Article feed recommendations using position-aware deep cross network

Abstract

The presence of social media platforms has recently transformed the digital landscape, with an ever-increasing user base. As these platforms become central to daily life, the need for recommendation systems that genuinely cater to individual preferences has never been more paramount. While current recommendation algorithms excel at curating content based on user interests, they often overlook inherent biases, notably the positional bias where users engage with content due to its placement rather than its inherent relevance. This oversight is particularly evident in every social media recommendation system. Addressing this challenge, we proposed a position-aware methodology within the Deep and Cross framework, aptly termed 'Position-Aware DCN.' By explicitly accounting for positional preferences, our proposed model aims to provide more genuine, unbiased recommendations, ensuring that users are presented with content that aligns with their interests and is not just influenced by its position in the feed. Evaluations conducted on Thai social media datasets reveal that our proposed model offers a marked improvement over traditional recommendation systems, underscoring its potential to foster a more user-centric digital experience. The author also implements the proposed model as an application programming interface (API) in an online deployment format by showcasing its functionality and seamless integration into the front-end web app.