Evaluating ChatGPT as a Recommender System: A Rigorous Approach (2309.03613v2)

Abstract: 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.

Evaluating the Potential of ChatGPT as a Recommender System

Introduction

Recommendation systems are an integral part of our digital lives, guiding us through vast seas of options to suggest products, content, and information tailored to our tastes. A paper delved into the capabilities of a particular type of system, ChatGPT, within this context.

Recommender Systems and ChatGPT

ChatGPT, a conversational agent based on the powerful GPT-3.5 LLM, has shown great potential as a recommender system. It has been trained on extensive data, learning to predict users' preferences and suggest items accordingly. Researchers have posed a critical question: Could this AI also excel as a recommender system, a tool to personalize suggestions in various domains like movies, music, and books?

The Study's Methodology

The paper employed a robust method to evaluate ChatGPT's recommendations against standard algorithms in the field. Researchers used three public datasets—MovieLens Small, Last.FM, and Facebook Book—to compare the performances. The evaluative benchmarks included classic metrics like Mean Average Precision and normalized Discounted Cumulative Gain, considering various aspects such as accuracy, diversity, novelty, and bias. It also included an assessment of how well ChatGPT could handle the notorious cold start problem faced by recommender systems.

Findings and Insights

The findings indicate that ChatGPT, even without optimization for the tasks, showcases promising capabilities. Performing comparably to state-of-the-art systems, it excels in understanding user preferences and recommending new items. The paper also suggests that ChatGPT, along with other LLMs, can manage cold-start scenarios effectively, where a user’s historical data is scarce or non-existent.

Despite its potential, the observations also indicate that ChatGPT's recommendations may exhibit popularity biases depending on the dataset. In terms of system similarity, the AI shows alignment with hybrid and collaborative systems rather than purely content-based approaches. Additionally, when presented with lists for re-ranking based on user preferences, it demonstrated significant improvements, leaning towards more personalized suggestions.

Future Directions

This research lays the groundwork for future studies that might explore prompt engineering or domain-specific fine-tuning. The consistent and high-performance of ChatGPT suggests it could revolutionize recommendation tasks, though it necessitates further research on enhancing performance and addressing biases.

Conclusion

The versatility of ChatGPT as a LLM extends to the field of recommenders, holding promises for personalized, efficient, and contextually relevant recommendations. Its potential application in various domains opens up possibilities for richer user experiences across digital services.