Leveraging User Profiling in Click-through Rate Prediction Based on Zhihu Data
Yueqi Sun, Bin Guo, Zhimin Li, Jiahui Cheng, Liang Wang, Zhiwen Yu · 2019
With the advent of the Web 2.0 era, the prediction of Click-through Rate (CTR) has been essential to improve the user experience and loyalty for the newly emerged industry, Content Marketing. Additionally, an incisive understanding of online users is not only vital for many scientific disciplines, but also plays an important role in providing personalized products and recommendation services. In this paper, we propose a Profile-CTR model, which leverages user profiles and historical behavior data to predict CTR of certain items on Zhihu, a popular social Q&A platform. Specifically, we predict the user profiles, which include gender and occupation, using a CNN-based model on their textual data. Then, the user profiles and historical behavior records are applied to DeepFM simultaneously to predict CTR. We evaluate our method with extensive experiments and the result reveals that our approaches outperform baselines, showing that combining the user profiles with the historical behavior records can significantly improve the performance of the CTR prediction in the recommendation system.