Reproduce “DialogueGCN: A Graph Convolutional Neural Network for Emotion Recognition in Conversation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Deepanway Ghosal, Navonil Majumder, Soujanya Poria, Niyati Chhaya, Alexander Gelbukh

DialogueGCN, a Graph Convolutional Neural Network, introduces a method for emotion recognition in conversation by modeling dialogues as directed graphs that explicitly capture inter-speaker and temporal dependencies. The model establishes new state-of-the-art performance on the IEMOCAP, AVEC, and MELD datasets, demonstrating improved contextual understanding in conversational AI.

Key insights

  • Representing conversations as graphs with explicit speaker and temporal relations allows for a more robust and nuanced propagation of contextual information than sequential models.
  • Explicitly modeling inter-speaker and self-speaker (emotional inertia) dependencies is critical for accurate emotion recognition in dialogue.
  • Graph Convolutional Networks can effectively mitigate the long-term dependency problems prevalent in recurrent neural network-based models for emotion recognition in conversation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.