Social influence does matter: User action prediction for in-feed advertising.

• Main Authors: WANG, Hongyang, MENG, Qingfei, FAN, Ju, LI, Yuchen, CUI, Laizhong, ZHAO, Xiaoman, PENG, Chong, CHEN, Gong Chen, DU, Xiaoyong
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2020
• Subjects: Numerical Analysis and Scientific Computing
• Online Access: https://ink.library.smu.edu.sg/sis_research/6112; https://ink.library.smu.edu.sg/context/sis_research/article/7115/viewcontent/5357_Article_Text_8582_1_10_20200508__1_.pdf
• Institution: Singapore Management University

Social in-feed advertising delivers ads that seamlessly fit insidea user’s feed, and allows users to engage in social actions(likes or comments) with the ads. Many businesses payhigher attention to “engagement marketing” that maximizessocial actions, as social actions can effectively promote brandawareness. This paper studies social action prediction for infeedadvertising. Most existing works overlook the social influenceas a user’s action may be affected by her friends’actions. This paper introduces an end-to-end approach thatleverages social influence for action prediction, and focuseson addressing the high sparsity challenge for in-feed ads. Wepropose to learn influence structure that models who tendsto be influenced. We extract a subgraph with the near neighborsa user interacts with, and learn topological features ofthe subgraph by developing structure-aware graph encodingmethods.We also introduce graph attention networks to learninfluence dynamics that models how a user is influenced byneighbors’ actions.We conduct extensive experiments on realdatasets from the commercial advertising platform ofWeChatand a public dataset. The experimental results demonstratethat social influence learned by our approach can significantlyboost performance of social action prediction.