Can Language Models Improve the Performance of SVD-based Recommender Systems?

Abstract

Traditional recommendation algorithms cannot provide personalized recommendations based on user preferences provided through text, e.g., “I like movies which take me into a dreamland”. Large Language Models (LLMs) have emerged as one of the most promising tools for natural language processing in recent years. This research proposes a framework that leverages the capabilities of LLMs to enhance movie recommendation systems by refining the recommendations of traditional recommendation systems and integrating them with language-based user preference inputs. We employ a Singular Value Decomposition (SVD) algorithm to generate initial movie recommendations. The base SVD algorithm is implemented from the Surprise Python library and trained on the MovieLens 32M dataset.