Emotion classification based on structured information

In the era of information explosion, more social network applications present a platform for people to share various news and information sources, which brings people into the era of big data. And the processing of structured information attracts more researchers' attention. In this paper, we propose a method of feature extraction based on the syntactic and grammar structure to discover the emotion of a sentence. Firstly, an emotional lexicon is constructed by the combination of Chi-square test, PMI with word2vec which is based on different types of neural networks. Secondly, we improve the quality of selected features by exploring Part-Of-Speech features, capturing various types of relationships through syntactic analysis, and focusing on the emotional words features in context. Then we experiment with diverse linguistically motivated features. The experimental results validate the feasibility of our approach in selecting informative features, and the existing problems and the future works are also present in the end.