面向警情数据的交互式可视分析方法

Interactive Visual Analysis of Police Records Data

警情数据的分析和挖掘是公安部门信息工作的核心环节, 数据具有时空分布不均匀、信息记录高冗余和文 本内容难关联等特点, 导致现有信息工作中存在人工分析效率低、传统静态图表无法支持深入分析等问题. 为此, 提 出一种交互式可视分析数据特征和事件模式的方法. 首先提出一种结合多种聚类方法的 Random-STR R 树算法, 划 分出密度不同的数据簇, 随机化结点最大容纳量并构建索引结构;然后结合数据分布特征和文本特征, 利用杰卡德 距离和皮尔逊距离分析区域内的异常情况;最后基于关键词的关联程度构建加权网络图, 利用社区检测算法对其进 行关联分析. 此外, 设计并实现警情数据可视分析原型系统--PRDVis, 包含时空特征分析视图、频繁信息要素分析 视图和异常信息要素分析视图. 通过某城市警情数据进行案例分析和专家评估, 结果证明了所提方法的可用性和有 效性.

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 dis-tribution,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 vis-ual analysis of data features and event patterns.Firstly,the Random-STR R-tree algorithm combining multi-ple 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 asso-ciation,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.