Insights: From Social Psychology to computation User Modeling
Abstract:
In the field of social psychology a huge amount of research has been conducted to understand human behaviors by studying their physical space and belongings. Inspired from these fruitful findings, can we design corresponding computational models to characterize diverse online user behaviors by exploring their posted comments and their established connections? Can we further build a unified model for each user by integration of different types of behavioral data?
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 aim to understand users by building computation user models automatically, thereby to save time and efforts. And the principles of social psychology serve as good references for building such computation models.
In this dissertation, we develop new techniques to model online user behaviors based on user-generated data to better understand user preferences and intents. We get inspired from social psychology principles in user behavior modeling and these developed computation models provide alternatives to explain human behaviors in the physical world. More specifically, we focus on two challenges: (1) model users’ diverse ways of expressing attitudes or opinions; (2) build unified user models by integration of different modalities of user-generated data.
To tackle the challenge of capturing users’ diverse opinions, we borrow the concept of social norms evolution to achieve personalized sentiment classification. By realizing the consistency existing in users’ attitudes, we further perform clustered model adaptation to better calibrate such opinion coherence. To understand users from a comprehensive perspective, we utilize different modalities of user-generated data to form multiple companion learning tasks. And each individual user is modeled as a mixture over these companion learning tasks to realize the behavior heterogeneity. To better characterize the correlation among different modalities of user-generated data, joint learning of different embeddings, together with their relatedness are performed, in order to achieve a comprehensive understanding of user intents and preferences. This dissertation borrows principles from social psychology to better perform effective computation user modeling. It also provides a foundation for making user behavior modeling useful for many other applications as well as offers new directions for designing more powerful and flexible models.
Committee:
- Hongning Wang (Advisors)
- Mary Lou Soffa (Committee Chair)
- Yanjun Qi
- Quanquan Gu (UCLA)
- Nikolaos Sidiropoulos (Minor Representative)