# Zero-Shot Next-Item Recommendation using Large Pretrained Language Models

**Authors:** Lei Wang, Ee-Peng Lim  
**arXiv:** [2304.03153](https://arxiv.org/abs/2304.03153)  
**Date:** 2023-04-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.03153v1/

## Key points

- Large language models (LLMs) can be used for next-item recommendations in the zero-shot setting.
- LLMs have achieved impressive zero-shot performance in various natural language processing tasks, but their application as recommenders is unexplored.
- Two major challenges for LLMs as recommenders are the large recommendation space and lack of knowledge about user preferences.
- The authors propose a prompting strategy called Zero-Shot Next-Item Recommendation (NIR) prompting to address these challenges.
- NIR prompting involves using an external module to generate candidate items based on user or item filtering.
- The proposed approach incorporates a 3-step prompting process to guide GPT-3, a specific LLM, in capturing user preferences and recommending movies.
- Evaluation on the MovieLens 100K dataset shows strong zero-shot performance, outperforming some sequential recommendation models trained on the entire dataset.
- This research opens up opportunities for using LLMs as recommenders and provides a novel strategy for next-item recommendations in the zero-shot setting.
- Code for implementing this strategy is available on GitHub at https://github.com/AGI-Edgerunners/LLM-Next-Item-Rec.

## Summary

The paper discusses the potential of large language models (LLMs) to perform next-item recommendations in the zero-shot setting. While LLMs have achieved impressive zero-shot performance in various natural language processing tasks, their application as recommenders has not been explored. The authors identify two major challenges that need to be addressed for LLMs to act effectively as recommenders: the large recommendation space and the lack of knowledge about the target user's past interacted items and preferences. To address these challenges, the authors propose a prompting strategy called Zero-Shot Next-Item Recommendation (NIR) prompting. This strategy involves using an external module to generate candidate items based on user-filtering or item-filtering. The proposed approach incorporates a 3-step prompting process that guides GPT-3, a specific LLM, to capture the user's preferences, select representative previously watched movies, and recommend a ranked list of 10 movies. The authors evaluate their approach using GPT-3 on the MovieLens 100K dataset and demonstrate its strong zero-shot performance. In fact, their approach outperforms some strong sequential recommendation models trained on the entire training dataset. These promising results highlight the research opportunities to use LLMs as recommenders. In conclusion, this paper presents a novel strategy for next-item recommendations in the zero-shot setting using large pretrained language models. The proposed approach shows significant potential and opens up avenues for further research in this area. The code for implementing this strategy is available on GitHub at https://github.com/AGI-Edgerunners/LLM-Next-Item-Rec., providing researchers with an easy way to explore these ideas further.