Sensible Tweets: Speech Act Filtering for Effective Sentiment Analysis
Thad Chee, Erika Wu, Harris Wu · 2022
Existing sentiment analyses of twitter data often develop ad hoc rules of selecting or discarding tweets. To move towards a rigorous methodology for the selection of short messages or so-called microtexts in sentiment analysis, we develop a sentiment analysis framework based on Speech Act theory with a focus on speech act filtering to identify sensible tweets – the sentiment-bearing microtexts that are perceptible to sentiment analysis. We develop a neural network classifier to implement speech act filtering and apply the sentimental analysis framework to twitter data. Using Twitter API, our speech act classifier and an online sentiment analysis tool, we conduct a pilot study on financial tweets to show that speech act filtering leads to higher quality sentiment analysis results.