This thesis presents a framework known as User Interest Modeling and Personalization (UIMAP) which builds a model by identifying and aggregating an individual user's interest expressed through their interactions with different applications at different times. To do this, we have implemented a content consumer/producer architecture. For this thesis, Microsoft Word and PowerPoint are treated as content producer applications while a web browser is used as a content consumer application. We unobtrusively observe user interactions with these applications as well as the actual content consumed/prepared in them. The challenge is to understand the importance of each application towards the user's real interest. Based on user activity data in these applications, Multilayer Perceptron (MLP), Support Vector Machine (SVM) and Weighted K-Nearest Neighborhood (WKNN) techniques are compared in their ability to combine these kinds of heterogeneous interest indicators into a single model. Thus, each application is weighted differently based on its contributing indicators to predict the relevant content for the specific need of an individual. We found that textual content from content producer applications plays an equally important role as content from consumer applications. Implicit feedbacks from consumer applications also have a major role in user's interest. The results indicated that WKNN is preferred if feature weighting is the primary goal while SVM is the preferred choice if identifying relevant content is the main objective.