Indian Journal of Science and Technology

Abstract

Background/Objectives: The website is composed of permanent and temporary pages. Deriving a prediction model which considers the dynamic pages generated on the website requires to consider new aspects. Methods/Statistical Analysis: We adopt supervised learning models as they give better prediction results for new input data. After reading the log files and applying preprocessing, we build the user navigation patterns and then apply the prediction of pages. The main parameter on which we have modified is the time stamp. In earlier approaches the time stamp was divided into day, month and year and based on the timestamp granule selected the prediction model was formed. In our work we consider the same granule with introduction to new timestamp namely event. Findings: Markov model are very good in predicting the pages for n length, but the model doesn’t focus on temporal aspect for prediction. Temporal n-gram model covers the temporal aspect of prediction by forming the granules of time. This model gives good accuracy in predicting pages that are permanent for any given website, but doesn’t tend to be good for pages that are temporary in nature. Our model focuses on temporal aspect for both types of pages by creating an event based temporal n-gram model. Event means creating a special named interval for which the pages are made available on the website. This means that after the interval specified in the event the page will be no more visible on the website. The pages are predicted based on the nature of pages, we form broadly two types of nature of pages 1. Regular for permanent pages and 2. Event for temporary pages. By introducing this temporal aspect the prediction algorithm considers the specified interval only for the event specific pages, after the interval is over the pages are not considered for prediction. Application/Improvements: Specifying events help to derive better accuracy in prediction when we consider permanent and temporary pages, as we predict the pages based on the condition whether they are regular or event based pages.
Keywords: Classification Algorithms, Conditional Probability, Event Based Granule Model, Naive Baysian, Prediction Model