Sentiment Uncertainty and Spam in Twitter Streams and Its Implications for General Purpose Realtime Sentiment Analysis
This work highlights a critical limitation in realtime sentiment analysis for social media applications, offering a new perspective and dataset to improve accuracy.
The paper addresses the problem that over half of tweets lack a distinct sentiment, which standard benchmarks ignore, by proposing to incorporate uncertainty into Twitter sentiment analysis and providing a high-quality public dataset for evaluation.
State of the art benchmarks for Twitter Sentiment Analysis do not consider the fact that for more than half of the tweets from the public stream a distinct sentiment cannot be chosen. This paper provides a new perspective on Twitter Sentiment Analysis by highlighting the necessity of explicitly incorporating uncertainty. Moreover, a dataset of high quality to evaluate solutions for this new problem is introduced and made publicly available.