知识图谱和表示学习在道路交通事故数据挖掘中的应用

Application of knowledge graph and representation learning in the mining of road traffic accident data

交通安全领域数据量庞大且蕴含丰富的语义信息,从海量道路交通事故数据中挖掘潜在的价值信息可为交通事故预防和智能研判提供支撑。然而,传统的事故分析方法在处理复杂且多语义交叉的事故数据时,存在一定的局限性。研究提出了一种基于知识图谱和知识表示学习的事故数据挖掘方法。通过翻译距离嵌入(Translating Embedding,TransE)模型对道路交通事故知识图谱进行表示学习,将事故实体和致因关系映射到向量空间,并在向量匹配运算中捕捉向量之间的语义信息,进而挖掘潜在的交通事故信息。研究采用真实的事故数据进行试验验证,结果表明所提方法具有较高的准确率和较强的语义解析性能,可为道路交通事故碎片化信息的最大化利用提供新的方法和思路。

The volume of data in traffic safety is substantial,containing rich semantic information.Extracting value from extensive road traffic accident data can provide crucial knowledge support for accident prevention and informed decision-making.However,conventional accident analysis methods face limitations when handling complex and multi-semantic intersecting accident data.Hence,this study proposes a method for mining accident data based on knowledge graph and knowledge representation learning.This method concatenates accident information by leveraging the graph structure and extracts associated information in spatial vectors,thereby enhancing the likelihood of discovering potential value in traffic accident data mining tasks.We establish a knowledge structure schema aligned with the general logic structure of road traffic accident occurrence and then proceed to construct a road traffic accident knowledge graph.In the mining process,we employ the TransE model for representation learning,wherein accident entities and causal relationships are mapped into vector space.Subsequently,semantic information between vectors is captured through vector matching operations to extract potential traffic accident information.In our initial experiments,we compare the model's performance on various datasets and vector dimensions,and we assess our method with real road accident data in different prediction tasks.The experimental results indicate that the proposed method exhibits high accuracy and robust semantic parsing capabilities.Additionally,utilizing vector spaces with appropriate dimensions enhances the accuracy of predictions for traffic-related data.This approach enables effective extraction of potential value from road traffic accident data.In fact,it helps mitigate biases caused by missing accident data in accident analysis and prevention decision-making processes,thereby enhancing the overall completeness of road traffic accident data.Therefore,the research results propose a novel method and idea for maximizing the utilization of fragmented information on road traffic accidents.