Tweet Sentiment Polarity Detection Based on Semantic Similarity

Abstract

Nowadays, social networks such as Twitter or Facebook become a robust means of learning about the users’ opinions and share their emotions towards specific subjects in a form of comments, to analysis these emotions sentiment analysis process is applied, which is used to discover the opinions of people on social media sites. It focuses on detection the polarity (positive, negative, or neutral). In recent years, it has been demonstrated that deep learning models are promising solution to the challenges of natural language processing (NLP). This study is devoted to apply semantic similarity approach for sentiment classification in addition to use lexical approach and Bag-of-Words model to perform a comparison among them. For examining the performance, precision, recall, accuracy, and F1 scores measurements with two datasets (STS-Test & SS-Tweet) for testing and sentiment140 for training have been used. The experimental results show the accuracy of the proposed approach about 81.0%.