Insights: From Social Psychology to Computational User Modeling
Abstract:
Can we create computational models to characterize diverse online user behaviors? If so, can we build a unified and interpretable model for each online user? As we can understand people by studying their physical space and belongings, we are also able to investigate users by studying their online connections, postings and behaviors. The advent of participatory web has created massive amounts of user-generated data, which enables the study of online user attributes and behaviors. Traditional social psychology studies commonly conduct surveys and experiments to collect user data in order to infer attributes of individuals, which are expensive and time-consuming. In contrast, we propose to understand users by building computational user models automatically, thereby to save time and efforts. And the principles of social psychology can still serve as good references for building such computational models. The goal of my proposed research is to build flexible and explainable user models which can provide alternative to explain human behaviors in the physical world. The proposed research gets inspired from social psychology principles and develops new techniques to model online user behaviors based on user-generated data, thus to better understand their intents and preferences. Preliminary results indicate the approach can accurately capture user intents and provide proper explanations. We believe, without exaggeration, the proposed method can be a replacement to traditional methods for social psychologists in the near future.
To achieve the modeling of online user behaviors, we focus on two challenges: (1) model users’ diverse ways of expressing attitudes; (2) build unified and interpretable user models. To address the first challenge, we get inspired by the evolution of social norms and perform personalized sentiment classification via shared model adaptation for each user over time. Suggested by Social Comparison Theory, we further exploit the clustering property of users’ ways of expressing opinions. In order to achieve a comprehensive understanding of users to address the second challenge, we propose to model each user’s unified behavior patterns by taking a holistic view of sentiment analysis and social network analysis, which is supported by Self Consistency Theory. In order to make the learned user behaviors flexible and explainable, we propose to model latent users explicitly with compact distribution representations, enabled by the large ever-growing textual corpora and social network. Therefore, each user’s diverse behavior patterns can be captured by his/her own distributed representation while the explanation of the behaviors can be achieved via his/her correlations to other users or latent semantics.
Committee Members:
Hongning Wang (Advisor), Mary Lou Soffa (Chair), Yanjun Qi, Quanquan Gu, Nikolaos Sidiropoulos (Minor Representative)