Design and Evaluation of Cross-Domain Recommender Systems

Abstract

The proliferation of e-commerce sites and online social media has allowed users to provide preference feedback and maintain profiles in multiple systems, reflecting a spectrum of their tastes and interests. Leveraging all the user preferences available in several systems or domains may be beneficial for generating more encompassing user models and better recommendations, e.g., through mitigating the cold-start and sparsity problems, or enabling cross-selling recommendations for items from multiple domains. Cross-domain recommender systems, thus, aim to enhance recommendations in a target domain by exploiting knowledge from source domains. In this chapter, we formalize the cross-domain recommendation problem, unify the perspectives from which it has been addressed, and analytically categorize and describe various recommendation techniques, from the simple legacy ones to the sophisticated ones based on deep-learning.Authors Suppressed Due to Excessive Length Zalando recommends clothing, and Spotify recommends songs and playlists. These domain-specific systems have been successfully deployed by numerous websites, and the single-domain recommendation functionality is not perceived as a limitation, but rather pitched as a focus on a certain market segment.Nonetheless, large e-commerce sites like Amazon and Alibaba often store user feedback for items across multiple domains, and social media users often express their tastes and interests for a variety of topics. It may, therefore, be beneficial to leverage all the available user data provided in various systems and domains, in order to generate more encompassing user models and better recommendations. Instead of treating each domain (e.g., movies, books and music) independently, knowledge acquired in a source domain could be transferred to and exploited in another target domain. The research challenge of transferring knowledge and the business potential of delivering recommendations spanning multiple domains, have triggered an increasing interest in cross-domain recommendations.Consider two motivating use cases for cross-domain recommendations. The first refers to the well known cold-start problem, which makes it difficult to generate recommendations due to the lack of sufficient information about users or items. In a cross-domain setting, a recommender may draw on information acquired from other domains to alleviate such a problem, e.g., user's favorite movie genres may be derived from her favorite book genres. The second refers to the generation of personalized cross-selling or bundle recommendations for items from multiple domains, e.g., a movie accompanied by a music album similar to the movie soundtrack. This recommendation may be informed by the user's movie tastes, extracted from rating correlations within a joined movie-music rating matrix.These use cases are underpinned by an intuitive assumption that there are correspondences between user and item profiles in the source and target domains. This assumption has been validated in several marketing, behavioral, a...