Abstract: The analysis and exploitation of police records data constitute an important part of the information work of public security departments, and the data has the characteristics of unequal temporal and spatial distribution, high redundancy of information records and difficulty in correlating textual content, which leads to the problems of low efficiency of manual analysis and traditional static charts that cannot support intensive analysis in existing information work. To address the above issues, a method is proposed for interactive visual analysis of data features and event patterns. Firstly, the Random-STR R-tree algorithm combining multiple clustering methods is proposed, which divides the data clusters with different density, randomizes the maximum capacity of nodes and constructs the index structure. Secondly, the distribution characteristics and text features of the data are combined to analyze the anomalies in the region using the Jaccard distance and Pearson distance. Thirdly, the weighted network graph is constructed based on the degree of keyword association, and correlation analysis is performed by the community detection algorithm. In addition, a prototype system for visual analysis of police records data, PRDVis, is designed and implemented, which includes view for analyzing spatiotemporal features, view for analyzing frequent information elements and view for analyzing abnormal information elements. Finally, the case study and expert evaluation are conducted through the police records data in a city, and the result demonstrates the usability and effectiveness of the above method.