Leveraging Behavioral Heterogeneity Across Markets for Cross-Market Training of Recommender Systems

Abstract

Modern recommender systems are optimised to deliver personalised recommendations to millions of users spread across different geographic regions exhibiting various forms of heterogeneity, including behavioural-, content-and trend specific heterogeneity. System designers often face the challenge of deploying either a single global model across all markets, or developing custom models for different markets. In this work, we focus on the specific case of music recommendation across 21 different markets, and consider the trade-off between developing global model versus market specific models. We begin by investigating behavioural differences across users of different markets, and motivate the need for considering market as an important factor when training models. We propose five different training styles, covering the entire spectrum of models: from a single global model to individual market specific models, and in the process, propose ways to identify and leverage users abroad, and data from similar markets. Based on a large scale experimentation with data for 100M users across 21 different markets, we present insights which highlight that markets play a key role, and describe models that leverage market specific data in serving personalised recommendations.

Cross-Market Product Recommendation

“…While there exists much work on domain adaptation, marketadaptation is relatively unstudied. CMR has attracted attention in music recommendation [11,43] where Ferwerda et al [11] analyze and study music diversity across countries and propose to use country-based diversity measurements for system evaluation. Roitero et al [43] studied user behavior in 21 different markets on Spotify and highlight the need for market-specific algorithms, as opposed to a global algorithm.…”

“…CMR has attracted attention in music recommendation [11,43] where Ferwerda et al [11] analyze and study music diversity across countries and propose to use country-based diversity measurements for system evaluation. Roitero et al [43] studied user behavior in 21 different markets on Spotify and highlight the need for market-specific algorithms, as opposed to a global algorithm. We take one step further in this direction by expanding our study to various item categories in e-commerce where users purchase items (rather than having a monthly subscription) and express their opinion and experience with item in the form of ratings and reviews.…”

“…The significance of Cross-Market Recommendation (CMR) has been pointed out in the literature [43]. However, small progress has been made in this area, mainly due to a lack of experimental data.…”

Aligning AI Optimization to Community Well-Being

“…Each of these communities may have different concerns at any given moment. Incorporating social network analysis or countryspecific data can improve the performance of recommender systems as measured by traditional relevance metrics (Chen et al 2018;Roitero et al 2020) but the question of how a recommender system impacts pre-existing communities, e.g. a city, has not been explored.…”

Multi-stage Ensemble Model for Cross-market Recommendation

“…These catalogs often overlap between different markets, and e-commerce companies have to deal with the recommendation of similar sets of items in different scenarios. This allows sharing both experience and information across markets, with the risk to spread market-specific biases and impose trends of data-rich markets to others [1,11]. How to effectively exploit information from different markets to improve recommendation quality remains an open challenge, which is the focus of the WSDM Cup 2022 competition on cross-market recommendation.…”