Novel Sentiment-Detection Algorithms for Tweets
Detecting the sentiment contained in a text has been studied thoroughly and we are able to detect whether a given text is sad or happy at a high accuracy. We can even detect the strength of the detected sentiment. However, when it comes to short messages (such as tweets containing only 140 characters), where the amount of information is limited, there is still potential for improving current approaches. In this master thesis, we aim to gather a deep understanding of sentiment analysis and detection methods for short texts, particularly for tweets and to improve those methods. We aim at facilitating so-called factorization machines and ensemble methods for improving the state-of-the-art approaches in sentiment detection. Further, we aim at participating at the Semantic Evaluation Exercises (SemEval) to be able to directly compare our findings with others.