Natural Language Processing Techniques for Political Opinion & Sentiment

Abstract

In the field of quantitative research for political opinions and sentiment extractions, the most common method for sampling and data collection is surveys and the most common method for analysis is regression (linear, multiple, and / or logistic). However, surveys can be costly to execute, can be limited on where and who can be sampled, and can be lagged / outdated by several months. With the ubiquity of the Internet and the mainstream adoption of social media, utilizing text data from online can provide researchers fast, real-time access to current political sentiments / opinions. Moreover, the data from online can be more plentiful, more geographically precise, and more cost-effective. In using textual data from online, practitioners can employ powerful tools from language modeling, machine learning, and deep learning for analysis. This thesis creates a platform for NLP as a service / on demand. This platform currently offers political bias / sentiment identification, ranking the text input from very conservative to very liberal, and issue identification, providing relevance scores onto a variety of common political issues for the text input. Note this platform for now focuses exclusively on US-based politics.