Abstract: In this paper, contextual and semantic features are modeled in detail, and Emotion Cause Pair Extraction (ECPC) is carried out on the fusion of two modal features. For context modal, BiLSTM is used to convert word embedding into clause embedding to get emotion and cause representation, and the global context matrix is obtained by two-factor attention mechanism. For semantic modal, local semantic features are obtained by constructing Graph Convolution networks (GCN) through inter-clause semantics. Finally, the fusion features are obtained by the main and auxiliary mode matching method for multi-task prediction, including emotional sentence, cause sentence and emotion-cause pair extraction task. Extensive experiments have been done to verify that contextual and semantic fusion system (CSF-ECPE) is significantly improved by 2.2% compared with the best baseline system in the classical Chinese ECPC data.

Key words: Emotion Cause Pair Extraction, Global Context, Local semantics, Modal Matching