Predictive Behavior Modeling Through Web Graphs: Enhancing Next Page Prediction Using Dynamic Link Repository

The goal of this experiment is to improve user interaction predictability for better search interface design and efficient search intent satisfaction. In this work, we optimize the Web Usage Mining (WUM) process reconstructing a web session for predicting navigation paths specifically on fragile websites. Our research is based on the most current algorithmic approach called the Complete Session Reconstruction Algorithm (CSRA), which outperforms state-of-the-art session reconstruction algorithms for predicting the possible single next pages. In this work, we extend the newest session reconstruction problem-solving algorithm and improve its next page prediction performance by extending the web topology taking into account dynamic targets. We implement a Bayesian network, integrate and evaluate it in our evaluation framework for the prediction of the next visited pages. The presented experiments show how our approach outperforms the compared next page prediction state-of-the-art methods.