Using logistic regression method to classify tweets into the selected topics

Abstract

Topics about health, music, sport, and technology are widely discussed in social network sites, especially in Twitter. Sharing information about those topics can enrich one's knowledge as well as increase the awareness of the current trends pertinent to the area of interests. Hence, this research aims to develop a web-based application that can classify tweets of netizens into these four categories of topics using one of machine learning methods called Logistic Regression. There are four main processes applied in this application that are fetching tweets, preprocessing, text feature extraction and machine learning. There are 1800 labeled tweets for each topic used as training data. Several processes were done in the pre-processing phase, including removal of URLs, punctuation, and stop words, tokenization, and stemming. Later, the application automatically converted the pre-processed tweets into set of features vector using Bag of Words. The set of features vector was applied to the Logistic Regression algorithm for the classification task. The trained classifier was then evaluated using 1800 tweets with 450 for each topic. Using Confusion Matrix, the results showed the accuracy of tweets classification into the selected topics is 92% which is considered very high.