Event influence path analysis method based on multi-modal data fusion

Abstract

The invention discloses an event influence path analysis method based on multi-modal data fusion, and relates to the technical field of artificial intelligence application. Comprising the following steps: 1, converting a natural language problem into a semantic vector by utilizing a domain embedding model, efficiently recalling related evidence fragments in a multi-source heterogeneous document library, and performing semantic retrieval in document libraries of agricultural meteorological reports, agricultural product market daily reports, supply chain logistics data, historical disaster influence analysis and statistical departments, 2, based on an economic knowledge map, entity mapping and multi-hop causal reasoning are carried out, deep association between entities is mined, and the entity mapping and multi-hop causal reasoning are carried out; wherein the identified related entities comprise extreme meteorological events, crops and market prices, mapping the entities to corresponding nodes in an agricultural knowledge graph, starting multi-hop reasoning, traversing causal and association relationships in the graph, and forming a structured influence path; and 3, integrating the initially retrieved text segments, the influence paths and historical data, and generating an analysis result with a clear structure and sufficient basis by using a large language model.

Description

Technical Field

[0001] This invention discloses an event impact path analysis method based on multimodal data fusion, which relates to the field of artificial intelligence application technology. Background Technology

[0002] The current mainstream search engines and agricultural databases have significant limitations: First, information is overloaded and fragmented, with systems returning a large amount of isolated information such as weather warnings, vegetable wholesale prices, and traffic control, requiring a significant amount of time to manually piece together causal chains for analysis; second, they lack multi-hop reasoning capabilities, such as the inability to automatically complete deep logical deductions like "typhoon → disaster in major producing areas → reduced production → transportation disruption → regional supply shortage → nationwide price transmission," and are particularly unable to identify the indirect impacts of non-directly affected areas; third, qualitative and quantitative information are separated, for example, key data such as the proportion of production reduction caused by historical typhoons, the number of days of logistics delays, and price increases are scattered across different sources, making it difficult to unify them into a model and use for quantitative prediction. Summary of the Invention

[0003] This invention addresses the problems of existing technologies by providing an event impact path analysis method based on multimodal data fusion. Combining semantic retrieval, agricultural knowledge graphs, and large-scale model generation capabilities, it accurately identifies key entities and performs structured reasoning along the path of "event—agricultural production—supply chain disruption—market supply and demand—price fluctuation." By integrating historical data and real-time information, it ultimately generates a clear and well-supported analysis report. This not only significantly improves the efficiency and accuracy of agricultural product price forecasting but also provides strong support for decision-making and supply chain management.

[0004] The specific solution proposed in this invention is as follows: This invention provides a method for event impact path analysis based on multimodal data fusion, comprising: Step 1: Utilize a domain embedding model to transform natural language questions into semantic vectors, efficiently retrieving relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries containing agricultural meteorological reports, daily agricultural product market data, supply chain logistics data, historical disaster impact analyses, and statistical department documents, returning text fragments containing information related to typhoon impacts on crops, distribution of major crop-producing areas, disruption of crop transportation, and crop price fluctuations. Step 2: Based on the economic knowledge graph, conduct entity mapping and multi-hop causal reasoning to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. Map the entities to the corresponding nodes in the agricultural knowledge graph, initiate multi-hop reasoning, traverse the causal and relational relationships in the graph, and form a structured influence path. Step 3: Integrate the initially retrieved text fragments, influence paths, and historical data, and use a large language model to generate well-structured and well-supported analysis results.

[0005] Furthermore, in step 2 of the event impact path analysis method based on multimodal data fusion, when traversing the causal and correlation relationships in the graph, starting from the extreme weather event node, along the impact relationship, the area where extreme weather events often occur is located. In these areas, crop nodes are searched to identify whether they contain the crop to be queried. If they do, a yield loss path is generated. At the same time, the impact path of transportation infrastructure is generated by extending from the extreme weather event node.

[0006] Furthermore, in step 2 of the event impact path analysis method based on multimodal data fusion, when traversing the causal and correlation relationships in the graph, if a traffic infrastructure impact path is generated, and if other major crop producing areas are not affected, then the capacity substitution capability of other major producing areas is assessed: whether supply can be increased to alleviate the gap; and each sub-path is integrated into a complete impact path sub-graph.

[0007] Furthermore, step 3 of the event impact path analysis method based on multimodal data fusion includes: The initially retrieved text fragments, the structured influence paths derived from the graph, and historical data are fused to form evidence. All evidence is then organized into a structured context and submitted to the large language model. The analysis report is generated based on evidence using a large language model, including: a summary of the impact path, quantitative supporting data, analysis of mitigation factors, and uncertainty alerts.

[0008] This invention also provides an event impact path analysis system based on multimodal data fusion, including a retrieval and management module, a mapping path generation module, and an analysis module. The retrieval management module utilizes a domain embedding model to transform natural language questions into semantic vectors, efficiently retrieving relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries containing agricultural meteorological reports, daily agricultural product market reports, supply chain logistics data, historical disaster impact analyses, and statistical department documents, returning text fragments containing information related to typhoon impacts on crops, the distribution of major crop-producing areas, crop transportation disruptions, and crop price fluctuations. The mapping path generation module relies on the economic knowledge graph to carry out entity mapping and multi-hop causal reasoning, and to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. The entities are mapped to the corresponding nodes in the agricultural knowledge graph, and multi-hop reasoning is initiated to traverse the causal and relational relationships in the graph to form a structured influence path. The analysis module integrates the initially retrieved text fragments, influence paths, and historical data, and uses a large language model to generate well-structured and well-supported analysis results.

[0009] Furthermore, when the mapping path generation module of the event impact path analysis system based on multimodal data fusion traverses the causal and correlation relationships in the graph, it starts from the extreme weather event node, follows the impact relationship, locates the area where extreme weather events often occur, searches for crop nodes in these areas, identifies whether they contain the crop to be queried, and if they do, generates a yield loss path. At the same time, it extends from the extreme weather event node to generate a transportation infrastructure impact path.

[0010] Furthermore, when the mapping path generation module of the event impact path analysis system based on multimodal data fusion traverses the causal and correlation relationships in the graph, if a traffic infrastructure impact path is generated, and if other major crop producing areas are not affected, the module assesses the capacity substitution capability of other major producing areas: whether supply can be increased to alleviate the gap; and integrates each sub-path into a complete impact path sub-graph.

[0011] Furthermore, the analysis module of the event impact path analysis system based on multimodal data fusion integrates the initially retrieved text fragments, the structured impact paths inferred from the graph, and historical data to form evidence. All evidence is then organized into a structured context and submitted to the large language model. The analysis report is generated based on evidence using a large language model, including: a summary of the impact path, quantitative supporting data, analysis of mitigation factors, and uncertainty alerts.

[0012] The advantages of this invention are: Deep reasoning and path discovery: Through multi-hop traversal of the knowledge graph, hidden influence transmission paths are automatically discovered and revealed, far exceeding simple retrieval.

[0013] The analysis process is interpretable: the generated structured report not only contains conclusions, but also clearly shows the data sources, reasoning logic, and chain of evidence on which the analysis was based, greatly enhancing the credibility of the conclusions.

[0014] Increased efficiency: The time required for analysts to conduct literature reviews, data correlation, and report writing, which used to take hours or even days, can be reduced to minutes, greatly improving the efficiency of data analysis. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0016] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention. Example

[0017] This invention provides a method for event impact path analysis based on multimodal data fusion, comprising: Step 1: Use a domain embedding model to transform natural language questions into semantic vectors, and efficiently retrieve relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries such as agricultural meteorological reports, daily agricultural product market reports, supply chain logistics data, historical disaster impact analysis, and statistical departments to return text fragments containing information related to typhoon impacts on crops, distribution of major crop production areas, interruption of crop transportation, and fluctuations in crop prices.

[0018] A professional embedding model trained on corpus can be used to convert the natural language problem into a domain-adaptive semantic query vector; the semantic query vector can be compared with a preset document vector library to retrieve the Top-K most relevant text fragments.

[0019] Step 2: Based on the economic knowledge graph, conduct entity mapping and multi-hop causal reasoning to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. Map the entities to the corresponding nodes in the agricultural knowledge graph, initiate multi-hop reasoning, traverse the causal and relational relationships in the graph, and form a structured influence path.

[0020] When traversing the causal and correlation relationships in the graph, starting from the extreme weather event node, the system locates the areas where extreme weather events frequently occur along the influence relationship. In these areas, it searches for crop nodes and identifies whether they contain the crop to be queried. If they do, it generates a path for yield loss. At the same time, it extends from the extreme weather event node to generate a path for the impact on transportation infrastructure.

[0021] If a transportation infrastructure impact path is generated, and other major crop-producing areas are not affected, then assess the capacity substitution capabilities of other major producing areas: whether they can increase supply to alleviate the gap; and integrate each sub-path into a complete impact path sub-graph.

[0022] For example, key entities can be identified and disambiguated: typhoons are extreme weather events, chili peppers are agricultural products, and national prices are market prices, and then mapped to the corresponding nodes in the agricultural knowledge graph.

[0023] Initiate multi-hop reasoning to traverse the causal and relational relationships in the graph: Starting from the "typhoon" node, locate its frequently landed areas such as coastal provinces like A, B, and C along the impact relationships; in these areas, search for major crop nodes and identify whether they contain the "chili pepper" node; if it is confirmed that the typhoon landed area is a major chili pepper producing area, such as C, which supplies chili peppers to the whole country in winter, then trigger the yield loss path: typhoon → strong winds and heavy rain → farmland flooding / crop lodging → chili pepper yield reduction; at the same time, extend from the "typhoon" node to the impact path on transportation infrastructure: typhoon → port closure, highway interruption, air transport suspension → logistics obstruction → delay in agricultural product transportation.

[0024] Further reasoning: Disruptions in chili transportation → regional supply shortages → rising wholesale market prices → transmission to the national market through inter-regional transportation; If other major producing areas such as D and F are not affected, assess their capacity substitution capabilities: can they increase supply to alleviate the gap; Finally, consider market expectations and speculative behavior: if the market expects tight supply, it may trigger hoarding, further pushing up prices.

[0025] Finally, the above multiple paths are integrated into a complete "influence path sub-map", covering meteorological, agricultural, logistics, market and behavioral factors.

[0026] Step 3: Integrate the initially retrieved text fragments, influence paths, and historical data, and use a large language model to generate well-structured and well-supported analysis results.

[0027] The specific process is as follows: the initially retrieved text fragments, the structured influence paths derived from the graph, and historical data are fused to form evidence. All evidence is then organized into a structured context and submitted to the large language model. The analysis report is generated based on evidence using a large language model, including: a summary of the impact path, quantitative supporting data, analysis of mitigation factors, and uncertainty alerts.

[0028] For example, the initially retrieved text fragments, such as recent typhoon warnings, statistics on planting area in major producing areas, and structured paths derived from graph inference, are integrated with historical data, such as the average price increase of chili peppers during typhoon seasons over the past five years and the proportion of chili peppers from region C in the national winter supply. All evidence is organized into a structured context and submitted to the large language model.

[0029] The analysis report generated by the large model will include: a summary of the impact path: typhoon → disaster in major producing areas → reduced output + transportation disruption → short-term supply contraction → increase in wholesale and retail prices; quantitative support: citing historical data, such as "Typhoon Haikui in 2023 caused a 30% reduction in chili pepper production in region C, and the national wholesale price rose by 18% within a week"; analysis of mitigation factors: pointing out that if non-major producing areas have sufficient inventory or substitutes such as dried chili peppers or imported chili peppers can quickly fill the gap, the price increase may be limited; uncertainty warning: emphasizing that the degree of impact depends on the intensity and duration of the typhoon's landfall, whether it coincides with the harvest season, and whether government departments activate emergency supply guarantee mechanisms. Example

[0030] This invention also provides an event impact path analysis system based on multimodal data fusion, including a retrieval and management module, a mapping path generation module, and an analysis module. The retrieval management module utilizes a domain embedding model to transform natural language questions into semantic vectors, efficiently retrieving relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries containing agricultural meteorological reports, daily agricultural product market reports, supply chain logistics data, historical disaster impact analyses, and statistical department documents, returning text fragments containing information related to typhoon impacts on crops, the distribution of major crop-producing areas, crop transportation disruptions, and crop price fluctuations. The mapping path generation module relies on the economic knowledge graph to carry out entity mapping and multi-hop causal reasoning, and to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. The entities are mapped to the corresponding nodes in the agricultural knowledge graph, and multi-hop reasoning is initiated to traverse the causal and relational relationships in the graph to form a structured influence path. The analysis module integrates the initially retrieved text fragments, influence paths, and historical data, and uses a large language model to generate well-structured and well-supported analysis results.

[0031] The information interaction and execution process between the modules in the above system are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description in the method embodiment of the present invention, and will not be repeated here.

[0032] Similarly, the advantages of the system of the present invention are: Deep reasoning and path discovery: Through multi-hop traversal of the knowledge graph, hidden influence transmission paths are automatically discovered and revealed, far exceeding simple retrieval.

[0033] The analysis process is interpretable: the generated structured report not only contains conclusions, but also clearly shows the data sources, reasoning logic, and chain of evidence on which the analysis was based, greatly enhancing the credibility of the conclusions.

[0034] Increased efficiency: The time required for analysts to conduct literature reviews, data correlation, and report writing, which used to take hours or even days, can be reduced to minutes, greatly improving the efficiency of data analysis.

[0035] It should be noted that not all steps and modules in the above processes and system structures are mandatory; some steps or modules can be omitted as needed. The execution order of each step is not fixed and can be adjusted as required. The system structures described in the above embodiments can be physical or logical structures. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0036] The embodiments described above are merely preferred embodiments for fully illustrating the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. An event impact path analysis method based on multimodal data fusion, characterized by: include: Step 1: Utilize a domain embedding model to transform natural language questions into semantic vectors, efficiently retrieving relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries containing agricultural meteorological reports, daily agricultural product market data, supply chain logistics data, historical disaster impact analyses, and statistical department documents, returning text fragments containing information related to typhoon impacts on crops, distribution of major crop-producing areas, disruption of crop transportation, and crop price fluctuations. Step 2: Based on the economic knowledge graph, conduct entity mapping and multi-hop causal reasoning to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. Map the entities to the corresponding nodes in the agricultural knowledge graph, initiate multi-hop reasoning, traverse the causal and relational relationships in the graph, and form a structured influence path. Step 3: Integrate the initially retrieved text fragments, influence paths, and historical data, and use a large language model to generate well-structured and well-supported analysis results.

2. The event impact path analysis method based on multimodal data fusion according to claim 1, characterized in that: In step 2, when traversing the causal and correlation relationships in the graph, starting from the extreme weather event node, along the influence relationship, the area where extreme weather events often occur is located. In these areas, crop nodes are searched to identify whether they contain the crop to be queried. If they do, a yield loss path is generated. At the same time, the impact path of transportation infrastructure is generated by extending from the extreme weather event node.

3. The event impact path analysis method based on multimodal data fusion according to claim 1, characterized in that: In step 2, when traversing the causal and relational relationships in the graph, if a path affecting transportation infrastructure is generated, and if other major crop-producing areas are not affected, the capacity substitution capability of other major producing areas is assessed: whether supply can be increased to alleviate the gap; and each sub-path is integrated into a complete sub-graph of impact paths.

4. The event impact path analysis method based on multimodal data fusion according to claim 1, characterized in that: Step 3 includes: The initially retrieved text fragments, the structured influence paths derived from the graph, and historical data are fused to form evidence. All evidence is then organized into a structured context and submitted to the large language model. The analysis report is generated based on evidence using a large language model, including: a summary of the impact path, quantitative supporting data, analysis of mitigation factors, and uncertainty alerts.

5. An event impact path analysis system based on multimodal data fusion, characterized by: It includes a retrieval management module, a mapping path generation module, and an analysis module. The retrieval management module utilizes a domain embedding model to transform natural language questions into semantic vectors, efficiently retrieving relevant evidence fragments from multi-source heterogeneous document libraries. Semantic retrieval is performed in document libraries containing agricultural meteorological reports, daily agricultural product market reports, supply chain logistics data, historical disaster impact analyses, and statistical department documents, returning text fragments containing information related to typhoon impacts on crops, the distribution of major crop-producing areas, crop transportation disruptions, and crop price fluctuations. The mapping path generation module relies on the economic knowledge graph to carry out entity mapping and multi-hop causal reasoning, and to explore the deep connections between entities. The relevant entities identified include extreme weather events, crops, and market prices. The entities are mapped to the corresponding nodes in the agricultural knowledge graph, and multi-hop reasoning is initiated to traverse the causal and relational relationships in the graph to form a structured influence path. The analysis module integrates the initially retrieved text fragments, influence paths, and historical data, and uses a large language model to generate well-structured and well-supported analysis results.

6. The event impact path analysis system based on multimodal data fusion according to claim 5, characterized in that it maps... When the path generation module traverses the causal and relational relationships in the graph, it starts from the extreme weather event node, follows the influence relationship, locates the area where extreme weather events often occur, searches for crop nodes in these areas, identifies whether they contain the crop to be queried, and if they do, generates a path for yield loss. At the same time, it extends from the extreme weather event node to generate a path for the impact of transportation infrastructure.

7. The event impact path analysis system based on multimodal data fusion according to claim 5, characterized in that it maps... When the path generation module traverses the causal and relational relationships in the graph, if a path affecting transportation infrastructure is generated, and if other major agricultural production areas are not affected, the module assesses the capacity substitution capability of other major production areas: whether supply can be increased to alleviate the gap; and integrates each sub-path into a complete impact path sub-graph.

8. The event impact path analysis system based on multimodal data fusion according to claim 5, characterized in that: The analysis module integrates the initially retrieved text fragments, the structured influence paths derived from the graph, and historical data to form evidence. It then organizes all the evidence into a structured context and submits it to the large language model. The analysis report is generated based on evidence using a large language model, including: a summary of the impact path, quantitative supporting data, analysis of mitigation factors, and uncertainty alerts.

Images