Abstract:

This paper investigates with the automated classification of tweets which turns out to be a very complicated problem because of its nature, heterogeneity and the amount of data. According to internet live stats, nearly 500 million tweets are tweeted per day, where the user’s opinion about different topics is shared. An automated decision support system is developed to analyze the tweets related to crime against women and children. The problem is viewed in a big data perspective because of the nature of data. The proposed work focuses on developing two systems: Hadoop MapReduce and Apache Spark framework for programming with Big Data. The algorithm based on hierarchical domain lexicon classifies different types of crime in a parallel and distributed manner. Moreover, the crime classification tool is based on hybridized Machine Learning techniques combined with Natural Language Processing techniques. To predict the location of twitter users, multinomial Naive Bayes classifier trained on Location Indicative terms and other vital parameters (such as city/country names, #hash tags and @mentions) is implemented. Our approach outperforms in terms of classification accuracy, mean and median error distance when compared with other algorithms based on parameters such as Location Indicative terms, #hash tags and city/country names.

Keywords: