Table of Contents
Fetching ...

Is ChatGPT Equipped with Emotional Dialogue Capabilities?

Weixiang Zhao, Yanyan Zhao, Xin Lu, Shilong Wang, Yanpeng Tong, Bing Qin

TL;DR

The paper investigates ChatGPT's emotional dialogue capabilities by separating understanding and generation tasks across ERC, Emotion Cause Recognition (CEE), and Dialog Act Classification (DAC), as well as empathetic response generation and emotional support conversations. It benchmarks ChatGPT against SOTA baselines using zero- and few-shot prompts, revealing consistent gaps in understanding tasks but promising performance in generating empathetic and supportive responses. Limitations include alignment with dataset guidelines, evaluation metric gaps, and limited model scope. The work points to future directions in prompt engineering, dataset-standard alignment, and developing evaluation methods that better reflect human judgments in affective dialogue.

Abstract

This report presents a study on the emotional dialogue capability of ChatGPT, an advanced language model developed by OpenAI. The study evaluates the performance of ChatGPT on emotional dialogue understanding and generation through a series of experiments on several downstream tasks. Our findings indicate that while ChatGPT's performance on emotional dialogue understanding may still lag behind that of supervised models, it exhibits promising results in generating emotional responses. Furthermore, the study suggests potential avenues for future research directions.

Is ChatGPT Equipped with Emotional Dialogue Capabilities?

TL;DR

The paper investigates ChatGPT's emotional dialogue capabilities by separating understanding and generation tasks across ERC, Emotion Cause Recognition (CEE), and Dialog Act Classification (DAC), as well as empathetic response generation and emotional support conversations. It benchmarks ChatGPT against SOTA baselines using zero- and few-shot prompts, revealing consistent gaps in understanding tasks but promising performance in generating empathetic and supportive responses. Limitations include alignment with dataset guidelines, evaluation metric gaps, and limited model scope. The work points to future directions in prompt engineering, dataset-standard alignment, and developing evaluation methods that better reflect human judgments in affective dialogue.

Abstract

This report presents a study on the emotional dialogue capability of ChatGPT, an advanced language model developed by OpenAI. The study evaluates the performance of ChatGPT on emotional dialogue understanding and generation through a series of experiments on several downstream tasks. Our findings indicate that while ChatGPT's performance on emotional dialogue understanding may still lag behind that of supervised models, it exhibits promising results in generating emotional responses. Furthermore, the study suggests potential avenues for future research directions.
Paper Structure (41 sections, 5 figures, 10 tables)

This paper contains 41 sections, 5 figures, 10 tables.

Figures (5)

  • Figure 1: The emotional dialogue capability of a chatbot can be divided into two aspects: understanding and generation capability, with several downstream tasks.
  • Figure 2: An example poria2019emotion of Emotion Recognition in Conversations (ERC).
  • Figure 3: An example from RECCON-DD dataset poria2021recognizing for identifying the causal utterances.
  • Figure 4: An example of empathetic conversation from EmpatheticDialogue dataset rashkin2018towards.
  • Figure 5: An example of emotional support conversation from ESConv dataset liu2021towards.