Twitter Sentiment Analysis with a Deep Neural Network: An Enhanced Approach using User Behavioral Information

Ahmed Alharbi

Department of Computer Science
Western Michigan University
Doctoral Committee:
Prof. Elise DeDoncker, Chair
Prof. Alvis Fong, Member
Prof. Ikhlas Abdel-Qader, Member

Date: Oct. 4 (Thu) at 10:00 am

Room: The Parkview Room (D132)
Sentiment analysis on social media such as Twitter has become a very important and challenging task. Due to the characteristics of such data— tweet length, spelling errors, abbreviations, and special characters— the sentiment analysis task in such an environment requires a non-traditional approach. Moreover, social media sentiment analysis is a fundamental problem with many interesting applications. Most current social media sentiment classification methods judge the sentiment polarity primarily according to textual content and neglect other information on these platforms. In this research, we propose a deep learning based framework that also incorporates user behavioral information within a given document (tweet). Within this framework, there are several models based on a variety of neural network architectures. Each one of these models is trained on a specific aspect of user behavior. Then, the framework exploits these multi-aspect learning models to jointly aim for a mutual task (the sentiment analysis task). The results of the preliminary experiments, which are reported in (Alharbi & DeDoncker, 2017), demonstrate that going beyond the content of a document (tweet) is beneficial in sentiment classification, because it provides the classifier with a deep understanding of the task.