The task of sentiment analysis aims at understanding the opinions in the text. For example, in the sentence ''It is great that he was promoted.'' versus ''It is great that he was fired.'', there is an positive sentiment in both sentences because of the positive sentiment word ''great''. Previous work of sentiment analysis may stop here. However, the sentiment toward ''he'' in the former sentence is positive, while the sentiment toward ''he'' in the later sentence is negative. Although there is no sentiment word directly modifying ''he'' in the sentences, they are actually indicated. While previous work cannot recognize such sentiment expressed toward ''he'', this talk contributes to developing a sentiment analysis system to recognize such sentiments in the sentences.
Specifically, the sentiments toward ''he'' cannot be recognized merely relying on sentiment lexicons since such sentiments are not directly associated with sentiment words. Instead, they need to be inferred. We present two sets of inference rules, and conduct the experiments to demonstrate the inference ability of the rules. Based on the positive result of the experiments, we then proceed to develop computational models to automatically infer the sentiments, using the rules as soft constraints of the models. What’s more important, the models take into account the information not only from sentiment analysis tasks but also from other Natural Language Processing tasks including information extraction and semantic role labelling. The models jointly solve different NLP tasks in one single model and improve the performances of the tasks.