Vietoris-Rips Complex: A New Direction for Cross-Domain Cold-Start Recommendation
Ajay Krishna Vajjala, Dipak Falgun Meher, Shrunal Pothagoni, Ziwei Zhu, David S. Rosenblum · Society for Industrial and Applied Mathematics eBooks · 2024
Cross-domain recommendation (CDR) has emerged as a promising solution to alleviating the cold-start problem by leveraging information from an auxiliary source domain to generate recommendations in a target domain. Most CDR techniques fall into a category known as bridge-based methods, but many of them fail to account for the structure and rating behavior of target users from the source domain into the recommendation process. Therefore, we present a novel framework called Vietoris-Rips Complex for Cross-Domain Recommendation (VRCDR), which utilizes the Vietoris-Rips Complex (a technique from computational geometry) to understand the underlying structure in user behavior from the source domain, and includes the learned information into recommendations in the target domain to make the recommendations more personalized to users' niche preferences. Extensive experiments on large, real-world datasets demonstrate that VRCDR consistently improves recommendations compared to state-of-the-art bridge-based CDR methods.