GenRec: Large Language Model for Generative Recommendation (2307.00457v2)

Abstract: In recent years, LLMs (LLM) have emerged as powerful tools for diverse natural language processing tasks. However, their potential for recommender systems under the generative recommendation paradigm remains relatively unexplored. This paper presents an innovative approach to recommendation systems using LLMs based on text data. In this paper, we present a novel LLM for generative recommendation (GenRec) that utilized the expressive power of LLM to directly generate the target item to recommend, rather than calculating ranking score for each candidate item one by one as in traditional discriminative recommendation. GenRec uses LLM's understanding ability to interpret context, learn user preferences, and generate relevant recommendation. Our proposed approach leverages the vast knowledge encoded in LLMs to accomplish recommendation tasks. We first we formulate specialized prompts to enhance the ability of LLM to comprehend recommendation tasks. Subsequently, we use these prompts to fine-tune the LLaMA backbone LLM on a dataset of user-item interactions, represented by textual data, to capture user preferences and item characteristics. Our research underscores the potential of LLM-based generative recommendation in revolutionizing the domain of recommendation systems and offers a foundational framework for future explorations in this field. We conduct extensive experiments on benchmark datasets, and the experiments shows that our GenRec has significant better results on large dataset.

GenRec: LLM for Generative Recommendation

The paper "GenRec: LLM for Generative Recommendation" presents an innovative approach to recommender systems by integrating LLMs through a generative recommendation paradigm. The primary focus is leveraging the extensive capabilities of LLMs to directly generate recommendations based on user interactions and item characteristics captured in textual data.

Overview of GenRec

The innovative aspect of GenRec lies in its departure from traditional discriminative recommendation systems, where each candidate item is assessed individually through ranking scores. Instead, GenRec utilizes the inherent expressive power of LLMs to generate the target item for recommendation directly, offering a paradigm that potentially elevates efficiency and personalization in recommender systems.

The process involves formulating specialized prompts to enhance the comprehension abilities of the LLM regarding recommendation tasks. These prompts are used to fine-tune the LLaMA backbone LLM, a model noted for its flexibility and adaptability, on datasets comprising user-item interactions documented as text. Such an approach allows the capture of both user preferences and item attributes effectively, with GenRec demonstrating significant improvement over traditional methods, particularly when operating on large datasets.

Methodology and Experimental Results

The paper elaborates on the architecture of GenRec, highlighting its method of sequence generation, where user-item interaction sequences are reformatted through specifically designed prompts. This formatting accounts for the rich semantic information embedded in item names, potentially enhancing the model's ability to recommend accurately. The training employed the LLaMA-LoRA architecture, optimizing resource usage and ensuring efficient training on limited GPU capacity.

Experimental comparisons against baseline methods such as P5 showcased strong performance in scenarios utilizing the MovieLens 25M dataset, indicating GenRec's adeptness in extracting and leveraging rich interaction data. However, P5 maintained a slight advantage in scenarios typical of Amazon datasets, suggesting areas for potential refinement within the GenRec framework.

Implications and Future Directions

The implications of this paper are twofold: practical and theoretical. Practically, GenRec promises enhanced recommendation systems capable of delivering more personalized user experiences by understanding complex interaction patterns through textual data. Theoretically, it underscores the potential of generative approaches to challenge and possibly outperform traditional paradigms grounded in discriminative methodologies.

Future directions for GenRec include refining sequence generation with more sophisticated prompt formulations and expanding input data to include complex interaction types like ratings or reviews. Testing GenRec with different LLM frameworks could also reveal specific benefits and compromises associated with varying model architectures, suggesting a pathway to further improvements in recommendation systems.

Conclusion

The paper convincingly argues for the integration of LLMs into the recommendation domain, demonstrating that a generative approach can significantly enhance recommendation efficacy. By focusing on the rich semantic layers within textual data, GenRec provides a promising foundation for further exploration and potential innovation across the field of recommendation systems, offering a glimpse at the future of personalized digital experiences.