Integrating Large Language Models into Recommendation via Mutual Augmentation and Adaptive Aggregation

TL;DR

This work introduces Llama4Rec, a model-agnostic framework that jointly leverages conventional recommender models and large language models (LLMs) to improve recommendation. It achieves mutual augmentation through data augmentation for conventional models and prompt augmentation for LLMs, followed by an adaptive aggregation that accounts for user tail behavior. The approach demonstrates consistent gains across direct, sequential, and rating-prediction tasks on three real-world datasets, with notable improvements in top-k metrics and reduced data-sparsity effects. The findings highlight the practical viability of integrating LLMs with traditional recommender systems and point to future work in iterative augmentation and grounding for further gains.

Abstract

Conventional recommendation methods have achieved notable advancements by harnessing collaborative or sequential information from user behavior. Recently, large language models (LLMs) have gained prominence for their capabilities in understanding and reasoning over textual semantics, and have found utility in various domains, including recommendation. Conventional recommendation methods and LLMs each have their strengths and weaknesses. While conventional methods excel at mining collaborative information and modeling sequential behavior, they struggle with data sparsity and the long-tail problem. LLMs, on the other hand, are proficient at utilizing rich textual contexts but face challenges in mining collaborative or sequential information. Despite their individual successes, there is a significant gap in leveraging their combined potential to enhance recommendation performance. In this paper, we introduce a general and model-agnostic framework known as \textbf{L}arge \textbf{la}nguage model with \textbf{m}utual augmentation and \textbf{a}daptive aggregation for \textbf{Rec}ommendation (\textbf{Llama4Rec}). Llama4Rec synergistically combines conventional and LLM-based recommendation models. Llama4Rec proposes data augmentation and prompt augmentation strategies tailored to enhance the conventional model and LLM respectively. An adaptive aggregation module is adopted to combine the predictions of both kinds of models to refine the final recommendation results. Empirical studies on three real-world datasets validate the superiority of Llama4Rec, demonstrating its consistent outperformance of baseline methods and significant improvements in recommendation performance.

Figures (8)

• Figure 1: An example that illustrates the motivation of Llama4Rec.
• Figure 2: (i) The overall framework architecture of the proposed Llama4Rec consists of two main components: mutual augmentation and adaptive aggregation. The red dashed lines denoted the data augmentation and prompt augmentation process, respectively. (ii) Illustration of the data augmentation process, encompassing three diverse recommendation scenarios. (iii) The pipeline of the adaptive aggregation module, which includes the aggregation of predictions from both types of models. The symbols $u$, $i$, and $a$ represents user, item, and attributes, respectively.
• Figure 3: Examples of instructions for top-k recommendation and rating prediction. The prompt augmentation component is highlighted. The similar user histories and conventional model predictions are prompt augmented, which are highlighted in light purple and beige. We keep the prompt template structure and convert the bullet-point format into cohesive sentences for use in experiments. In the top-k recommendation, both direct and sequential recommendations follow the same prompt structure, with the key difference being that the user interaction history for sequential recommendations is time-series aware.
• Figure 4: Impact of hyper-parameters $\alpha_1$ and $\alpha_2$ on ML-1M dataset with backbone model LightGCN.
• Figure 5: Performance comparison w.r.t different LLaMA-2 size for training Llama4Rec on the Bookcrossing dataset.