BiLSTM-CRF Model for Named Entity Recognition in Railway Accident and Fault Analysis Report

Li, Xinqin; Shi, Tianyun; Li, Ping; Yang, Lianbao; Ma, Xiang

doi:10.1145/3321619.3321623

Cited by 5 publications

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

A deep learning approach for robust traffic accident information extraction from online chinese news

Ling,

Ma,

Dong

et al. 2024

IET Intelligent Trans Sys

View full text Add to dashboard Cite

Road traffic accidents are the leading causes of injuries and fatalities. Understanding the traffic accident occurrence pattern and its contributing factors are prerequisites for effective traffic safety management. The paper proposes a deep learning approach for traffic accident recognition and information extraction from online Chinese news to extract and organize traffic accidents automatically. The approach consists of three modules, including automated news collection, news classification, and traffic accident information extraction. The automated news collection module crawls news from online sources, cleans and organizes it into a general news database with different categories of news. The news classification module robustly recognizes the traffic accident news from all types of news by fusing the sentence‐wise and context‐wise semantic news information. The accident information extraction module extracts the key attributes of traffic accidents (e.g. causes, times, locations) from news text using the SoftLexicon‐BiLSTM‐CRF method. The proposed approach is validated by comparing it with state‐of‐the‐art text mining methods using Chinese news data crawled online. The results show that the approach can achieve a high information extraction performance in terms of precision, recall, and F1‐score. It improves the performance of the best benchmark model (BiLSTM‐CRF) by 18.8% in precision and 12.08% in F1‐score. In addition, the potential value of the automatically extracted accident data is illustrated from online news in complementing traditional authority accident data to drive more effective traffic safety management in practice.

show abstract