IMETA-GNN: Meta Learning-Based Cold Start Optimization for Recommendation System

Abstract

Recommendation systems are becoming essential components of contemporary online goods and services, and they significantly affect customer satisfaction. Recommendation systems are designed to empower customers in their decision-making process by providing personalized recommendations. Meta-learning has proven to be effective in addressing user cold-start problems in recommendation systems. Many meta-learning-based recommendation systems designed for the cold-start problem are gradient-based. The existing frameworks require optimization techniques to maximize the potential of hyper networks, enabling them to adapt and generate compatible parameters for efficient learning as meta-learners. However, these frameworks often lack contextual knowledge from users to provide suitable initial guidelines in the recommendation network for new users. We propose a meta-learning-based framework that improves cold-start recommendation accuracy by incorporating user behavior and preferences. Evaluations on MovieLens 100K and DBook datasets show our IMETA-GNN model outperforms state-of-the-art baselines, achieving accuracies of 82% and 75%, respectively. Performance evaluation demonstrates that our proposed method outperforms several state-of-the-art metalearning recommendation systems in addressing the user cold-start conundrum.