Towards a hybrid approach for a predictive modeling of user navigational behaviors: State of the art

Internet is a digital network which is providing a great deal of heterogeneous and divided Web resources, the scope of which is continuously increasing. Faced with a mass of information, the user finds it very difficult to find suitable and pertinent resources that satisfy his/her needs in a quite reasonable amount of time. Due to this mixture of noised and heterogeneous resources, learning the regularities in navigational paths and a prediction for modeling user behavior seem to be necessary. In this paper, we will present a dynamic and hybrid approach of the users behavior modeling based on the extraction and analysis of implicit usage traces. Our aim is to optimize the predictive modeling of the user behaviors online by combining two complementary approaches. Derived from the Web Usage Mining field (WUM), the first approach consists in Sequential Patterns Mining (SPM). The second approach stems from the probabilistic model: Conditional Random Fields (CRFs). Therefore, our challenge is to show that the application of the CRFs on the patterns (sequence of resources) does optimize the metric performance of the SPM approach in terms of accuracy, coverage and complexity.