An Enhanced Twitter Sentiment Analysis Model using Negation Scope Identification Methods

Abstract

Sentiment Analysis (SA), which is also known as Opinion Mining, is a hot-fastest growing research area, making it challenging to follow all its activities. It intends to study peoples' thoughts, feelings, and attitudes about topics, events, issues, entities, individuals, and attributes in social media (e.g., social networking sites, forums, blogs, etc.) expressed by either text comments or tweets. Twitter is one of the world's largest online microblogging platforms that allows its users to freely post texts called tweets. It offers a wealth of information, therefore utilizing SA to analyze this information into positive or negative will assist organizations' and customers' decision-making that will have a significant impact on daily life. SA draws the attention of scientific research in the Natural Language Processing community due to the text structure challenges that may contain negation. Negation is a widespread linguistic structure that changes the text meaning to the opposite and affects text polarity. Therefore, it needs to be considered in sentiment analysis systems. In this paper, supervised machine learning models have been used as a baseline to categorize the sentiment of a Twitter dataset using (Bag of Words) and (Term Frequency Inverse Document Frequency) feature representation methods. Then we applied negation scope identification methods to find negated tokens and investigate how embedding these tokens can raise SA classifiers' accuracy. The results of the sentiment classification task show an improvement once considering these tokens.