the event context.

6. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, Step 301 further includes: A problem rewriting mechanism is introduced to decouple and simplify the structure of complex semantic retrieval by breaking it down into multiple retrieval sub-problems.

7. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, The iteration termination condition in step 4 includes any of the following: All missing items were successfully filled in; Reach the preset maximum number of iterations; The semantic gain from the newly added information is less than the preset threshold.

8. The iterative event context construction method based on semantic integrity as described in claim 1, characterized in that, The method further includes: sorting all event nodes globally by timestamp before outputting the final event timeline, merging semantically duplicate entries, and eliminating conflicts.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of claim 1.

Images