Reliable Sentiment Analysis in Social Media
- Publication Type:
- Thesis
- Issue Date:
- 2020
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Sentiment analysis in social media is critical yet challenging because the source materials (i.e., reviews posted in social media) are with high complexity, low quality, and uncertain credibility. For example, words and sentences in a textual review may couple with each other, and they may have heterogeneous meanings under different contexts or in different language locales. These couplings and heterogeneities essentially determine the sentiment polarity of the review but are too complex to be captured and modeled. Also, social reviews contain a large number of informal words and typos (a.k.a., noise) but a rare number of vocabularies (a.k.a., sparsity). As a result, most of the existing natural language processing (NLP) methods may fail to represent social reviews effectively. Furthermore, a large proportion of social reviews are posted by fraudsters. These fraud reviews manipulate social opinion, and thus, they disturb sentiment analysis.
This research focuses on reliable sentiment analysis in social media. It systematically investigates the sentiment analysis techniques to tackle three major challenges in social media: high data complexity, low data quality, and uncertain credibility. Specifically, this research focuses on two research problems: general sentiment analysis in social media and fraudulent sentiment analysis in social media. The general sentiment analysis targets on tackling high data complexity and low-quality of social articles that are credible. The fraudulent sentiment analysis handles the uncertain credibility issue, which is common and profoundly affects the precise sentiment analysis in social media. Based on these investigations, this research proposes a serial of methods to achieve reliable sentiment analysis: It studies the polarity-shift characteristics and non-IID characteristics in general paragraphs to capture the sentiment more accurately. It further models multi-granularity noise and sparsity in short text, which is the most common data in social media, for robust short text sentiment analysis. Finally, it tackles the uncertain credibility problem in social media by studying fraudulent sentiment analysis in both supervised and unsupervised scenarios.
This research evaluates the performance and properties of the proposed reliable sentiment analysis methods by extensive experiments on large real-world data sets. It demonstrates that the proposed methods are superior and reliable in social media sentiment analysis.
Please use this identifier to cite or link to this item: