Achieving Few-Shot and Chain of Thought Prompting in Movie Recommendation: A ChatGPT-Based Solution

Recommender systems have a decisive impact across various platforms, such as streaming services and e-commerce websites. Traditional systems require extensive user-item interaction data for training, which can be challenging to collect and maintain. This paper explores the potential of utilizing pre-trained language models, specifically ChatGPT, as standalone recommender systems and presents a comparative analysis of different prompting techniques. We evaluate ChatGPT's performance in movie recommendation and rating prediction tasks and discuss its strengths and limitations. Additionally, we introduce an open-source codebase for constructing similar systems, demonstrate the applicability of GPT in scenarios with limited data, and showcase its use as a proof of concept for projects with data scarcity. The study aims to inspire further research on leveraging large-scale language models in recommendation systems.

Keywords:

Recommendation System , LLM , ChatGPT , Chain of Though Prompting , Few-Shot Prompting