An Effective Deep Learning Approach for Personalized Advertisement Service Recommend

Abstract

Personalized advertisement service recommend represents one of the largest scales and most sophisticated industrial recommendation systems. The key challenge of advertisement recommend is to find relevant users for a specific advertisement. Traditional recommendation approaches suffered from selecting effective features on both advertisements and user profiles. In this paper, we studied a deep neural network to learning effective representation on advertisement recommend. Specifically, we connected one user profile features and one advertisement features together as one input vector, and then employed deep neural network to predict the whether the user is relevant to the advertisement. We conducted our experiments on Tencent advertising competition data set, and the experiment results show that (1) the DNN method obtained better predictive performance than traditional approaches; (2) the DNN method with 6 layers hidden nodes achieved best performance; (3) the single-perception method overcame the multi-perceptron method on Advertisement Recommend.