A Research on Event Extraction for illegal Cultivated land use Case Texts

Abstract. Cultivated land is fundamental to human survival and development. China attaches great importance to the utilization and protection of cultivated land. In the field of administrative law enforcement, the Ministry of Natural Resources, as the competent authority, has regularly released text of cases involving illegal occupation of cultivated land for a long time. Such unstructured data is of low value density and limited usability. In this paper, the structured information such as location, time, event type, offence and occupied area can be quickly and efficiently extracted from the administrative cases through event extraction technology, so as to promote the use of historical case information and the study of intelligent law enforcement and case handling for administrative agencies. Firstly, the real-case data is collected from the official website of the Ministry of Natural Resources using a thematic web crawler. Subsequently, it constructs an entity argument recognition framework, defining the types of illegal cultivated land use events and their argument compositions. A training dataset is built for illegal cultivated land use cases through manual annotation. Finally, two sets of models based on statistical-based machine learning and deep learning are employed to optimize the extraction results of event arguments. Experiments show that the F1 score for the event argument extraction reaches 90.81% with BiLSTM-CRF model. The research in this paper achieves effective extraction of case information, verifies the effectiveness of existing models in event extraction from administrative law enforcement case texts, and provides effective support for deeper research in this field in the future.