Rough set based incremental crime report labelling in dynamic environment

摘要

The proposed work describes a rough set based incremental crime report labelling technique. The named entities are recognized from the available crime reports to identify the phrases existing between entity pairs. The phrases are vectorized considering intervening context words and a quick reduct generation algorithm is applied to minimize their dimensions. A graph based clustering algorithm has been applied to cluster the phrases and later they have been labelled based on some centrality measure techniques. Each current report, represented by a set of phrases, is labelled by the labels of clusters in which its phrases reside. Thus, a current report may be labelled as a multi-class report. Each new report is labelled incrementally using existing labelled clusters with the help of rough set theory. The phrases of the new report are partitioned into equivalence classes using indiscernibility relation and the lower approximations of the clusters of phrases are obtained considering these equivalence classes as target sets. If the lower approximation of a cluster contains all the phrases of the new report, then the report is labelled by the label of that cluster; otherwise, probability based membership values have been assigned to the report to represent the degree of its belongingness into different classes. The existing clusters and equivalence classes of phrases generate a new set of labelled clusters, which are further used for labelling upcoming reports. Thus, the proposed method labels crime reports in an incremental way. The work is validated by various indices and compared with many state-of-the-art clustering and classification algorithms. Experimental results show the effectiveness of the method in the crime report labelling. (C) 2019 Elsevier B.V. All rights reserved.