Open-domain sentiment analysis

Abstract

The proliferation of social media on the Internet in recent years has led to an increased amount of user-generated information worldwide. These platforms have allowed people of different backgrounds to share their opinions regarding the news and events occurring around them. Research regarding sentiment analysis had always been conducted within a closed domain, under the assumption of a static vocabulary. However, new words or phrases constantly appear in social media under various contexts. Thus, the sentiment lexicon built offline from the training data may be inaccurate when being used to make predictions. The proposed framework can extend a sentiment lexicon for Twitter based on incoming tweets dynamically. Results of the experiment has proven that expanding the sentiment lexicon dynamically does improve the accuracy and precision of the sentiment classifier, although the efficacy of the classifier is affected due to the added load.