Novel Sentiment-Detection Algorithms for Tweets
Twitter has grown to one of the most popular microblogging websites in recent years and therefore, has become a major source of opinionated texts. Since the length of these Twitter messages is limited to 140 characters, detecting whether such a tweet conveys a positive or negative sentiment is a challenging task and has become a popular field of study in natural language processing. In this thesis, we propose an approach called SentTwi to classify the sentiment of tweets as either positive, neutral or negative. The main goal of SentTwi is to demonstrate the suitability of so-called factorization machines and ensemble methods in conjunction with state-of-
the-art approaches like part-of-speech tagging, sentiment lexicons and term frequency weighting to detect the sentiment of tweets. A quality evaluation based on the datasets of the International Semantic Evaluation (SemEval) challenge 2016 is performed to evidence that SentTwi provides comparable results than other sentiment detection systems.