Fuzzy Subjective Sentiment Phrases: A Context Sensitive and Self-Maintaining Sentiment Lexicon

In this paper, we present a novel self-maintaining, domain-independent, and context-sensitive Sentiment Lexicon (SL) which finds and maps opinion words and phrases to a fuzzy sentiment score ranging from strong negative to strong positive. We show that our automatically built SL has advantage over other already existing lexicons in various aspects, namely, reducing the number of word false-matches by using context information. Furthermore, through iterative learning, our lexicon not only learns the sentiment score of a target word but also the sentiment scores of phrases and top word collocations with which the target word frequently co-occurs. Our experimental results suggest that our proposed method is highly effective and can produce lexicons which outperform a standard manually built SL, such as the AFINN lexicon, which was used in our experiments.