Event information enhancement and deduction prediction method combined with knowledge graph

Abstract

The invention relates to an event information enhancement and deduction prediction method combined with a knowledge graph, and belongs to the field of natural language processing and knowledge graphs. The objective of the invention is to solve the problem of insufficient accuracy and robustness of a deduction result due to the fact that multi-dimensional information cannot be effectively fused in an existing event deduction method. In order to solve the problems that a traditional event deduction method mostly depends on a large amount of entity link information and is low in efficiency and prone to being influenced by data scarcity in practical application, the method comprises the five steps of data input and preprocessing, knowledge graph construction and optimization, event semantic background enhancement, event deduction prediction and event deduction prediction. According to the method, the limitation of a traditional method in event deduction can be effectively overcome by introducing technologies such as a cross-domain knowledge graph and a graph neural network, rich event semantic backgrounds are constructed by utilizing the knowledge graph, event information is enhanced and deduced in combination with a time sequence and a causal relationship between events, and the event deduction efficiency is improved. Therefore, the accuracy and reliability of event deduction are improved.

Description

Technical Field

[0001] The present invention belongs to the fields of natural language processing and knowledge graphs, and particularly relates to a method for enhancing and inferring event information in combination with a knowledge graph and making predictions. Background Art

[0002] Event inference and prediction are tasks of inferring future events based on historical data and associated information. The goal is to provide accurate support for decision-making by inferring and predicting current and future possible events on the basis of understanding and analyzing historical events and their interrelationships. Such tasks are widely applied in fields such as financial market analysis and legal case prediction. During the event inference process, the system needs to combine multiple types of information such as time sequence, causal relationship, and space in historical events to infer and predict the subsequent development of events.

[0003] Existing event inference technologies mainly rely on data-driven methods, and use machine learning or deep learning models to learn the laws of events from historical data for predicting future events. However, traditional event inference methods usually rely on static data and lack effective integration of the deep semantic relationships and cross-domain knowledge behind events, resulting in their insufficient performance when facing complex and dynamic event inference tasks. For example, in financial market prediction, although machine learning models can extract certain trends from historical data, due to the lack of understanding of the causal relationships behind market turmoil, it is difficult for the models to accurately predict market events. In addition, most existing event inference systems rely on a single data source when processing event information, lack the ability to comprehensively process multi-source and multi-dimensional information, and are difficult to fully reflect the spatio-temporal changes and causal connections of events. Therefore, how to combine cross-domain knowledge graphs with event inference tasks to enhance the inference ability of event information has become a major challenge in the current technology.

[0004] The present invention proposes a method for enhancing and inferring event information in combination with a knowledge graph. By introducing technologies such as cross-domain knowledge graphs and graph neural networks (GNNs), it can effectively overcome the limitations of traditional methods in event inference. The innovation of the present invention lies in using the knowledge graph to construct a rich semantic background for events, and combining the time sequence and causal relationship between events to enhance and infer event information, thereby improving the accuracy and reliability of event inference. Summary of the Invention

[0005] (I) Technical Problems to be Solved The technical problem to be solved by the present invention is how to provide an event information enhancement and deduction prediction method combined with a knowledge graph to solve the problems that existing event deduction methods usually fail to effectively integrate multi-dimensional information, resulting in insufficient accuracy and robustness of deduction results; traditional event deduction methods mostly rely on a large amount of entity link information, resulting in low efficiency of the system in practical applications and being vulnerable to data scarcity.

[0006] (II) Technical Solution To solve the above technical problems, the present invention proposes an event information enhancement and deduction prediction method combined with a knowledge graph, and the method includes the following steps: Step 1, data input and preprocessing: Extract event information from multi-source heterogeneous data, complete data cleaning and standardization, and unify the time and geographical data formats; Step 2, knowledge graph construction and optimization: Construct a knowledge graph based on the preprocessed data, including event nodes and their time, space, and causal relationship chains, and perform knowledge graph optimization; Step 3, event semantic background enhancement: Based on the knowledge graph, adopt multi-dimensional semantic enhancement technology, generate time features through sine-cosine encoding, embed geographical locations using GeoEncoder, and extract potential causal chains in combination with a causal analysis model based on RoBERTa; subsequently, perform multi-hop semantic propagation using GCN, and enhance the information aggregation ability of key nodes through a self-attention mechanism, thereby enhancing the global consistency of event representation; Step 4, event deduction prediction: Based on the knowledge graph after semantic enhancement, use the graph convolutional neural network GCN to capture the explicit relationships between nodes, combine the graph attention network GAT to assign node weights, and construct an event deduction model; when the model deduces the path, use the heuristic A* algorithm to find the optimal path, and normalize the path weights through the Softmax function to generate the probability distribution of future events; then, train the model; Step 5, dynamic optimization and result output: Real-time update the knowledge graph through consistency detection rules, quickly verify and add new nodes and relationships; use variational Bayesian inference to quantitatively model the uncertainty of events and provide a confidence score for the deduction results; in the output stage, provide the prediction results in JSON format, visually display the deduction path and the knowledge graph structure through D3.js, and generate a detailed event report to support practical scenario applications.

[0007] (III) Beneficial Effects The present invention proposes an event information enhancement and deduction prediction method combined with a knowledge graph. The present invention discloses an event information enhancement and deduction prediction method combined with a knowledge graph, and its advantages are mainly reflected in the following aspects: (1) Multi - dimensional information fusion improves the accuracy of event deduction: By constructing a cross - domain knowledge graph and introducing a multi - dimensional event information enhancement algorithm, the present invention significantly improves the event deduction model's ability to understand multiple information such as time, space, and causality. This enables the system to deduce more accurately in complex event scenarios, enhancing the accuracy and robustness of the deduction.

[0008] (2) Reducing the dependence on entity link information: The present invention proposes an event deduction technology that does not rely on a large amount of entity link information. By combining graph neural networks and knowledge graphs, it effectively reduces the negative impact of entity link errors on event deduction results. This technology reduces the dependence on labeled data, improves the applicability and stability of the model in practical applications, and performs more excellently especially in scenarios with scarce or dynamically changing data.

[0009] (3) Graph neural network optimizes the deduction ability: The present invention uses graph neural networks to model event information, which can deeply mine the complex associations and evolution laws between events, enhancing the reasoning ability and prediction accuracy during event deduction. When dealing with complex event relationships and long - term event predictions, the system demonstrates higher adaptability and stronger deduction accuracy, capable of coping with dynamically changing event environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is the overall framework of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0011] To make the objectives, content, and advantages of the present invention clearer, the following further describes in detail the specific implementation manners of the present invention with reference to the drawings and embodiments.

[0012] The present invention proposes an event information enhancement and deduction prediction method combined with a knowledge graph, mainly solving the following two problems: First, existing event deduction methods usually fail to effectively fuse multi - dimensional information (such as time, space, causal relationships, etc.), resulting in insufficient accuracy and robustness of deduction results. For this reason, the present invention proposes an event information enhancement method based on a knowledge graph. By fusing multi - dimensional information, the semantic background of event information is enhanced, thereby improving the effect of event deduction. Second, traditional event deduction methods rely on a large amount of entity link information, resulting in low efficiency and susceptibility to data scarcity in practical applications. The present invention proposes a deduction technology that does not require a large number of entity links. By introducing an optimized structure of graph neural networks and knowledge graphs, the dependence on entity links is reduced, while the accuracy and training efficiency of the event deduction model are improved.

[0013] The present invention provides an event information enhancement and deduction prediction method combined with a knowledge graph, and the overall framework is as Figure 1As shown in the figure, it mainly includes five steps: data input and preprocessing, knowledge graph construction and optimization, event semantic background enhancement, event deduction and prediction, and event deduction and prediction: Step 1: Data input and preprocessing: Extract event information from multi-source heterogeneous data, and complete data cleaning and standardization, including denoising, filling in missing values, and unifying the time and geographical data formats (ISO 8601 and WGS84).

[0014] Among them, extracting event information from multi-source heterogeneous data includes: using NER for text feature extraction to identify event subjects, objects, and actions, and combining BERT to encode time and causal clues; while for image feature extraction, YOLOv8 is used to detect key objects, and the I3D model for video understanding and action recognition is combined to identify dynamic events, providing high-quality input for subsequent knowledge graph construction.

[0015] Step 2: Knowledge graph construction and optimization: Construct a knowledge graph based on the preprocessed data, including event nodes and their time, space, and causal relationship chains, and perform knowledge graph optimization.

[0016] Among them, the time relationship is generated by timestamp sorting, the space relationship is calculated based on Euclidean distance, and the causal relationship is identified through Granger causality test. During the process of optimizing the constructed knowledge graph, Node2Vec is used to generate node vector representations, learning node semantics through random walks, and combining a pruning algorithm based on node degree to reduce redundancy and improve the efficiency and deduction ability of the knowledge graph.

[0017] Step 3: Event semantic background enhancement: Based on the knowledge graph, multi-dimensional semantic enhancement technology is adopted. Time features are generated through sine-cosine encoding, geographical locations are embedded using GeoEncoder, and potential causal chains are extracted by combining a causal analysis model based on RoBERTa. Subsequently, GCN is used for multi-hop semantic propagation, and the information aggregation ability of key nodes is improved through self-attention mechanism, thereby enhancing the global consistency of event representation.

[0018] Step 4: Event deduction and prediction: Based on the knowledge graph after semantic enhancement, use the Graph Convolutional Network (GCN) to capture explicit node relationships, combine the Graph Attention Network (GAT) to assign node weights, and construct an event deduction model. When the model deduces the path, the heuristic A* algorithm is used to find the optimal path, and the path weights are normalized through the Softmax function to generate the probability distribution of future events. Then, the model is trained, and during the training phase, the prediction results are optimized through the cross-entropy loss function to improve the deduction accuracy and robustness.

[0019] Step 5: Dynamic Optimization and Result Output: The knowledge graph is updated in real time through consistency detection rules to quickly verify and add new nodes and relationships. Variational Bayesian inference is used to quantitatively model the uncertainty of events and provide a confidence score for the deduction results. In the output stage, the system provides the prediction results in JSON format, visually displays the deduction path and the knowledge graph structure through D3.js, and generates a detailed event report to support practical scenario applications. Embodiment

[0020] The following is a detailed introduction to the five steps respectively: Step 1: Data Input and Preprocessing: The system extracts event-related information from multi-source heterogeneous data to build the basis for the subsequent knowledge graph.

[0021] In the data cleaning stage, data cleaning rules are adopted, including removing redundant items, filling in missing values, and correcting inconsistent data formats. The time data format is unified into the ISO 8601 standard format, and the geographical location information is represented in the WGS84 coordinate system.

[0022] In terms of event feature extraction, a Named Entity Recognition (NER) model is used to identify the subject, object, and action relationships of events. At the same time, a pre-trained BERT model is used to perform context feature encoding on the text description of events to extract time and causal clues. For image and video data, the YOLOv8 object detection model is used to extract key object features, and further combined with the I3D model of the action recognition algorithm to obtain dynamic event features. These processed data provide high-quality standardized input for the subsequent construction of the knowledge graph.

[0023] Step 2: Knowledge Graph Construction and Optimization: After completing data preprocessing, the system constructs a knowledge graph based on the preprocessed data and optimizes its structure. The nodes of the knowledge graph include the subject, object of the event, and their associated attributes (time, location, causal relationship, etc.). The types of relationships between nodes are determined by a predefined set of time relationships, spatial relationships, and causal relationships. The time relationship constructs a time sequence relationship chain through the sorting of event timestamps; the spatial relationship calculates the Euclidean distance between nodes based on geographical coordinates and sets a threshold for spatial association; the causal relationship uses the Granger causality test method for event pairs to identify the trigger and causal chain between events. In the process of graph optimization, the Node2Vec embedding algorithm is used to generate vector representations of nodes, and the random walk and Skip-Gram models are used to learn node semantic representations, thereby enhancing the computational efficiency and retrieval ability of the knowledge graph. For sparse optimization, a pruning algorithm based on node degree is used to delete low-degree nodes and their edges to reduce redundancy and improve the deduction efficiency.

[0024] Step 3: Event Semantic Background Enhancement: Based on the knowledge graph, the system enhances the background semantic representation of event nodes through multi-dimensional semantic enhancement methods. In the time dimension, the periodic encoding technique (such as sine-cosine encoding) is used to characterize the timestamps, generating the temporal context features of events; in the space dimension, GeoEncoder is used to embed the geographical location information; in the causal dimension, a pre-trained causal relationship extraction model (such as the causal analysis model based on RoBERTa) is used to identify potential causal chains and supplement them to the node attributes. Subsequently, the Graph Convolutional Network (GCN) is used to perform semantic propagation on the knowledge graph, capturing implicit semantic associations between distant nodes through multi-layer message passing. In addition, the graph neural network structure adopts the Self-Attention mechanism to improve the information aggregation ability of high-importance nodes, thereby enhancing the global semantic consistency of events.

[0025] Step 4: Event Deduction and Prediction: On the semantically enhanced knowledge graph, the system predicts the likelihood and paths of future events through an event deduction model. The deduction model is based on a multi-layer graph neural network architecture. First, GCN is used to capture the explicit relationships between nodes, and then the Graph Attention Network (GAT) is further used to assign weights to different nodes to highlight key event nodes. The search for the deduction path of the model adopts the heuristic A* search algorithm, with the shortest path as the optimization goal, and combines the relationship weights to find the most likely deduction path between events. To calculate the probability of future events occurring, the model normalizes the cumulative weights of each path through the Softmax function to generate a prediction distribution. During the model training process, the cross-entropy loss function is used to optimize by comparing the true event chain labels with the prediction results, improving the accuracy and robustness of the deduction.

[0026] Step 5: Dynamic Optimization and Result Output: To adapt to the update of dynamic events, the system designs a real-time knowledge graph update mechanism. When new event data enters, the system automatically verifies the relationship between the new nodes and the existing graph through consistency detection rules, and quickly updates the graph structure through the new node module. At the same time, based on the variational Bayesian inference method, the uncertainty in the event data is quantitatively modeled to provide a confidence score for the deduction results, facilitating users to understand the uncertainty of the prediction. In terms of result output, the predicted events and their possible paths are output in JSON format, and a visualization tool based on D3.js is provided to support the interactive visualization of the knowledge graph structure and the deduction path. In addition, the system generates a detailed event report, including the probability of the event occurring, key influencing factors, and deduction basis, facilitating application to actual scenarios.

[0027] The present invention discloses an event information enhancement and deduction prediction method combined with a knowledge graph, and its advantages are mainly reflected in the following aspects: (1) Multi-dimensional information fusion improves the accuracy of event deduction: By constructing a cross-domain knowledge graph and introducing a multi-dimensional event information enhancement algorithm, the present invention significantly improves the understanding ability of the event deduction model for multiple information such as time-space and causality, enabling the system to deduce more accurately in complex event scenarios and improving the accuracy and robustness of deduction.

[0028] (2) Reducing the dependence on entity link information: The present invention proposes an event deduction technology that does not rely on a large amount of entity link information. By combining graph neural networks with a knowledge graph, it effectively reduces the negative impact of entity link errors on the event deduction results. This technology reduces the dependence on labeled data and improves the applicability and stability of the model in practical applications, especially showing more superiority in scenarios with scarce or dynamically changing data.

[0029] (3) Graph neural network optimizes the deduction ability: The present invention uses a graph neural network to model event information, which can deeply mine the complex associations and evolution laws between events, and improve the reasoning ability and prediction accuracy in the event deduction process. When dealing with complex event relationships and long-term event predictions, the system shows higher adaptability and stronger deduction accuracy, and can cope with the dynamically changing event environment.

[0030] The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. An event information enhancement and deduction prediction method combined with a knowledge graph, characterized in that, The method includes the following steps: Step 1, Data Input and Preprocessing: Extract event information from multi-source heterogeneous data, complete data cleaning and standardization, and unify the time and geographical data formats; Step 2, Knowledge Graph Construction and Optimization: Construct a knowledge graph based on the preprocessed data, including event nodes and their time, space, and causal relationship chains, and perform knowledge graph optimization; Step 3, Event Semantic Background Enhancement: Based on the knowledge graph, adopt multi-dimensional semantic enhancement techniques, generate time features through sine-cosine encoding, embed geographical locations using GeoEncoder, and extract potential causal chains by combining a causal analysis model based on RoBERTa; Subsequently, use GCN for multi-hop semantic propagation and enhance the information aggregation ability of key nodes through self-attention mechanism, thereby enhancing the global consistency of event representation; Step 4, Event Deduction and Prediction: Based on the knowledge graph after semantic enhancement, use the graph convolutional neural network GCN to capture explicit relationships between nodes, combine the graph attention network GAT to assign node weights, and construct an event deduction model; When the model deduces the path, use the heuristic A* algorithm to find the optimal path, and normalize the path weights through the Softmax function to generate the probability distribution of future events; Then, train the model; Step 5, Dynamic Optimization and Result Output: Real-time update the knowledge graph through consistency detection rules, quickly verify and add new nodes and relationships; Use variational Bayesian inference to quantitatively model the uncertainty of events and provide a confidence score for the deduction results; In the output stage, provide the prediction results in JSON format, visually display the deduction path and the knowledge graph structure through D3.js, and generate a detailed event report to support practical scenario applications.

2. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 1, wherein In the first step, the data cleaning stage adopts data cleaning rules, including removing redundant items, filling in missing values, and correcting inconsistent data formats; The time data format is unified into the ISO 8601 standard format, and the geographical location information is represented using the WGS84 coordinate system.

3. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 1, characterized in that In the first step, extracting event information from multi-source heterogeneous data includes: Using NER to identify event subjects, objects, and actions for text feature extraction, and combining BERT to encode time and causal clues; For image feature extraction, detect key objects through YOLOv8 and identify dynamic events by combining the I3D model for video understanding and action recognition, providing high-quality input for subsequent knowledge graph construction.

4. The method for enhancing event information and inferring and predicting in combination with a knowledge graph according to claim 1, wherein In the second step, the types of relationships between nodes are determined by a predefined set of time relationships, space relationships, and causal relationships; The time relationship constructs a time sequence relationship chain through event timestamp sorting; The space relationship calculates the Euclidean distance between nodes based on geographical coordinates and sets a threshold for spatial association; The causal relationship uses the Granger causality test method for event pairs to identify trigger and causal chains between events.

5. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 4, wherein In the process of graph optimization, use the Node2Vec embedding algorithm to generate vector representations of nodes, learn node semantic representations through random walks and the Skip-Gram model, and enhance the computational efficiency and retrieval ability of the knowledge graph.

6. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 4, wherein For sparse optimization, a pruning algorithm based on node degree is used to delete low-degree nodes and their edges to reduce redundancy.

7. The method for enhancing event information and inferring prediction by combining a knowledge graph according to claim 1, wherein In step 3, the time dimension uses the periodic sine-cosine encoding technique to characterize timestamps and generate the time context features of events. The space dimension uses GeoEncoder to embed and represent geographical location information; the causal dimension identifies potential causal chains through a RoBERTa-based causal analysis model and supplements them to node attributes; subsequently, the graph convolutional network GCN is used for semantic propagation of the knowledge graph, and implicit semantic associations between distant nodes are captured through multi-layer message passing.

8. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 1, wherein In step 4, the search for the deduction path uses the heuristic A* search algorithm, with the shortest path as the optimization goal, and combines the relationship weights to find the most likely deduction path between events; to calculate the probability of future events occurring, the cumulative weights of each path are normalized through the Softmax function to generate the prediction distribution.

9. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 8, wherein In step 4, during the training phase, the prediction results are optimized through the cross-entropy loss function to improve the deduction accuracy and robustness.

10. The method for event information enhancement and deduction prediction combined with a knowledge graph according to claim 1, wherein In step 5, the predicted events and their possible paths are output in JSON format, and a visualization tool based on D3.js is provided to support the interactive visualization of the knowledge graph structure and deduction paths; in addition, the system generates a detailed event report, including the probability of event occurrence, key influencing factors, and deduction basis, which is convenient for application to actual scenarios.

Images