Integrated Weighted PageRank algorithm with Multi-Layer Perceptron for Predicting Web User Behaviour from Streams of User Interactions

Abstract

In this paper, a robust Multi-Layer Perceptron (MLP) classifier is designed to improve the efficacy of weighted PageRank algorithm that predicts the user behaviour on online e-commerce websites. Initially, the Weighted PageRank is designed to serve this purpose but with increasing users and e-commerce website, it is essential to design a fast response mechanism that assigns the weight for PageRank to possible estimate the user behaviour and tracking their behaviour over time. To achieve this, the study uses two different mechanisms i.e. user behaviour modelling and user interaction modelling to estimate their behaviour and both the models uses MLP to find the maximum extent of weights based on their interaction and behaviour. The simulation is conducted to check the efficacy of MLP-Weighted PageRank method with other existing machine and deep learning frameworks. The simulation carried out consists of following activities: Finding a web site, Building a web map, Finding the root set, Finding the base set and Check the clickstream. The results of simulation show a substantial improvement in extracting the relevant features by breaking down the problem of data dimensionality through the process of pre-training. Most of the conventional methods reported in the study failed to do so. The results show that the MLP-Weighted PageRank achieves higher rate of classification, F-measure, sensitivity, specificity and reduced mean absolute percentage error (MAPE) than other methods.