Emotions Polarity of Tweets Based on Semantic Similarity and User Behavior Features

In social networks, people share their emotions and opinion towards specific subjects in the form of comments. These comments are positive, negative, or neutral based on their content. Therefore, people try to pick out the comment's content; this content reflects the behaviour and style of expressing their emotions. In recent years, researchers interested in sentiment analysis (SA) achieved results using different methods, the lexical features and semantic similarity. But most of them neglect parts of the content of comments, considering it has no importance such as URLs, Numbers and Marks. They use the value of semantic similarity between comments direct to identify emotions polarity or to find synonyms of lexical features. This paper introduces an implementation that uses a combination of user behaviour, semantic and lexical features together for finding polarity emotions of Tweets. This proposed method examines some of the neglected content of tweets as features and adapts the semantic similarity value to emotional polarity as a feature in addition to lexical features. The main objective of this paper is to improve the classification accuracy. The performance of the proposed method evaluated using Naive Bayes and SVM as two popular machine learning classifiers. The best-obtained result is 94% using Naive Bayes and Sentiment140 dataset.