Predicting Users’ Revisitation Behaviour Based on Web Access Contextual Clusters

摘要

Most modern browsers record all previously visited web pages for future revisitation. However, not all users utilize such feature. One of the reasons is that the records are displayed at once as a single list, which may overwhelm the users. This paper proposes a predictive model to decide whether a web page will be revisited in the future based on a particular visit. The model can be used to filter web records so that only web pages that may be re-visited are presented. According to our evaluation, the model is considerably effective. It can generate 53.195% accuracy when measured with 10-fold cross-validation and 95% meaningful topic identification. Further, attributes rooted from the same website’ access frequency are the most salient ones for prediction. In addition, contextual similarities based on k-means clustering and cosine similarity (which are used for defining some attributes) are considerably effective.