تحليل المشاعر والسمات الدلالية في نصوص وسائل التواصل الاجتماعي

Social media sites have turned into abundant sources of real-time opinionated text. The present work investigates if incorporating semantic features can improve the sentiment analysis accuracy of social media texts. It creates and contrasts two models: one baseline model based on the traditional text-based features, and one enhanced model based on the semantic features such as contextual word embedding and sentiment lexicons. The results indicate that the semantic-enhanced model performed much better using a large amount of tweets, especially when it came to detecting sentiment in more complex and context-dependent scenarios like sarcasm and informal slang expressions. The results validate the hypothesis that semantic understanding plays an important role for sentiment classification. Its results are relevant for any organization aiming to gain more trustworthy information from social media, and also have research value by combining surface analysis with deeper semantic understanding.