Recognizing Opinion Sources Based on a New Categorization of Opinion Types / 2775
Lingjia Deng, Janyce Wiebe
Recognizing sources of opinions is an important task in sentiment analysis. Different from previous works which categorize an opinion according to whether the source is the writer or the source is a noun phrase, we propose a new categorization of opinions according to the role that the source plays. The source of a participant opinion is a participant in the event that triggers the opinion. On the contrary, the source of a non-participant opinion is not a participant. Based on this new categorization, we classify an opinion using phrase-level embeddings. A transductive learning method is used for the classifier since there is no existing annotated corpora of this new categorization. A joint prediction model of Probabilistic Soft Logic then recognizes the sources of the two types of opinions in a single model. The experiments have shown that our model improves recognizing sources of opinions over baselines and several state-of-the-art works.