Issue |
ITM Web Conf.
Volume 70, 2025
2024 2nd International Conference on Data Science, Advanced Algorithm and Intelligent Computing (DAI 2024)
|
|
---|---|---|
Article Number | 03024 | |
Number of page(s) | 10 | |
Section | Image Processing and Computer Vision | |
DOI | https://doi.org/10.1051/itmconf/20257003024 | |
Published online | 23 January 2025 |
Advancing Computational Humor: LLaMa-3 Based Generation with DistilBert Evaluation Framework
1 Department of Computer Science, The University of Hong Kong, Hong Kong, 999077, China
2 School of Data Science, The Chinese University of Hong Kong, Shenzhen, 518000, China
Corresponding author: u3594927@connect.hku.hk
Humor generation presents significant challenges in the field of natural language processing, primarily due to its reliance on cultural backgrounds and subjective interpretations. These factors contribute to the variability of human-generated humor, necessitating computational models capable of mastering diverse comedic styles with minimal subjectivity and maximal generalizability. This study introduces a novel approach to humor generation by fine-tuning the LLaMA-3 language model with Low-Rank Adaptation (LoRA). The study developed a comprehensive dataset sourced from diverse online platforms, supplemented by non-humorous content from scientific literature and press conferences to enhance the model's discriminative capabilities. Utilizing DistilBERT for efficient evaluation, the fine-tuned LLaMA-3 achieved an impressive accuracy of 95.6% and an F1-score of 97.75%, surpassing larger models such as GPT-4o, and Gemini. These results demonstrate the model's exceptional capability in generating humor, offering a more efficient and scalable solution for applications such as conversational agents and entertainment platforms. This research advances the field by showcasing the benefits of comprehensive dataset preparation and targeted fine-tuning, providing a foundation for future developments in humor-related artificial intelligence applications.
© The Authors, published by EDP Sciences, 2025
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.