An Improved Feature Selection Method for Sentiments Analysis in Social Networks

The increasing growth of user oriented media and users' preferences in using social networks for communication, causes these virtual communities to become a valuable source of data. These communities provide users with the possibility of being aware of useful and reliable opinions. Many organizations employ efficient classifiers for determining the polarity of users' opinions in order to make valid decisions in different business domains. However, most of the existing approaches suffer from low accuracy results due to performing classification task in a high-dimensional feature space. To this end, an efficient feature selection method based on a modified version of firefly algorithm is presented in this article. The main contribution of the proposed method is employing a weighted combination of classification performance measures in constructing a fitness function for the firefly algorithm. The proposed model for the fitness function leads to establishing a trade-off between the performance measures while trying to reduce the number of dimensions. The results obtained from experiments conducted on 11000 tweets show that the proposed method outperforms the existing counterparts in terms of polarity classification performance.