Cold-Start Product Recommendation Using Microblogging Information: Linking Social Media To E-Commerce

Abstract

The lines between online shopping and social networking have blurred in recent years. Social login allows users to access their favourite e-commerce sites by logging in with credentials from their existing account on a third-party social network like Facebook or Twitter. Additionally, customers can promote their recent purchases on microblogs by including links to the item pages on the merchant's website. To recommend products from e-commerce websites to users on social networking sites in "cold-start" situations, this paper proposes a novel approach to the under-explored problem of cross-site cold-start product recommendation. The challenge of figuring out how to use knowledge extracted from social networking sites is a major one in implementing cross-site cold-start product recommendations. To facilitate the mapping of social networking features to another feature representation for a product recommendation, we propose using users who have accounts on both social networking sites and e-commerce sites as a bridge. To be more specific, we propose using recurrent neural networks to learn feature representations for users and products (termed user embedding’s and product embedding’s, respectively) from data collected from e-commerce websites, and then employing a modified gradient boosting trees method to transform users' social networking features into user embedding’s. After acquiring user embedding’s, we develop a feature-based matrix factorization approach to cold-start product recommendations. Experimental results on a large dataset constructed from SINA WEIBO, the largest Chinese microblogging service, and JINGDONG, the largest Chinese B2C e-commerce website, confirmed the efficacy of our proposed framework.