Enhancement of frequent sequential pattern mining using co occurrence for web recommender systems

Abstract

The web recommendation systems developed using traditional collaborative filtering and content based filtering approaches suffer from certain drawbacks The content based filtering technique recommends a webpage based on the past experience upon visiting a website by the user the main drawback in this technique is that seldom a user provides correct ratings for a website even if it would help them in future The collaborative filtering technique recommends a webpage to a specific user based on the webpage preference by a similar kind of the user the similarity among the users is calculated by collecting all the information about the user log activities on the website from the webserver The main drawback of this technique is that it suffers from sparsity and scalability The sparsity drawback highlights the limit of available ratings as against the required ratings that are to be predicted Recently web recommendation systems are designed based on Web Usage Mining WUM to make decision on how to organize a web content of a website based on the recommendations Many works were dedicated towards the development of recommendation systems based on WUM which operates in two stages i Data preprocessing and ii Pattern discovery In the first stage data preprocessing or preparation of the data in the form of web logs are sourced from the web server that maintains users log details and prepares them appropriately for the application of pattern discovery algorithms During the next phase in the pattern discovery stage the data mining tools are applied to extract the patterns Using the extracted patterns the decisions for website restructuring web recommendation site modifications etc can be made