The System for Extracting Crime Elements and Predicting Excavation-Type Heritage Crimes Based on Deep Learning Models

Abstract

Heritage crimes can result in the significant loss of cultural relics and predicting them is crucial. To address the issues of inconsistent textual information format and the challenge of preventing and combating heritage crimes, this paper develops a system that extracts crime elements and predict heritage crime occurrences. The system comprises two deep-learning models. The first model, Bi-LSTM + CRF, is constructed to automatically extract crime elements and perform spatio-temporal analysis of crimes based on them. By integrating routine activity theory, social disorder theory, and practical field experience, the research reveals that holidays and other special days (SD) perform a critical role as influential factors in heritage crimes. Building upon these findings, the second model, LSTM + SD, is constructed to predict excavation-type heritage crimes. The results demonstrate that the model with the introduction of the holiday factor improves the RMSE and MAE by 6.4% and 47.8%, respectively, when compared to the original LSTM model. This paper presents research aimed at extracting crime elements and predicting excavation-type heritage crimes. With the ongoing expansion of data volume, the practical significance of the proposed system is poised to escalate. The results of this study are expected to provide decision-making support for heritage protection departments and public security authorities in preventing and combating crimes.