A Graph-Based Push Service Platform

Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, Xiuqiang He

Abstract: Learning users’ preference and making recommendations is critical in information-exploded environment. There are two typical modes for recommendation, known as pull and push, which respectively account for recommendation inside and outside the item market. While previously most recommender systems adopt only pull-mode, push-mode becomes popular in today’s mobile environment. This paper presents a push recommendation platform successfully deployed for Huawei App Store, which has reached 0.3 billion registered users and 1.2 million Apps by 2016. Among the various modules in developing this push platform, we recognized the task of target user group discovery to be most essential in terms of CTR. We explored various algorithmic choices for mining target user group, and highlighted one based on recent advance in graph mining, the Partially Absorbing Random Walk [13], which leads to substantial improvement for our push recommendation, compared to the state-of-the-art including the popular PageRank. We also covered our practice in deploying our push platform in both single server and distributed cluster.