Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CFunModel: A "Funny" Language Model Capable of Chinese Humor Generation and Processing

Zhenghan Yu, Xinyu Hu, Xiaojun Wan

TL;DR

CFunSet aggregates multiple Chinese humor datasets and Tieba JokeBar to form a multi-task Chinese humor corpus, enabling training for diverse tasks such as Humor Recognition, Crosstalk Response Selection, Joke Generation, and Crosstalk Generation. Using CFunSet, the authors fine-tuned Qwen2.5-7B-Instruct to create CFunModel, the first open-source model specialized in Chinese humor. CFunModel achieves strong results across tasks, outperforming baselines in Humor Recognition and Crosstalk-related tasks, and demonstrates coherent joke and Crosstalk generation in qualitative cases. The work provides open datasets and an open-source model to advance Chinese humor research and practical applications.

Abstract

Humor plays a significant role in daily language communication. With the rapid development of large language models (LLMs), natural language processing has made significant strides in understanding and generating various genres of texts. However, most LLMs exhibit poor performance in generating and processing Chinese humor. In this study, we introduce a comprehensive Chinese humor-related dataset, the Chinese Fun Set (CFunSet). This dataset aggregates existing Chinese humor datasets and includes over 20,000 jokes collected from Tieba-JokeBar, a Chinese online platform known for joke sharing. The resulting corpus comprises more than 160,000 entries. Leveraging CFunSet, we developed the Chinese Fun Model (CFunModel), the first large language model designed to handle various Chinese humor-related tasks including Crosstalk Response Selection, Humor Recognition, Joke Generation, etc. Experimental results demonstrate that CFunModel outperforms popular large language models in these tasks. Our CFunSet is available at https://huggingface.co/datasets/ZhenghanYU/CFunSet and CFunModel is available at https://huggingface.co/ZhenghanYU/CFunModel. A demostration video of our work is available at https://youtu.be/MOsISOJ66Ms.

CFunModel: A "Funny" Language Model Capable of Chinese Humor Generation and Processing

TL;DR

CFunSet aggregates multiple Chinese humor datasets and Tieba JokeBar to form a multi-task Chinese humor corpus, enabling training for diverse tasks such as Humor Recognition, Crosstalk Response Selection, Joke Generation, and Crosstalk Generation. Using CFunSet, the authors fine-tuned Qwen2.5-7B-Instruct to create CFunModel, the first open-source model specialized in Chinese humor. CFunModel achieves strong results across tasks, outperforming baselines in Humor Recognition and Crosstalk-related tasks, and demonstrates coherent joke and Crosstalk generation in qualitative cases. The work provides open datasets and an open-source model to advance Chinese humor research and practical applications.

Abstract

Humor plays a significant role in daily language communication. With the rapid development of large language models (LLMs), natural language processing has made significant strides in understanding and generating various genres of texts. However, most LLMs exhibit poor performance in generating and processing Chinese humor. In this study, we introduce a comprehensive Chinese humor-related dataset, the Chinese Fun Set (CFunSet). This dataset aggregates existing Chinese humor datasets and includes over 20,000 jokes collected from Tieba-JokeBar, a Chinese online platform known for joke sharing. The resulting corpus comprises more than 160,000 entries. Leveraging CFunSet, we developed the Chinese Fun Model (CFunModel), the first large language model designed to handle various Chinese humor-related tasks including Crosstalk Response Selection, Humor Recognition, Joke Generation, etc. Experimental results demonstrate that CFunModel outperforms popular large language models in these tasks. Our CFunSet is available at https://huggingface.co/datasets/ZhenghanYU/CFunSet and CFunModel is available at https://huggingface.co/ZhenghanYU/CFunModel. A demostration video of our work is available at https://youtu.be/MOsISOJ66Ms.

Paper Structure

This paper contains 20 sections, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Chinese Humor Datasets
  4. Humor Evaluation
  5. CFunSet Dataset
  6. Overview
  7. Data Collection and Processing
  8. Humor-related Task Design
  9. Tieba—JokeBar
  10. CrossDial
  11. Chumor2.0
  12. HumorWB
  13. Chinese-Humor-Sentiment/Crosstalk-Generation
  14. CCL2019-Chinese-Humor-Computation
  15. CFunModel
  16. ...and 5 more sections

Figures (1)

  • Figure 1: Examples generated by Qwen2.5-7B-Instruct on the themes of lunch. This text is intended as a joke, but it falls short in delivering effective humor. The premise sets up a relatable situation where lunch occurs between breakfast and dinner, but the sentence highlighted in red lacks the punch typically needed for a joke, only conveying a sense of mild frustration or regret.