Solving cold start in news recommendations: a RippleNet-based system for large scale media outlet

A new scalable recommender system based on RippleNet has been developed for Onet.pl, one of Poland's largest online media platforms. This system addresses the cold-start problem commonly encountered in news recommendations, particularly for newly published content that lacks user interaction data. By employing content-based item embeddings, the system can effectively score unseen items, improving recommendation quality. The approach leverages RippleNet's capabilities to propagate user preferences through related content, enhancing the relevance of recommendations. Evaluations indicate that this solution is effective in overcoming the cold-start challenge, enabling better personalization for users. This development represents a significant advancement in large-scale media recommendation systems, particularly in handling fresh content. The innovation aligns with ongoing efforts to improve user engagement through more accurate and timely news delivery.