Loading...

Analysis and Modeling of User Behavior over Social Media

Khodadadi, Ali | 2019

2464 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 51993 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza
  7. Abstract:
  8. Nowadays many of us spend a big part of our daily times on social media.One of the most important research problems in social media analysis is how to engage users. The trace of user activity over these websites is a valuable resource for user understanding and engagement, but this data is very huge and unstructured. An approach to deal with this problem is user behavior modeling. In this process, first a behavioral model is considered for users, then using the activity data and the behavioral model, some parameters are learned. Finally, using the learned parameters, a user profile is constructed for each user. This profile can be used for user engagement and many other applications. Generally, we can consider 3 type of factors in the behavioral model. Contentual, Contextual, and Engagement factors. In the literature,most of the works only considered the content factors and less attention is paid to contextual and engagement factors.In this research we concentrate on user behavior modeling, putting more emphasis on contextual and engagement factors. Indeed, we want to enrich the behavior modeling, considering both contextual and engagement factors. To this end, we focus on two important categories of research problems. In the first category, we concentrate on modeling user engagement over social media services. First we model user behavior in presence of badges. Recently, many social media sites incorporate badges to engage users over their sites. We want to consider the effect of badges in the user behavior model. To this end, we introduce a model based on a continuous-time point process which considers the effect of badges on the intensity of user activities. We also propose a novel inference algorithm for model parameters based on Variational EM algorithm. Second, we investigate the problem of user churn prediction.Previous works on churn prediction converted the problem in to a binary classification problem. We changed the problem to a return time prediction one and proposed a temporal point process to model user return times to the service. We combined the point processes and variational recurrent neural networks to improve the expressive power of the model. We also proposed an inference algorithm based on Variational inference and used the back propagation through time to learn the model parameters.At the second category, we concentrate on modeling the content of user activities on online services. At first, we investigate the problem of diffusion over social networks. We propose a continuous-time model for user activity over social networks. Since the type of user activity is influenced by his friends, our model is capable to model the diffusion process. The proposed method is based on a temporal point process which is called Hawkes process, and do not suffer from the unrealistic assumption of cascades independence which is used in many previous works. In the second problem, we concentrate on continuous time recommender systems. We extend the standard poisson factorization framework to the recurrent version and call it recurrent poisson factorization. We propose a family of models that are able to model the impact of social network, item-item interaction and heterogeneity in users’ interests in the behavior model. At the end, we proposed a variation inference algorithm to infer the model parameters. Finally, we evaluated the performance of proposed methods on synthetic and real datasets using different performance metrics. The results demonstrate the superior performance of proposed methods over state of the arts
  9. Keywords:
  10. User Behavior Modeling ; User Engagement ; Badge ; Social Media ; Temporal Point Process ; Constitutive Model

 Digital Object List

 Bookmark

...see more