ASPIRE: Building a Sentiment Lexicon from Ratings of Social Reviews

Finding semantic orientation and intensity of sentiment phrases and words is a subst antial task of opinion mining. The problem is to give a score to each sentiment phrase, so that different expressions of opinions in different platforms, like social networks, can be processed. There have been several attempts to do this task, and this paper aims to score each sentiment phrase based on its occurrence in reviews with different overall ratings. The idea is that if a sentime nt phrase occurs more in 5-starred reviews than in 3-starred ones, it should be more positive and more intense. The results support this idea. Each sentiment phrase in the corpus is given a score based on a weighted average of their frequency in reviews with different ratings. When a sentiment phrase gets a high score, it means it is more likely to be positive and more likely to be intense. And if a sentiment phrase gets a low score, it means that it is negative. This score sets the threshold of negativity and positivity. The high precision and recall for this feature shows its significance in classifying positive and negative sentiment phrases.