Risk conduction prediction method and system based on combined deduction of time sequence diagram and large model

Abstract

The invention provides a risk conduction prediction method and system based on combined deduction of a time sequence diagram and a large model, and relates to the technical field of artificial intelligence and knowledge engineering, and the method comprises the steps: extracting a prospective time sequence fact based on a hierarchical cue word and a structured constraint mechanism; extracting conflicts and generating a sequential relationship; the historical time sequence knowledge graph is updated based on the time sequence relation and a confidence coefficient weighted updating strategy; determining a central entity, and initiating a structured query to the updated historical time sequence knowledge graph by taking the central entity as a starting point to generate a context knowledge sub-graph; converting the context knowledge sub-graph into a natural language description with a logic relationship, injecting a risk hypothesis event, constructing a cue word as a new input of a large model, and outputting to obtain a structured JSON object containing a complete reasoning chain; and converting the structured JSON object based on a risk path extraction and visualization algorithm influencing weight attenuation to obtain risk early warning information and a visual conduction path diagram.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and knowledge engineering technology, specifically to a risk transmission prediction method and system based on joint inference of time series diagrams and large models. Background Technology

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In today's rapidly globalizing and digitalizing world, the complexity and vulnerability of various systems, especially global supply chain networks, are becoming increasingly apparent. Even a small, localized disturbance, such as a production accident at a key supplier, can quickly propagate and amplify through complex and opaque networks, ultimately triggering systemic and catastrophic disruptions. Therefore, early, accurate, and interpretable prediction of such risks has become a core requirement for enterprises to maintain competitiveness and resilience. Traditional risk warning methods, particularly those based on statistical models and expert systems, primarily rely on statistical analysis of structured information such as historical financial and inventory data (e.g., time series forecasting), combined with a rule base of industry experts. The shortcomings of this approach are: heavy reliance on structured data, inability to utilize the explosive growth of unstructured information such as news, announcements, and research reports on the internet, leading to a significant lag in risk perception; and difficulty in maintaining and expanding expert rules, making it unable to cope with the rapidly changing and highly complex modern supply chain networks.

[0004] With the development of technology, schemes for risk prediction using machine learning models (such as LSTM) and graph computing techniques (such as graph neural networks GNN) have emerged. For example, by constructing the supply chain as a graph, GNN is used to learn how historical disruption events propagate in the graph. Compared with traditional methods, these schemes have improved prediction accuracy, but existing machine learning methods still have the following defects: (1) Lack of interpretability: The prediction process of deep learning models such as GNN is a "black box". It can tell you that "Company A's supply chain disruption risk probability is 90%", but it cannot tell you why it is 90%, where the risk will come from, which links it will go through, and who it will ultimately affect. This makes it difficult for decision-makers to trust the results, let alone make accurate interventions based on them. (2) Inability to deal with "black swan" events: The upper limit of the capabilities of this type of data-driven model is limited by the "historical experience" of the training data. For those "black swan" risk events that have never occurred in history or are rare (for example, the emergence of a new type of trade barrier), due to the lack of learning samples, the model often cannot make any effective predictions. (3) Limitations of information sources: Although some solutions have begun to try to use NLP technology to extract events from text, most of them are limited to extracting “strong signal” events with clear trigger words (such as “bankruptcy”, “fire”). Their ability to capture “weak signals” that are more forward-looking and hidden in analysts’ opinions or industry dynamics is very limited.

[0005] Recently, large language models (LLMs) have demonstrated powerful natural language understanding and preliminary reasoning abilities, leading some practitioners to attempt to directly ask LLMs to predict the impact of certain events. However, this "naive" approach of direct application still suffers from the following problems: (1) The problem of “fact illusion”: The general language model lacks real-time updated and accurate structured knowledge of specific domains (such as the supply chain of a specific enterprise). When it makes inferences, its factual basis is vague and unreliable, which makes it very easy to produce “fact illusion”, resulting in its predictions “sounding reasonable but not consistent with the facts”, which is unacceptable in serious business decisions.

[0006] (2) Shallowness and instability of reasoning: Lacking the guidance of structured knowledge, the reasoning process of LLM is often divergent and shallow, making it difficult to conduct multi-step and rigorous logical deductions, and the results are unstable and very sensitive to changes in prompts.

[0007] (3) It cannot adapt to complex and ever-changing scenarios and lacks an industrial-grade solution that deeply integrates temporally changing and structured domain knowledge with the powerful reasoning and natural language generation capabilities of large language models. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a risk transmission prediction method and system based on joint inference using time-series graphs and large-scale models. Through a hierarchical prompting mechanism, forward-looking time-series events are extracted from unstructured data to construct a dynamic time-series knowledge graph. Combined with a large-scale model joint inference framework, the contextual knowledge subgraphs of the time-series knowledge graph serve as a reliable factual basis for reasoning. By injecting "hypothetical events," the large language model is guided to perform chain-like, interpretable causal reasoning, thereby simulating and inferring the complete transmission path of risk events in complex networks.

[0009] According to some embodiments, the present invention adopts the following technical solution: Risk transmission prediction methods based on joint extrapolation of time series diagrams and large-scale models include: Acquire diverse heterogeneous unstructured data; Chain reasoning is performed on heterogeneous unstructured data. Based on hierarchical prompts and structured constraint mechanisms, forward-looking impact events are identified and extracted, and formatted into time-series facts. Retrieve existing facts that conflict with time-series facts from the historical time-series knowledge graph, and construct a set of conflicting facts. For each conflicting fact, use time-series relation algebra to make a conflict judgment and obtain the time-series relation. Update the historical time-series knowledge graph based on the time-series relation and a confidence-weighted update strategy. Initiate a risk transmission prediction request and identify the central entity. Starting from the central entity, initiate a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. The serialization method based on knowledge subgraph-natural language narrative converts the context knowledge subgraph into a natural language description with logical relationships, injects risk hypothetical events into the natural language description, constructs highly structured and formatted output constraint prompts as new inputs to the large model, and outputs a structured JSON object containing a complete inference chain. Based on the risk path extraction and visualization algorithm that affects weight decay, the structured JSON object is transformed to obtain risk warning information and a visualized transmission path diagram.

[0010] According to some embodiments, the present invention adopts the following technical solution: A risk transmission prediction system based on joint extrapolation of time series diagrams and large-scale models includes: The time-series event extraction module is used to acquire various heterogeneous unstructured data, perform chain reasoning on the heterogeneous unstructured data, identify and extract forward-looking impact events based on hierarchical prompt words and structured constraint mechanisms, and format them into time-series facts; The dynamic temporal knowledge graph management module is used to retrieve existing facts that conflict with temporal facts in the historical temporal knowledge graph, filter and construct a set of conflicting facts, and for each conflicting fact, use temporal relation algebra to make a conflict judgment to obtain the temporal relation; and update the historical temporal knowledge graph based on the temporal relation and a confidence-weighted update strategy. The large-scale model joint inference and prediction module is used to initiate risk transmission prediction requests and identify the central entity. Starting from the central entity, it initiates a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. Based on the knowledge subgraph-natural language narrative serialization method, the contextual knowledge subgraph is converted into a natural language description with logical relationships. Risk hypothetical events are injected into the natural language description, and highly structured and formatted output constraint prompts are constructed as new inputs to the large-scale model. The output is a structured JSON object containing a complete inference chain. The risk transmission path generation and early warning module is used to transform structured JSON objects based on risk path extraction and visualization algorithms with diminishing impact weights, to obtain risk warning information and a visualized transmission path diagram.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention presents a risk transmission prediction method based on joint inference using time-series graphs and large-scale models. Through a hierarchical prompting word mechanism, it utilizes a large language model to extract "prospective" time-series events from unstructured data, updating and constructing a dynamic time-series knowledge graph. Based on the large-scale model joint inference framework, the contextual knowledge subgraph of the time-series knowledge graph serves as a credible factual basis for the large language model's reasoning. By injecting hypothetical risk events, it guides the large language model to perform chain-like, interpretable causal reasoning, thereby simulating and deducing the complete transmission path of risk events in complex networks (such as supply chain networks). This fully realizes how to start from a vague, unstructured "weak signal" influencing event, dynamically model it using the time-series knowledge graph, and then utilize the "hypothetical reasoning" capability of the large language model to ultimately generate an interpretable, traceable, and visualized accurate risk warning.

[0012] The risk transmission prediction method based on the joint inference of time series diagrams and large-scale models of this invention, through the extraction of time series events and the deep reasoning capabilities of large-scale language models, identifies and extracts weak signals and forward-looking risk events from massive amounts of unstructured information that are undetectable by traditional methods. This advances the starting point of risk warning from "the event has already occurred" to the stage of "the risk is brewing," providing enterprises with a valuable window of opportunity to take countermeasures, achieving foresight in risk perception, and improving the timeliness of early warning.

[0013] The risk transmission prediction method based on joint inference of time series graphs and large-scale models in this invention constructs a framework for joint inference of time series graphs and large-scale models. This framework does not rely on the existence of similar risk patterns in historical data, but rather utilizes the common sense and world knowledge of large-scale language models to make reasonable and logically consistent hypothetical inferences about novel risk events that have never occurred in history. This can solve the problem that existing data-driven models are unable to cope with "black swan" risks, essentially achieving zero-sample inference for "black swan" events.

[0014] The risk transmission prediction method of this invention, based on joint inference of time series diagrams and large-scale models, is fundamentally different from the "black box" nature of traditional prediction models. This invention outputs not only a risk probability, but also a human-readable and logically clear reasoning chain generated by a chain-like causal reasoning process. Furthermore, this chain is transformed into a visualized risk transmission path diagram, making each prediction conclusion traceable. Decision-makers can clearly understand the source, transmission path, and ultimate impact of the risk, thereby greatly enhancing the trust in the early warning results, achieving "white box" interpretability of the prediction process, and enhancing the credibility of decision-making.

[0015] The risk transmission prediction method based on joint inference of time series graphs and large-scale models in this invention employs a rigorous, credibility-weighted update algorithm to maintain and update the time series knowledge graph. This algorithm can handle the conflict, overlay, and fusion of new and old information, ensuring that the contextual knowledge subgraph, which serves as the basis for large-scale model reasoning, is always dynamically updated, logically consistent, and of high quality. This provides a solid data guarantee for the accuracy of subsequent reasoning and prediction, ensuring the dynamism and consistency of the knowledge base and improving the accuracy of reasoning.

[0016] The risk transmission prediction method based on joint inference of time series diagrams and large models in this invention does not perform local optimization of a single link in existing risk prediction technologies. Instead, it provides a complete, end-to-end next-generation risk prediction solution from early signal perception to deep causal inference by deeply coupling a series of original processes, including "forward-looking event extraction," "dynamic time series knowledge modeling," "joint inference of large models," and "explainable path generation." This solution addresses the core pain points of traditional methods and existing AI models, such as lagging risk perception, the "black box" nature of the prediction process, and the inability to cope with new "black swan" risks. While significantly improving the timeliness and accuracy of early warnings, it ensures the transparency and explainability of the prediction process, and can drive the field of risk management towards higher-level intelligent and cognitive decision support, possessing high industrial value. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0018] Figure 1 This is a schematic diagram of the overall architecture of the risk transmission prediction method based on joint inference of time series diagrams and large models according to an embodiment of the present invention; Figure 2 This is an overall flowchart of the risk transmission prediction method based on joint inference of time series diagrams and large models according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the forward-looking impact event extraction process according to an embodiment of the present invention. Figure 4 This is a core flowchart of the joint deduction framework of an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0021] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0022] Example 1 One embodiment of the present invention provides a risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models, the method steps of which include: Step 1: Acquire various heterogeneous unstructured data, perform chain reasoning on the heterogeneous unstructured data, identify and extract forward-looking influencing events based on hierarchical prompts and structured constraint mechanisms, and format them into time-series facts; Step 2: Retrieve existing facts that conflict with the time-series facts in the historical time-series knowledge graph, and construct a set of conflicting facts. For each conflicting fact, use time-series relation algebra to make a conflict judgment and obtain the time-series relation. Update the historical time-series knowledge graph based on the time-series relation and a confidence-weighted update strategy. Step 3: Initiate a risk transmission prediction request and identify the central entity. Starting from the central entity, initiate a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. Based on the knowledge subgraph-natural language narrative serialization method, convert the contextual knowledge subgraph into a natural language description with logical relationships. Inject hypothetical risk events into the natural language description, construct highly structured and formatted output constraint prompts as new inputs to the large model, and output a structured JSON object containing a complete inference chain. Step 4: Based on the risk path extraction and visualization algorithm that affects weight decay, the structured JSON object is transformed to obtain risk warning information and a visualized transmission path diagram.

[0023] As one embodiment, the risk transmission prediction method based on joint inference of time-series graphs and large-scale models of the present invention utilizes a large language model to extract forward-looking impact events from unstructured data to update a dynamic time-series knowledge graph. More importantly, it proposes a "joint inference of large-scale models" framework: it uses the "contextual knowledge subgraph" of the time-series knowledge graph as a credible factual basis for the large language model's reasoning, and guides the large language model to perform chain-like, interpretable causal reasoning by injecting hypothetical risk events, thereby simulating and inferring the complete transmission path of risk events in complex networks. The specific implementation process is as follows: Step 1: Acquire various heterogeneous unstructured data; perform chain reasoning on the heterogeneous unstructured data, identify and extract forward-looking influencing events based on hierarchical prompts and structured constraint mechanisms, and format them into time-series facts; The core task of step 1 is to identify and extract various events affecting the target system (such as the supply chain) in real time from massive, heterogeneous, and unstructured data sources. The innovation of this process lies in the fact that it can not only identify "strong signal" events that have already occurred, but also discover and structure "weak signal" and "prospective" impact events that are difficult to capture by traditional methods through a certain method.

[0024] In one specific embodiment, prompt words are used to assign explicit expert roles to the LLM model and inject domain knowledge. The prompt words decompose the task of extracting heterogeneous unstructured data into a series of simple and continuous thinking steps, guiding the model to think in a chain. Finally, JSON Schema constraints are provided as the output format template, and the model is finally extracted to extract forward-looking, predictive weak signal influencing events. Each extracted influencing event is formatted as a structured time-series fact data unit containing metadata of event subject, object, relationship, timestamp, source and confidence level.

[0025] This involves continuously monitoring various data sources, including but not limited to industry news portals, relevant departmental announcements, listed company financial reports, brokerage analysis reports, social media, and internal company emails and meeting minutes. Its workflow differs from existing event extraction methods that rely on explicit keywords; instead, it employs an event extraction method based on hierarchical prompt words and structured constraint templates. This method does not present a single, general question to the Large Language Model (LLM) all at once, but rather ensures the accuracy and consistency of extraction through a multi-layered, progressively focused prompt chaining. This transforms the uncertainty of LLM into a stable and reliable structured data generator. The specific implementation steps are as follows: First layer: Role-playing and domain knowledge injection.

[0026] By using prompts to define clear expert roles for the LLM (Limited Language Model) and injecting core domain knowledge required for the task, subsequent LLM analysis can focus on specific domains and make judgments using a domain-specific perspective. For example, in a supply chain risk warning scenario, the beginning of the prompt might define: "You are a supply chain risk analysis expert specializing in the semiconductor industry. The key risk categories you need to focus on include: [geographical risks, capacity bottlenecks, technology substitution, and fluctuations in key raw material prices]." The second layer: task decomposition and chain thinking guidance.

[0027] The prompts break down complex extraction tasks into a series of simple, sequential thought processes, guiding the model to perform chain-of-thought (CoT). This guidance forces the model to demonstrate its logical reasoning process before giving a final conclusion, thereby improving the accuracy of complex tasks. For example, the prompts might continue to guide the analysis of the text by following these steps: (1) Read through the entire text and identify all mentioned company entities and geographical locations. (2) For each entity, find sentences that describe its current or future state. (3) Determine whether these state descriptions are relevant to the key risk categories we are concerned with. (4) If relevant, determine whether this is a 'realized fact' or a 'potential future risk'. Third layer: Structured output constraint template.

[0028] This layer is a crucial step in ensuring output quality and machine readability. The final part of the prompt provides a strict JSON schema as the output format template, optionally supplemented with one or more "few-shot" examples to further standardize the output. This strong constraint ensures that regardless of changes in the input text, the LLM output is uniform, structured data that can be directly parsed and consumed by downstream modules. A typical output constraint template example is as follows: / / ... (Content of the first two layers of prompts) ... Please strictly adhere to the following JSON format when outputting your analysis results, without adding any additional explanations:\n "{\n" "\"events\": [\n" "{\n" ""event_description": "<Text description of the event>",\n" "entities": [{"name": "<company name>", "role": "subject / object"}],\n" ""event_type": "<Previously Occurred Facts | Potential Future Risks>",\n" "risk_category": "<Geopolitical risks | ...>",\n" "\"time_window\": {\"start\": \"YYYY-MM-DD\", \"end\": \"YYYY-MM-DD\"},\n" "\"confidence_score\":<floating-point number between 0.0 and 1.0>,\n" "source_quote": "<Original quote>" "}\n" "]\n" "}".

[0029] By employing the aforementioned layered, progressively guided, and ultimately strongly constrained output format approach, LLM's deep semantic understanding and reasoning capabilities can extract forward-looking, predictive "weak signal" events from analyst reports, such as "Company B is overly reliant on its upstream supplier C, and its production stability faces a high risk in the next six months." Each extracted event is formatted as a structured "time-series fact" data unit containing metadata such as the event subject, object, relationship, timestamp, source, and confidence level, providing high-quality input for the next module to construct a "foresightful" knowledge graph.

[0030] Step 2: Retrieve existing facts that conflict with the time-series facts in the historical time-series knowledge graph, and filter and construct a set of conflicting facts. For each conflicting fact, use time-series relation algebra to make a conflict judgment and obtain the time-series relation. Update the historical time-series knowledge graph based on the time-series relation and a confidence-weighted update strategy. Step 2 is responsible for constructing, updating, storing, and dynamically maintaining a temporal knowledge graph (TKG) that accurately reflects the evolution of the real world over time. The core data structure of this TKG differs from traditional static triples; instead, it employs an enhanced quintuple form that includes confidence scores: F = (h, r, t, T, C), where h is the head entity, r is the relation, t is the tail entity, T represents the valid time interval of the fact [T_start, T_end], and C represents the confidence score of the fact [0, 1]. The historical temporal knowledge graph records every confirmed event relation valid within a historical timeframe. For example, a fact (Company A, Purchasing CPUs, Company B, [2023-01-01, 2023-12-31], 1.0) precisely records a confirmed purchasing relationship valid in 2023.

[0031] Specifically, when a new temporal fact is extracted, the module first retrieves all existing facts in the historical temporal knowledge graph that have the same subject-verb-object triple (h, r, t) as the temporal fact, forming a candidate fact set. From this candidate fact set, all conflicting facts where time intervals T_existing and T_new overlap are selected, forming a conflicting fact set. For each existing fact in the conflicting fact set, the module uses formalized temporal relation reasoning theory to determine one of the 13 possible temporal relations between T_new and T_existing. Based on the determined temporal relations and the confidence level of the fact, different historical temporal knowledge graph update strategies defined by formulas are executed.

[0032] As one embodiment, the core function of step 2 is to implement event-driven dynamic updates of the temporal knowledge graph based on a confidence-based temporal fact update algorithm, thereby ensuring the consistency, accuracy, and timeliness of the graph knowledge. When a new temporal fact F_new = (h, r, t, T_new, C_new) is received from the temporal event extraction module, the module does not simply add it, but performs the following rigorous conflict resolution and fusion process: Step (1): Retrieve the conflicting fact set; First, retrieve all existing facts in the graph that have the same subject-verb-object triple (h, r, t) as the temporal fact F_new obtained in step 1, forming a candidate fact set. Then, from this candidate set, select all facts where the time intervals T_existing and T_new overlap (i.e., T_existing ∩ T_new ≠ ∅), forming the final conflict fact set S_conflict.

[0033] Step (2): Apply temporal relational algebra to determine conflicts; For each existing fact F_existing in the conflict fact set S_conflict, the module uses Allen's Interval Algebra (a formal theory of temporal relational reasoning) to accurately determine one of the 13 possible temporal relations between T_new and T_existing (such as Equals, Starts, Finishes, Overlaps, During, etc.).

[0034] Step (3): Credibility-weighted historical time-series knowledge graph update strategy and formulaic decision-making; Based on the determined temporal relationships and the confidence level of the facts, different update strategies defined by formulas are executed, rather than vague business rules.

[0035] A: Override & Invalidate Strategy: This strategy is applied when the confidence level of a new time-series fact is significantly higher than that of an older time-series fact. This invention defines the confidence advantage function Adv(C_new, C_existing): Adv(C_new, C_existing) = C_new / (C_existing + ε) (where ε is a small constant to prevent the denominator from being zero); If Adv(C_new, C_existing) > θ_override (e.g., coverage threshold θ_override = 2.0), then the time-series fact F_new is considered overwhelmingly dominant. In this case, the validity of the time interval for all F_existing instances that overlap with T_new will be "truncated" by T_new. For example, if T_existingOverlapsT_new, then the end time of F_existing, T_existing_end, will be updated to T_new_start. The time-series fact F_new is then fully added to the historical time-series knowledge graph. This ensures that information from more reliable sources (such as official announcements C=1.0) can overwrite previous market rumors (such as C=0.5).

[0036] B: Merge & Enhance strategy: This strategy applies when two facts describe the same continuous state and have similar confidence levels (i.e., Adv values ​​are between 1 / θ_override and θ_override). For example, if T_new Meets T_existing (i.e., T_new_start equals T_existing_end), these two facts are merged into a new time-series fact F_merged with a longer time span. Its new confidence level C_merged can be calculated using a confidence fusion function, such as a weighted average. C_merged = (C_new * len(T_new) + C_existing * len(T_existing)) / (len(T_new) + len(T_existing)) (where len(T) represents the length of the time interval); This ensures that consecutive descriptions of the same event from different reports can be integrated.

[0037] C: Tag & Coexist Strategy: When new and old facts contradict each other in content within overlapping time intervals, but the confidence level is not overwhelmingly dominant, the information is considered conflicting. In this case, both temporal facts F_new and F_existing are retained in the historical temporal knowledge graph, but they are each given a special "conflict" tag, and a "conflict pointer" is created pointing to the other fact ID. This allows higher-order reasoning modules (such as the third module) to recognize the uncertainty in the knowledge at this point during deduction and to make more prudent judgments accordingly.

[0038] Step 3: Initiate a risk transmission prediction request and identify the central entity. Starting from the central entity, initiate a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. Based on the knowledge subgraph-natural language narrative serialization method, convert the contextual knowledge subgraph into a natural language description with logical relationships. Inject hypothetical risk events into the natural language description, construct highly structured and formatted output constraint prompts as new inputs to the large model, and output a structured JSON object containing a complete inference chain. Step 3 proposes a joint inference framework based on contextual knowledge subgraphs and large language models (LLMs) for hypothetical risk events. Its innovation lies in the fact that it eliminates the need for fine-tuning or specialized training of the large language model. Instead, it utilizes and constrains the zero-shot reasoning capabilities of the general large language model through an engineering approach, thereby achieving "white-box" prediction of risk transmission paths.

[0039] This process, based on risk transmission prediction requests initiated from time-series facts (e.g., analyzing the impact of "Company B's production halt" on "Company A"), executes a rigorous, multi-step deduction process. The specific implementation process is as follows: Step (1): Identify the central entity. Starting from the central entity, initiate a structured query to the updated historical time-series knowledge graph. The query extracts all entities within a preset number of hops related to the central entity and their inter-entity temporal facts that are still valid up to the current time. The query returns a local contextual knowledge subgraph that is highly relevant to the current prediction task. This step aims to provide an accurate, complete, and clearly defined factual foundation for subsequent LLM inference. First, the "Contextual Knowledge Subgraph Extraction" operation is performed. It does not use the entire massive knowledge graph, but rather the following: A. Identify the central entity: Based on the risk transmission forecast request, identify one or more central entities (e.g., "Company A").

[0040] B. Perform an N-degree relational query: Starting from the central entity, initiate a structured query into the historical time-series knowledge graph. This query will extract all entities within a preset number of hops (N degrees, e.g., N=3) of the central entity and their respective time-series facts that are still valid up to the current time point.

[0041] C. Generating a Subgraph: The query returns a local, contextual knowledge subgraph that is highly relevant to the current task. This subgraph is the complete and reliable set of background facts that LLM needs to refer to when performing inference.

[0042] Step (2): Based on the knowledge subgraph-natural language narrative serialization method, the context knowledge subgraph is converted into a natural language description with logical relationships. This includes: taking the central entity of the context knowledge subgraph as the starting point of the narrative, performing a breadth-first traversal starting from the central entity, and calling the time-series fact to natural language template converter to convert it into a complete sentence when each timeseries is reached. When multiple relations point to the same entity, logical connectors are used to organize the independent sentences into a logically clear natural language description that is easy for large language models (LLM) to predict. This step adopts an innovative knowledge subgraph-to-natural language narrative serialization method to enable a general large language model that has not been trained in a specific domain to truly "understand" and utilize the structural information of the graph. This method converts the machine-readable context knowledge subgraph extracted in the previous step into a coherent natural language description with logical relationships. The process is as follows: A. Determine the narrative center: Use the central entity of the contextual knowledge subgraph (such as "Company A") as the starting point of the narrative.

[0043] B. Relationship Template Transformation: Starting from the central entity, a breadth-first traversal is performed. Upon reaching a temporal knowledge graph (h, r, t, T, C), a temporal fact-to-natural language template converter is invoked to transform it into a complete sentence. For example, (Company A, CPU procurement, Company B, [2023-01-01, ...], 1.0) will be transformed into: "Fact: Based on accurate information (confidence level 1.0), since January 1, 2023, Company A's main CPU supplier has been Company B." C. Intelligent insertion of logical connectors: When multiple relations point to the same entity, logical connectors such as "at the same time", "in addition", and "its upstream supplier is..." are used to organize independent sentences into a logically clear paragraph that is easy for LLM to understand.

[0044] Ultimately, the input to LLM is no longer a simple data structure, but a background factual statement similar to an expert analysis report, which greatly reduces the cognitive burden of general large models in understanding complex domain knowledge.

[0045] Step (3): Inject hypothetical risk events into the natural language description. The hypothetical risk events are risk events that trigger the risk transmission prediction task. Injecting the hypothetical risk events into the natural language description forms a highly structured prompt word that includes role-playing, context injection, thought chain guidance, and formatted output constraints. This prompt word serves as a new input to the large language prediction model. The output is a structured JSON object that includes the reasoning_chain and the final_alert.

[0046] After generating the narrative "background fact statement", the following operations are performed: A. Inject the risk event that triggers this forecast simulation (e.g., "Hypothetical risk event: Company B suspends production due to a fire, with an estimated recovery time of 6 months") into the background fact statement as a new input.

[0047] B. Submit all the above information to the large language model through a highly structured prompt word that includes role-playing, context injection, chain-of-thought guidance, and formatted output constraints.

[0048] The design of these cue words is central to how this invention "uses" rather than "trains" a general-purpose LLM. A typical cue word structure is as follows: ### Role ### As a professional supply chain risk analysis expert, all your reasoning must be strictly based on the "background fact statements" provided below; ### Background Facts (Contextual Knowledge Subgraph) ### <Insert a story-based natural language description generated in step two here> ### Hypothetical Event ### <Insert hypothetical event description here> ### Task&Chain-of-Thought Instruction ### Based on the above "background factual statements" and the "hypothetical event" that occurred, please **infer step by step, in a chain-like manner** the impact of this event on Company A's next two quarters; your reasoning must form a causal chain, clearly indicating which fact led to which new change; ### Output Format Constraint ###; Please strictly follow the following JSON format to return your complete reasoning process and final conclusion, without adding any additional explanations; <Detailed JSON Schema is provided here, including fields such as reasoning_chain and final_alert>.

[0049] Through this rigorous framework, this invention transforms complex reasoning tasks into a "reading comprehension and continuation writing" task—one that LLM excels at, with clear context and logical guidance. Guided by this framework, the output of the large language model is no longer divergent and unreliable natural language text, but a structured JSON object containing the complete reasoning process. This object meticulously records how each step, starting from the initial event, affects a fact in the graph, triggering the next chain reaction, until the final assessment of the impact on the core business of target entity A. This process simulates the deep thinking and logical deduction processes of human experts facing complex problems, without relying on any modifications or training to the model itself.

[0050] Step 4: Based on the risk path extraction and visualization algorithm that affects weight decay, the structured JSON object is transformed to obtain risk warning information and a visualized transmission path diagram.

[0051] The core task of step 4 is to transform the complex, machine-readable inference results output by the large-scale model's joint inference and prediction module into risk warning information and visual pathways that decision-makers can intuitively understand, quickly focus on, and act upon. This is not a simple data format conversion, but rather the refinement and enhancement of the inference information through an algorithm.

[0052] In one specific implementation, a risk path extraction and visualization algorithm based on the decay of influence weights is initiated based on a structured JSON object containing `reasoning_chain` and `final_alert`. The specific implementation process of this algorithm is as follows: Step (1): Construction of the initial transmission path graph; First, the reasoning_chain array is traversed, and each JSON object in the array (representing a reasoning step) is created as a node in a graph. Directed edges are established between the nodes according to their order in the array, thereby constructing an initial risk transmission path graph. Each node contains raw information such as event description and impact analysis for that step.

[0053] Step (2): Quantify the allocation of influence weights; To rank and filter the importance of paths, the algorithm uses a quantified impact score (a floating-point number in the range [0, 1]) generated by the large language model for evaluating each inference step. This score represents the strength of the causal chain at the current step or the severity of its impact. The algorithm uses this impact score as the initial weight S of its corresponding directed edge.

[0054] Step (3): Filter key transmission paths with high influence intensity based on path pruning operation with weight decay; To avoid presenting numerous secondary, decision-confusion-inducing transmission paths (information overload) in the final visualization, the algorithm performs a path pruning operation. This operation is based on the calculation of a cumulative influence weight W, including: A. Weight initialization: Define the cumulative impact weight W(N_0) of the initial risk event (i.e., the first node N_0 in the reasoning_chain) as 1.0.

[0055] B. Weight decay propagation: The algorithm propagates forward along the propagation path, calculating the cumulative influence weight of each subsequent node N_i. The calculation formula can be expressed as: W(N_i) = W(N_{i-1}) * S_i Where W(N_{i-1}) is the cumulative impact weight of the previous node, and S_i is the weight of the edge from N_{i-1} to N_i (i.e., impact_score). This formula simulates the natural decay of energy during the transmission of risk.

[0056] C. Perform pruning: A preset "minimum influence weight threshold θ_prune" is used (e.g., 0.1, which can be configured by the user). During the calculation process, if the algorithm finds that the cumulative influence weight W(N_i) of a node N_i is less than θ_prune, then that node N_i and all its subsequent propagation paths will be pruned and removed from the graph.

[0057] This step ensures that only the critical transmission paths that maintain a high level of influence throughout are preserved.

[0058] Step (4): Visualize the mapping and highlighting; After obtaining the pruned, focused risk transmission path diagram, it is mapped to the final visualization chart. To further enhance readability, based on the key transmission paths, the calculated weight values ​​are mapped and highlighted onto the visual attributes of the chart, resulting in the visualized risk transmission path diagram, including: A. Node Highlighting: The color or size of a node is proportional to its cumulative impact weight W. The node that is most severely impacted (with a still high W value) will be the most prominent in the chart.

[0059] B. Edge Highlighting: The thickness or style of the edge (e.g., solid line vs. dashed line) is proportional to its initial weight S. This allows decision-makers to immediately identify which step in the transmission is the key link in amplifying risk.

[0060] Step (5): Generation of structured early warning information; Finally, based on the final_alert part of the output (containing risk level, summary, etc.), it is bound to the visualized risk transmission path diagram generated in the previous step. This will generate a final, complete alert package, which includes: A. High-level warning summary: For example, "Severe risk: The production stoppage of Tier 1 supplier B will have a severe impact on your core product line within 2 quarters." B. Interactive visual transmission path diagram: Users can click on the nodes in the diagram to view the detailed reasoning description and original evidence source for that step.

[0061] C. Recommended measures (optional): These can be initially generated by the LLM based on the risk type and transmission path, or matched by the system based on a preset rule base.

[0062] This invention uses a complete algorithm to extract information from original reasoning to final decision support information, transforming complex AI reasoning results into a decision support "battle map" that is clear in its priorities, highlights key points, provides clear cause and effect, and is interactive, thereby improving the readability, credibility, and practical value of early warning information.

[0063] Example 2 One embodiment of the present invention provides a risk transmission prediction system based on joint extrapolation of time series diagrams and large-scale models, comprising: The time-series event extraction module is used to acquire various heterogeneous unstructured data, perform chain reasoning on the heterogeneous unstructured data, identify and extract forward-looking impact events based on hierarchical prompt words and structured constraint mechanisms, and format them into time-series facts; The dynamic temporal knowledge graph management module is used to retrieve existing facts that conflict with temporal facts in the historical temporal knowledge graph, filter and construct a set of conflicting facts, and for each conflicting fact, use temporal relation algebra to make a conflict judgment to obtain the temporal relation; and update the historical temporal knowledge graph based on the temporal relation and a confidence-weighted update strategy. The large-scale model joint inference and prediction module is used to initiate risk transmission prediction requests and identify the central entity. Starting from the central entity, it initiates a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. Based on the knowledge subgraph-natural language narrative serialization method, the contextual knowledge subgraph is converted into a natural language description with logical relationships. Risk hypothetical events are injected into the natural language description, and highly structured and formatted output constraint prompts are constructed as new inputs to the large-scale model. The output is a structured JSON object containing a complete inference chain. The risk transmission path generation and early warning module is used to transform structured JSON objects based on risk path extraction and visualization algorithms with diminishing impact weights, to obtain risk warning information and a visualized transmission path diagram.

[0064] Example 3 In one embodiment of the present invention, a typical electronic product supply chain is used as a background to illustrate in detail how the risk transmission prediction system based on the joint inference of time series diagrams and large models provided by the present invention achieves the entire process of risk transmission prediction based on the joint inference of time series diagrams and large models, from early signal detection to in-depth inference and finally early warning.

[0065] I. Scene Setting: (1) Central entity (core enterprise): Company A, a smartphone manufacturer.

[0066] (2) Key component: high-performance CPU chip.

[0067] (3) Initial supply chain state (already exists in the historical time-series knowledge graph): Fact 1: (Company A purchases CPUs, Company B, [2023-01-01, ...], 1.0); Fact 2: (Company B, purchasing wafers, Company C, [2023-01-01, ...], 1.0); Fact 3: (Company A has a backup CPU supplier, none, [2023-01-01, ...], 1.0); As an example, based on the above scenario, the specific implementation process of the risk transmission prediction method based on the joint inference of time series diagrams and large models of the present invention for risk transmission prediction is as follows: Step 1: Identification of hypothetical events and generation of time-series facts based on the influence of weak signals; During the continuous monitoring of information across the entire network, the system's time-series event extraction module obtained an industry analysis report published by a securities firm regarding the region where Company C is located.

[0068] (1) First, the hierarchical prompt words and structured constraint template mechanism within the time-series event extraction module are activated. Under the role of a "semiconductor industry analyst" and guided by the CoT, the Large Language Model (LLM) did not find strong signal words such as "production stoppage" or "interruption," but it identified a key "weak signal" description: "...affected by increasingly tense labor relations in the region, wage negotiations between the union and major wafer manufacturer Company C have reached a stalemate. Analysts generally believe that if an agreement cannot be reached within the next three months, the possibility of a large-scale strike at Company C is significantly increasing..." (2) Based on the structured output template, LLM generates a "prospective" time-series fact with confidence and transmits it to the next dynamic time-series knowledge graph management module. The generated time-series fact is as follows: { "event_description": "Company C faces the potential risk of a large-scale strike", "entities": [{"name": "Company C", "role": "main entity"}], "event_type": "potential future risks", "risk_category": "Labor Relations Risk", "time_window": {"start": "2024-06-01", "end": "2024-08-31"}, "confidence_score": 0.70, "source_quote": "..." }

[0069] Step 2: Dynamic updating of the time-series knowledge graph; The dynamic temporal knowledge graph management module receives temporal facts sent by the temporal event extraction module: (1) Since this is a brand new “potential risk” event and there are no conflicting facts in the historical timeline knowledge graph, this new quintuple is directly added to the historical timeline knowledge graph: F4 = (Company C, facing strike risk, union negotiation deadlock, [2024-06-01, 2024-08-31], 0.70).

[0070] (2) At this point, the updated historical time series knowledge graph not only records the current stable supply relationship, but also explicitly marks a future, potential risk hypothetical event.

[0071] Step 3: Joint extrapolation and prediction based on hypothetical risk events; A supply chain risk manager wants to assess "what impact would a strike at Company C have on Company A's mobile phone production?" This can be achieved by initiating a risk propagation prediction task through system interface interaction.

[0072] (1) Triggering the simulation: After initiating the risk transmission prediction request, the task is sent to the large model joint simulation and prediction module.

[0073] (2) Extracting contextual knowledge subgraphs: The large model joint inference and prediction module takes "Company A" as the center and extracts facts 1, 2, 3, and 4, which contain the three companies A, B, and C and their interrelationships, to form the "background fact statement" for this inference.

[0074] (3) Hypothetical events of risk injection: The scenario set by the risk manager is: "Hypothesis: Company C will experience a complete interruption of wafer production due to a large-scale strike starting from July 1, 2024, which will last for at least one quarter." (4) Perform joint inference: The large model joint inference and prediction module submits the narrative "background fact statement" and "hypothetical event" to the general large language model through the designed "chain causal reasoning guidance" prompt words.

[0075] (5) Generate inference chain: Under the guidance of the framework, LLM outputs a structured JSON inference chain: { "reasoning_chain": [ { "step": 1, "event": "Company C's wafer production is interrupted", "impact": "Company C is unable to supply wafers to Company B.", "impact_score": 1.0}, { "step": 2, "event": "Company B loses wafer supply", "impact": "Company B's CPU production line will cease production within 2-4 weeks due to raw material depletion.", "impact_score": 0.95}, { "step": 3, "event": "Company B's CPU production has ceased", "impact": "Company B will be unable to fulfill its CPU supply contract with Company A.", "impact_score": 1.0}, { "step": 4, "event": "Company A is unable to obtain CPUs from Company B", "impact": "Because the knowledge base shows that Company A has no alternative suppliers, its smartphone production line will face a supply disruption of core components.", "impact_score": 1.0}, { "step": 5, "event": "Company A's production line is disrupted", "impact": "This is expected to delay the release of Company A's next-generation flagship phone by at least one quarter and result in significant revenue losses.", "impact_score": 0.85} ], "final_alert": { "level": "Severe", "summary": "The potential strike risk from Tier 2 supplier C will be directly transmitted to our company within 1-2 months through supplier B, causing a catastrophic impact on our core product lines."} }

[0076] Step 4: Visualization and early warning of risk transmission paths; The risk transmission path generation and early warning module receives the above structured JSON object: (1) Path Extraction: The risk transmission path generation and early warning module initiates a risk path extraction and visualization algorithm based on the decay of influence weights. It calculates the cumulative influence weight sequence from step 1 to step 5 of the structured JSON inference chain as 1.0 -> 0.95 -> 0.95 -> 0.95 -> 0.8075. Since the weights of all nodes are higher than the set pruning threshold (e.g., 0.1), the entire critical path is preserved intact.

[0077] (2) Visualization Generation: Based on these weights, a visual transmission path diagram is generated. In the diagram, the "Company C Strike" node is the starting point, and the arrows point sequentially to "Company B Production Stoppage", "Company A Supply Cutoff", and finally to a "Flagship Phone Release Delay" node marked in dark red. The thickness of the edges reflects the impact_score of each step.

[0078] (3) Warning push: Finally, a complete warning report containing the above-mentioned visual path diagram, the summary text in final_alert and the detailed reasoning steps is pushed to the decision dashboard of the supply chain risk manager.

[0079] This embodiment fully demonstrates how the present invention starts from a fuzzy, unstructured "weak signal," models it through dynamic updates of a time-series knowledge graph, and leverages the "hypothesis reasoning" capabilities of a large language model to ultimately generate an interpretable, traceable, and visualized accurate risk warning. This proves the advancements and industrial practical value of the present invention compared to existing technologies.

[0080] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0082] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models, characterized in that, include: Acquire diverse heterogeneous unstructured data; Chain reasoning is performed on heterogeneous unstructured data. Based on hierarchical prompts and structured constraint mechanisms, forward-looking impact events are identified and extracted, and formatted into time-series facts. Retrieve existing facts that conflict with chronological facts from the historical chronological knowledge graph, filter and construct a set of conflicting facts, and use chronological relation algebra to make a conflict judgment for each conflicting fact to obtain the chronological relation. The historical time-series knowledge graph is updated based on a time-series relationship and a confidence-weighted update strategy. Initiate a risk transmission prediction request and identify the central entity. Starting from the central entity, initiate a structured query to the updated historical time-series knowledge graph to generate a contextual knowledge subgraph. The serialization method based on knowledge subgraph-natural language narrative converts the context knowledge subgraph into a natural language description with logical relationships, injects risk hypothetical events into the natural language description, constructs highly structured and formatted output constraint prompts as new inputs to the large model, and outputs a structured JSON object containing a complete inference chain. Based on the risk path extraction and visualization algorithm that affects weight decay, the structured JSON object is transformed to obtain risk warning information and a visualized transmission path diagram.

2. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, By utilizing the LLM model and assigning explicit expert roles to the model using cue words, and injecting domain knowledge, the cue words decompose the task of extracting heterogeneous unstructured data into a series of simple, sequential thinking steps, guiding the model to perform chain-like thinking. Finally, JSON Schema constraints are provided as the output format template, ultimately extracting forward-looking, predictive weak signal influencing events. Each extracted influencing event is formatted as a structured time-series fact data unit containing metadata of event subject, object, relationship, timestamp, source, and confidence level.

3. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, The temporal knowledge graph is in the form of a quintuple including confidence: F = (h, r, t, T, C), where h is the head entity, r is the relation, t is the tail entity, T represents the valid time interval of the fact [T_start, T_end], and C represents the confidence score of the fact [0, 1]. The historical temporal knowledge graph records every confirmed event relation that is valid within a historical time period. When a new temporal fact is extracted, all existing facts with the same subject-verb-object triple (h, r, t) as the temporal fact are first retrieved from the historical temporal knowledge graph to form a candidate fact set. From this candidate fact set, all conflicting facts where the time intervals T_existing and T_new overlap are selected to form a conflicting fact set. For each existing fact in the conflicting fact set, the module uses formal temporal relation reasoning theory to determine one of the 13 possible temporal relations between T_new and T_existing. Based on the determined temporal relation and the confidence of the fact, different historical temporal knowledge graph update strategies defined by formulas are executed.

4. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, Based on this time-series fact, a risk transmission prediction request is initiated, a central entity is identified, and a structured query is initiated to the updated historical time-series knowledge graph starting from the central entity. The query extracts all entities within a preset number of hops related to the central entity and their time-series facts that are still valid up to the current time point. The query results are local context knowledge subgraphs that are highly relevant to the current prediction task.

5. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, The serialization method based on knowledge subgraph-natural language narrative converts the context knowledge subgraph into a natural language description with logical relationships. This includes: taking the central entity of the context knowledge subgraph as the starting point of the narrative, performing a breadth-first traversal starting from the central entity, and calling the time-series fact to natural language template converter to convert it into a complete sentence when each time a time-series fact is encountered. When multiple relations point to the same entity, logical connectors are used to organize the independent sentences into a logically clear natural language description that is easy for large language model LLM inference to predict.

6. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, Hypothetical risk events are injected into the natural language description. These hypothetical risk events are risk events that trigger the risk propagation prediction task. The injection of these hypothetical risk events into the natural language description forms a highly structured prompt word that includes role-playing, context injection, thought chain guidance, and formatted output constraints. This prompt word serves as a new input to the large language prediction model. The output is a structured JSON object that includes the reasoning_chain and the final_alert.

7. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, The structured JSON object is processed using a risk path extraction and visualization algorithm that utilizes the decay of influence weights. This includes: first, traversing the reasoning_chain array, creating a graph node for each JSON object in the array, establishing directed edges between nodes according to their order in the array, thereby constructing an initial risk transmission path graph, sorting and filtering the importance of the paths, using the quantitative impact score impact_score as the initial weight of the directed edges, and employing path pruning operations to filter key transmission paths with high influence intensity.

8. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, Based on the key transmission path, the calculated weight values ​​are mapped onto the visual attributes of the chart to obtain a visualized risk transmission path diagram. In this diagram, the color or size of the node is set to be proportional to its cumulative impact weight, and the node most severely impacted will be the most prominent in the path diagram. The thickness or style of the edges is set to be proportional to the initial weight, so that it can be visualized which step of transmission is the key link in amplifying the risk.

9. The risk transmission prediction method based on joint extrapolation of time series diagrams and large-scale models as described in claim 1, characterized in that, The final alert (final_alert) is bound to a visual risk transmission path diagram to generate a final, complete alert package, which includes a high-level alert summary, an interactive visual risk transmission path diagram, and recommended measures.

10. A risk transmission prediction system based on joint extrapolation of time series diagrams and large-scale models, characterized in that, include: The time-series event extraction module is used to acquire various heterogeneous unstructured data, perform chain reasoning on the heterogeneous unstructured data, identify and extract forward-looking impact events based on hierarchical prompt words and structured constraint mechanisms, and format them into time-series facts; The dynamic temporal knowledge graph management module is used to retrieve existing facts that conflict with temporal facts in the historical temporal knowledge graph, filter and construct a set of conflicting facts, and use temporal relation algebra to make conflict judgments for each conflicting fact to obtain the temporal relation. The historical time-series knowledge graph is updated based on a time-series relationship and a confidence-weighted update strategy. The large-scale model joint inference and prediction module is used to initiate risk transmission prediction requests and identify the central entity. Starting from the central entity, it initiates structured queries to the updated historical time-series knowledge graph and generates contextual knowledge subgraphs. The serialization method based on knowledge subgraph-natural language narrative converts the context knowledge subgraph into a natural language description with logical relationships, injects risk hypothetical events into the natural language description, constructs highly structured and formatted output constraint prompts as new inputs to the large model, and outputs a structured JSON object containing a complete inference chain. The risk transmission path generation and early warning module is used to convert JSON objects based on the risk path extraction and visualization algorithm with diminishing impact weight to obtain risk warning information and a visualized transmission path diagram.

Images