| Summary: | Recommendation systems have emerged as a pivotal tool in shaping our daily choices. Traditional systems face many challenges in providing users with accurate recommendations, especially when there is limited data. Furthermore, these systems fail to provide users with explanations for the recommendations made which is crucial in cultivating trust and transparency. In light of the recent focus on Large Language Models (LLMs), this work proposes a novel framework called LightGCNxGPT that improves recommender systems by employing effective methods such as neighbourhood aggregation and user and item refinement. The LLM based paradigm proposed leverages upon the power of GPT, a popular LLM, to enhance the recommendations made by the state-of-the-art LightGCN model through innovative techniques, namely (i) User Information Refinement (ii) Item Noise Filtering (iii) GPT-Based Explanation Generation. Furthermore, theoretical analysis is provided to support the rationale behind the work and chosen methodology. The experimental results evaluated on a benchmark dataset showcases that the LightGCNxGPT model demonstrates superior performance over current state-of-the-art models.
|
|---|