Reproduce “COSMIC: COmmonSense knowledge for eMotion Identification in Conversations” 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, Alexander Gelbukh, Rada Mihalcea, Soujanya Poria

A framework named COSMIC enhances emotion recognition in conversations by explicitly integrating commonsense knowledge derived from ATOMIC and COMET. This approach establishes new state-of-the-art performance across four benchmark datasets, including an increase to 65.28 Weighted Average F1 on IEMOCAP, by improving the model's understanding of speaker intentions, effects, and reactions.

Key insights

  • Explicitly integrating commonsense knowledge directly into the model significantly boosts performance in Emotion Recognition in Conversations, surpassing purely implicit contextual modeling.
  • Speaker-centric commonsense features, such as the speaker's intent and their internal/external reactions, are particularly crucial for accurately predicting emotional states.
  • The use of commonsense knowledge enables better detection of subtle emotion shifts and differentiation of fine-grained emotion classes, contributing to a more nuanced understanding of dialogue.

Summary by alphaXiv.

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