In this paper, we address the task of utterance level emotion recognition in\nconversations using commonsense knowledge. We propose COSMIC, a new framework\nthat incorporates different elements of commonsense such as mental states,\nevents, and causal relations, and build upon them to learn interactions between\ninterlocutors participating in a conversation. Current state-of-the-art methods\noften encounter difficulties in context propagation, emotion shift detection,\nand differentiating between related emotion classes. By learning distinct\ncommonsense representations, COSMIC addresses these challenges and achieves new\nstate-of-the-art results for emotion recognition on four different benchmark\nconversational datasets. Our code is available at\nhttps://github.com/declare-lab/conv-emotion.\n