Sentiment Analysis

Sentiment analysis—also called opinion mining—involves automatically extracting and classifying sentiment from text. Core tasks include binary sentiment classification (positive vs. negative), multi-class classification (positive, negative, neutral, mixed), and aspect-based sentiment extraction. Sentiment models power recommendation systems, brand monitoring, and increasingly, attack validation in adversarial text generation scenarios.

Key papers

• A large-scale COVID-19 Twitter chatter dataset for open scientific research - an international collaboration — Analyzes sentiment of 800+ million COVID-19 tweets in near real-time; enables stratified measurement of public emotional responses to pandemic and public health measures
• Detecting and Tracking the Spread of Astroturf Memes in Microblog Streams — Applies GPOMS sentiment analysis to detect astroturfed political memes; six-dimensional mood vectors contributed to classifier but ranked below network topology features
• Generating Sentiment-Preserving Fake Online Reviews Using Neural Language Models and Their Human- and Machine-based Detection — Uses BERT-based sentiment classification to validate that generated fake reviews preserve the desired sentiment