Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
Graph-Based Global Interaction Network with Assistant Prediction for Emotion-Cause Pair Extraction
Ist Teil von
2023 IEEE 6th International Conference on Pattern Recognition and Artificial Intelligence (PRAI), 2023, p.791-800
Ort / Verlag
IEEE
Erscheinungsjahr
2023
Quelle
IEL
Beschreibungen/Notizen
Emotion-cause pair extraction (ECPE) has received growing interest in recent years, which aims to extract all emotion clauses and corresponding cause clauses in a document. Previous methods first generate emotion and cause features separately and then generate pair features for pair extraction. These methods either fail to consider the relation between emotion and cause features encoder, or ignore the different importance of different information, resulting in unbalanced information in features, further leading to the wrong prediction. Besides, the original semantic information of clauses is ignored. In this paper, we propose a novel Graph-Based Global Interaction Network with Assistant Prediction (GGINAP) to address the problem. Our model can be divided into two stages: independent prediction and interactive prediction. In the first stage, independent pair features are generated from original clause representations for independent pair prediction. In the second stage, we define various types of meta-paths and apply Heterogeneous graph Attention Network (HAN) to model the global interaction between clauses and pairs, capturing contextual information and causal information simultaneously. Then interactive pair features are obtained for interactive pair prediction. To balance the negative effect of interaction, independent and interactive prediction are both considered for the final prediction. Extensive experiments demonstrate that our model outperforms existing methods and further analysis prove the effectiveness of our framework. Our code is available at https://github.com/klkkkkk/GGINAP.