Text Emotion Cause Detection Based On Emotional Common Sense Knowledge-base

With the rapid development of Internet and the rise of social media, the social media texts, especially the ones with emotion information, grow explosively. Thus, the public opinion monitoring and group event prediction on these texts became more important. Up to now, the text emotion analysis becomes practical. Base on this, as an important research direct, the study on text emotion cause detection attracts much research interest because it has shown great research value and practical significance.This study investigates the approach for constructing emotional common sense knowledge-base based on large scale text, and method for detecting emotion cause by incorporating the developed knowledge base. The main work of this study including: Firstly, design and construct a corpus with annotated emotion cause on news text. The corresponding annotation framework and quality control scheme are designed. Currently, the corpus with 2,105 annotated instances is constructed. Then, the statistical analysis are performed on the annotated corpus for estimate the distribution characteristics of cause structure and cause-related rules and distribution differences between news text and Weibo text. Secondly, investigate the approach for automatic construction and expansion the common sense knowledge-base. Here, the reader emotional voting data for Sina social news are utilized as the clues to automatic tagging the emotion of large scale social news text. These labeled texts are used to mining the common sense knowledge related to emotion for constructing the emotional common sense knowledge-base with arousal strength. Take into account such common sense knowledge-base has the problem of insufficient coverage, an automatic knowledge expansion approach using the How Net and the Tongyici Cilin are designed and implemented. The base knowledge is expanded for three times effectively. Thirdly, investigate the approach for detecting the emotion cause from text based on emotional common sense knowledge base. The evaluations show that this approach improves the F1 performance of emotion detection for 12.37% from the baseline. Based on this, the emotion cause detection approaches, which incorporate knowledge-base with rule-based approach and machine learning based approach, respectively, are investigated. The evaluations show that these two approaches further improve the F1 performance on emotion detection for 1.27% and 15.01%, respectively. This study shows that the emotional common sense knowledge base may be employed to improve the emotion cause detection approaches as effective knowledge source and additional...