INTEGRATING SOCIAL MEDIA WITH E-COMMERCE: COLD-START PRODUCT RECOMMENDATION USING MICROBLOGGING DATA

Abstract

In recent years, the boundaries between online shopping and social networking
have become increasingly blurred. Through social login, users can conveniently access
their preferred e-commerce platforms using credentials from third-party social networks
like Facebook or Twitter. Moreover, customers often share their recent purchases on
microblogs, including links to the corresponding product pages on the merchant's website.
Addressing the challenge of recommending products from e-commerce platforms to users
on social networking sites in "cold-start" scenarios, this paper proposes an innovative
solution to the relatively unexplored issue of cross-site cold-start product recommendation.
The primary hurdle in implementing cross-site cold-start product recommendations lies in
leveraging insights gleaned from social networking sites effectively. To overcome this
obstacle, we suggest leveraging users who maintain accounts on both social networking
platforms and e-commerce sites as intermediaries. Specifically, we propose employing
recurrent neural networks to learn feature representations for users and products (referred
to as user embeddings and product embeddings, respectively) using data collected from ecommerce
platforms. Subsequently, we utilize a modified gradient boosting trees method to
translate users' social networking features into user embeddings. With these user
embeddings in hand, we develop a feature-based matrix factorization approach for coldstart
product recommendations. Experimental evaluations conducted on a substantial
dataset compiled from SINA WEIBO, the largest Chinese microblogging service, and
JINGDONG, the largest Chinese B2C e-commerce website, validate the effectiveness of
our proposed framework.