Analyzing Subjectivity and Sentiment of Online Forums
Open Access
- Author:
- Biyani, Prakhar
- Graduate Program:
- Information Sciences and Technology
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- August 11, 2014
- Committee Members:
- Prasenjit Mitra, Dissertation Advisor/Co-Advisor
Prasenjit Mitra, Committee Chair/Co-Chair
John Yen, Committee Member
Alexander Klippel, Committee Member
Marcel Salathe, Committee Member
Cornelia Caragea, Special Member - Keywords:
- Subjectivity analysis
sentiment analysis
classification
supervised learning
semi-supervised learning
online forums - Abstract:
- Online social media has emerged as a popular medium for seeking and providing information, opinions and social support. Online sites such as discussion forums, blogs and health communities have tremendous amounts of user generated data in their archives. Analyzing this content for its subjectivity and sentiment has important applications such as improving information search in social media, understanding users for providing content personalization, identifying influential members in online communities, etc. In this dissertation, I will discuss my works on subjectivity analysis of online forum threads, identifying the type of social support (emotional or informational) present in and analyzing sentiment of user messages in an online health community (OHC). For subjectivity analysis, I show that thread-specific non-lexical features such as thread structure and dialogue acts expressed in thread posts are highly informative for inferring thread subjectivity. For sentiment analysis of messages of the OHC, I use unlabeled messages to augment a small training data using co-training and build highly accurate sentiment classifiers. For support identification, I build supervised classifiers using several generic and novel domain-specific features and analyze the posting behaviors of regular members and influential members in the OHC in terms of the type of support they provide in their messages. I find that influential members generally provide more emotional support as compared to regular members in the OHC. Experimental results demonstrate that all the proposed models significantly outperform various state-of-the-art models.