Abstract:

With the advances in information and communication technologies, social media and microblogging platforms serve as an important source of information. In microblogging platforms, people can share their opinions, complaints, sentiments and attitudes towards topics, current issues and products. Sentiment analysis is an important research direction in natural language processing, which aims to identify the sentiment orientation of source materials. Twitter is a popular microblogging platform, where people all over the world can interact by user-generated text messages. Information obtained from Twitter can serve as an essential source for several applications, including event detection, news recommendation and crisis management. In sentiment classification, the identification of an appropriate feature subset plays an important role. LIWC (Linguistic Inquiry and Word Count) is an exploratory text analysis software to extract psycholinguistic features from text documents. In this paper, we present a psycholinguistic approach to sentiment analysis on Twitter. In this scheme, we utilized five main LIWC categories (namely, linguistic processes, psychological processes, personal concerns, spoken categories and punctuation) as feature sets. In the experimental analysis, five LIWC categories and their ensemble combinations are taken into consideration. To explore the predictive performance of different feature engineering schemes, four supervised learning algorithms (namely, Naïve Bayes, support vector machines, k-nearest neighbor algorithm and logistic regression) and three ensemble learning methods (namely, AdaBoost, Bagging and Random Subspace) are utilized. The experimental results indicate that ensemble feature sets yield higher predictive performance compared to the individual feature sets. 

Keywords: