Web navigation prediction based on dynamic threshold heuristics

Abstract

Web Navigation Prediction is a popular area of research as users on the web are continuously growing. Capturing users’ behavior provides insights about the user demands, accessibility patterns, and inconsistencies in the website design. Several navigation models have been developed in the past. Most of the models rely upon fixed threshold mechanism for prediction, which is inadequate as they require multiple computational steps which not necessarily give the best possible value. To combat this issue, we propose two models based on dynamic thresholds: All-Kth Modified Markov Model based on Geometric threshold (KMMMG) and (b) All-Kth Modified Markov Model based on Branching Factor Threshold (KMMMBF). Our experiments show that longer navigations are rare and are highly co-related. However, smaller navigations are more and are less co-related. Dynamic threshold models produce more optimum predictions as compare to fixed threshold models. KMMMBF performs the best and achieves the highest prediction accuracy, 78.53%, 91.32%, and 61.37% on CTI, MSWEB, and BMS dataset, respectively.

Keywords

• Markov
• Web
• KMMM
• Threshold
• Geometric
• Branching factor