الفهرس 
Title : Dynamic modeling of users in social networks
ContentsOnly 14 pages are availabe for public view
 |  | from | 75 |  |  |
| | | from | 75 | | |
Abstract
Social Networks are popular platforms for users to express themselves, facilitate interactions, and share knowledge. Today, users in social networks have personalized profiles that contain both static attribute and dynamic attributes representing their demographic attribute, interest, activity, and behavior. Several models emerged that analyze dynamic profiles attributes that represent user interest over time such as published content, and location check-ins in order to build dynamic models. These models are used in different analysis tasks in social networks such as friend recommendation, link prediction and location prediction. However, these models suffer from the two main drawbacks. The first one is that multiple user models rely on a static snapshot of attributes which do not reflect the change in user interest and behavior over time. The second drawback is that multiple models don{u201F}t consider the semantic relation that may exist between different user{u201F}s attributes. This thesis proposes a dynamic model that captures user{u201F}s dynamic interest and behavior from social networks. This model has been used within a comprehensive framework that utilizes user{u201F}s topical interests and locations to represent user{u201F}s interest and behavior respectively. This model also utilized the semantic relationship between user published contents and user{u201F}s locations to integrate multiple aspects of the user{u201F}s behavior. The proposed model had been applied on real twitter dataset of 1452 users representing their interaction from 2011 till 2014. In order to approve the effectiveness of the proposed model, it has been applied in two different social networks analysis tasks: similarity engine and location prediction. Based on the scope of each of those tasks, a sample of the gathered dataset was used. Experiments showed that the proposed dynamic model outperformed multiple standard model that utilize static attributes and the models that don{u201F}t consider the semantic features.