Core Concepts
The author proposes the CKERC framework to enhance emotion recognition by incorporating commonsense knowledge into large language models, achieving state-of-the-art results on popular datasets.
Abstract
Emotion recognition in conversation is crucial for improving human-machine interactions. The CKERC framework leverages historical utterances and commonsense knowledge to enhance speaker information mining and achieve superior performance on emotion recognition tasks.
The dialogue emotion classification task is influenced by various factors such as dialogue context, speaker identity, and multi-party scenarios. Introducing more modal information can improve emotional understanding but may lead to dataset construction challenges.
Previous works have introduced commonsense into emotion recognition tasks, but CKERC stands out by utilizing historical utterances and large language models to generate accurate commonsense information for improved emotional responses.
By replacing speaker identification with interlocutor commonsense identification, CKERC addresses the challenge of insufficient speaker information mining during conversations. This approach leads to competitive performance on emotion recognition tasks across different datasets.
Stats
Based on the different history utterances,speaker generates different commonsense information for the same utterance"yeah." So, the same utterance "yeah." in the different conversation expresses different emotion.
The IEMOCAP dataset annotates 6 types of emotions: happy, sad, neutral, angry, excited, and frustrated.
The MELD dataset collected a total of 13118 dialogues covering 7 different emotional categories (anger, disgust, fear, happiness, neutrality, sadness, and surprise).
The EmoryNLP dataset includes 97 episodes with 12606 utterances and seven emotion labels: neutral, joyful, peaceful, powerful, scared, mad and sad.
Quotes
"The development of large language models has transformed natural language processing tasks."
"Our work introduces historical conversation information to predict common sense for current sentences."
"CKERC achieves state-of-the-art performance by leveraging commonsense knowledge in emotion recognition."