An Event-based emotion Corpus: Cause and Effect

This chapter will introduce a Chinese event-based emotion corpus and explore an important task in emotion analysis—emotion cause detection. The corpus design and the data collection and annotation procedures will be described, as will the emotion cause detection task, which aims to detect the triggering cause of an emotion automatically. We regard the detection task as a sequence labeling issue and determined whether a clause contained an emotion cause, accordingly. In concrete terms, the conditional random field (CRF) model was adopted with various features taken into consideration for the detection task, such as lexical features, part-of-speech features, contextual features, and linguistic features. The experiments demonstrated that all these features were effective in recognizing emotion causes, in particular, the contextual features. In addition, the sequence labeling model yielded better performance than the multi-label classification model when similar features were employed.