User sentiment detection: a YouTube use case

In this paper we propose an unsupervised lexicon-based approach to detect the sentiment polarity of user comments in YouTube. Polarity detection in social media content is challenging not only because of the existing limitations in current sentiment dictionaries but also due to the informal linguistic styles used by users. Present dictionaries fail to capture the sentiments of community-created terms. To address the challenge we adopted a data-driven approach and prepared a social media specific list of terms and phrases expressing user sentiments and opinions. Experimental evaluation shows the combinatorial approach has greater potential. Finally, we discuss many research challenges involving social media sentiment analysis.