# A Preliminary Study of ChatGPT on News Recommendation: Personalization, Provider Fairness, Fake News

**Authors:** Xinyi Li, Yongfeng Zhang, Edward C. Malthouse  
**arXiv:** [2306.10702](https://arxiv.org/abs/2306.10702)  
**Date:** 2023-06-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.10702v1/

## Key points

- Personalized news recommendation methods are commonly used in online news platforms.
- Previous studies have relied on language model techniques for user interests and news content understanding.
- Large language models like GPT-3 and T-5 have led to a new recommendation paradigm.
- ChatGPT is a popular language model due to its user-friendly interface and versatility in text-based tasks.
- Language models like ChatGPT are increasingly being used for news recommendations.
- This study focuses on personalized news recommendation, news provider fairness, and fake news detection using ChatGPT.
- The study aims to investigate constraints in ChatGPT's generated responses and explore prompt formats to alleviate these constraints.
- A webpage is developed to monitor ChatGPT's performance on a weekly basis for the investigated tasks and prompts.
- The goal of the study is to enhance news recommendation performance using large language models like ChatGPT.

## Summary

In the field of online news platforms, personalized news recommendation methods are commonly used to help users discover interesting articles. Previous studies have relied on language model techniques to capture user interests and understand news content. However, with the emergence of large language models like GPT-3 and T-5, a new recommendation paradigm has emerged that leverages pre-trained language models for making recommendations. One such language model is ChatGPT, which has gained popularity due to its user-friendly interface and versatility in text-based tasks. Given the increasing reliance on ChatGPT for language tasks and the importance of news recommendation in addressing social issues, there is a growing trend of using language models like ChatGPT for news recommendations. To explore the performance of ChatGPT in news recommendations, this study focuses on three perspectives: personalized news recommendation, news provider fairness, and fake news detection. However, it is important to note that ChatGPT has limitations in terms of sensitivity to input phrasing. Therefore, this study aims to investigate the constraints present in the generated responses of ChatGPT from each perspective. Additionally, the study explores whether specific prompt formats can alleviate these constraints or if further attention from researchers is required to address these limitations. To go beyond fixed evaluations, the researchers develop a webpage that monitors ChatGPT's performance on a weekly basis for the investigated tasks and prompts. The ultimate goal of this study is to contribute to enhancing news recommendation performance by utilizing large language models like ChatGPT. The findings aim to encourage more researchers to engage in this area of study and explore ways to improve personalized news recommendations through advanced language models.