Evaluating ChatGPT as a Recommender System: A Rigorous Approach

TL;DR

This paper introduces an open-source evaluation pipeline to rigorously assess ChatGPT as a standalone recommender across Books, Music, and Movies under zero-shot role-playing prompts. It assesses three tasks—Top-N, re-ranking, and cold-start—using both accuracy and beyond-accuracy metrics, and analyzes the similarity of ChatGPT-generated lists to content-based and collaborative-filtering baselines. Results show domain- and model-dependent strengths (notably ChatGPT-4 in Books and Movies, and strong re-ranking/cold-start performance) with clear trade-offs in diversity and popularity bias. The study provides evidence that ChatGPT engages latent collaborative signals and can complement traditional RS methods, and it publicly releases the evaluation framework for further research. The findings suggest a promising direction for leveraging LLMs in sparse or data-scarce regimes while highlighting the need for careful evaluation of beyond-accuracy criteria and ethical considerations in user-data usage.

Abstract

Large Language Models (LLMs) have recently shown impressive abilities in handling various natural language-related tasks. Among different LLMs, current studies have assessed ChatGPT's superior performance across manifold tasks, especially under the zero/few-shot prompting conditions. Given such successes, the Recommender Systems (RSs) research community have started investigating its potential applications within the recommendation scenario. However, although various methods have been proposed to integrate ChatGPT's capabilities into RSs, current research struggles to comprehensively evaluate such models while considering the peculiarities of generative models. Often, evaluations do not consider hallucinations, duplications, and out-of-the-closed domain recommendations and solely focus on accuracy metrics, neglecting the impact on beyond-accuracy facets. To bridge this gap, we propose a robust evaluation pipeline to assess ChatGPT's ability as an RS and post-process ChatGPT recommendations to account for these aspects. Through this pipeline, we investigate ChatGPT-3.5 and ChatGPT-4 performance in the recommendation task under the zero-shot condition employing the role-playing prompt. We analyze the model's functionality in three settings: the Top-N Recommendation, the cold-start recommendation, and the re-ranking of a list of recommendations, and in three domains: movies, music, and books. The experiments reveal that ChatGPT exhibits higher accuracy than the baselines on books domain. It also excels in re-ranking and cold-start scenarios while maintaining reasonable beyond-accuracy metrics. Furthermore, we measure the similarity between the ChatGPT recommendations and the other recommenders, providing insights about how ChatGPT could be categorized in the realm of recommender systems. The evaluation pipeline is publicly released for future research.

Figures (7)

• Figure 1: Multi-Stage Evaluation Pipeline for Assessing ChatGPT Recommendation Performance.
• Figure 2: Kiviat diagrams illustrate the performance of the models on various datasets. For each Kiviat diagram, a comparison is presented among the ChatGPT models, the BestCF (Best Collaborative Filtering), the BestCBF (Best Content-Based Filtering), and the MostPop (Most Popular). Higher values indicate better performance.
• Figure 3: Grouped bar charts illustrating the performance of the ChatGPT models across three domains. Each grouped bar chart displays the baseline nDCG score and the improvement in terms of nDCG score. Higher improvements in nDCG indicate a better ability of the model to re-rank and personalize a list of items according to individual user preferences.
• Figure 4: Mistral-7B Few-Shot prompt
• Figure 5: ChatGPT Response Example