A two-stage approach of named-entity recognition for crime analysis

Abstract

Crime against women in India is on increase over the past few years and enormous crime reports are being generated everyday. But it is difficult to manually access the crime reports to derive useful information that can provide insights to the law enforcement officers for analysing the crime trends. The present work emphasizes on a simple yet efficient two stage approach for analysing crime against women in India. Initially, the proposed framework extracts crime reports from online newspaper articles. Once the data is collected, the first stage approach provides an interesting aspect by identifying named entities like name of states, cities, person etc. from the dataset and a collection of top ten entities of various categories is ranked according to their frequency of occurrence. The preliminary assessment shows feasible results which are also compared with crime records drawn from National Crime Records Bureau. However, the identified subtypes of entities are mostly ignored whereas dealing only with the basic entities fails to provide in-depth recognition of crime trends. So considering the subtypes can really provide the prerequisites for finer distinction in the field of crime data mining. The second stage approach in the present work considers the sub-types of named entities as ‘Modus Operandi’ features (mode of operation) of the crime that caters exquisite perception of the crime performed against women in India. Though lot of research exists on crime analysis, considering modus operandi features is very less. The present work demonstrates the effectiveness of the method with high recall and precision for the identified named entities.