Multi-class Text Classification and Publication of Crime Data from Online News Sources

Over the past decade, India’s major cities have been deemed as some of the most unsafe places in the world for women. We know that crimes against women occur every day as newspapers show reports of the activities that occur to not just women but also minor girls. As a result of this, women’s safety is a growing concern for the government. The women in our lives deserve to feel secure wherever they go. This paper aims to help women select the states they travel to or relocate to based on recent criminal activity. The proposed methodology is a web application utilizing the Django framework. The web application provides a map with the crime hotspots of India sourced from reputed news articles on the Internet. Articles are first scraped using a web crawler we have designed. The scraped article is classified into a corresponding category to depict them on the heatmap. Classifying news articles is a multi-class classification problem that requires a robust and powerful machine learning model. A novel hybrid architecture of Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) is adopted and proposed in this paper. The ensemble model performs incredibly well on news sources collated from reputed websites and successfully classifies news articles with an accuracy of 95.1%.