Tag Based Recommender System Using Pareto Principle

Abstract

Recommender systems help users to cope up with large amount of data. Today large amount of data comes in very high speed and in different formats. This huge amount of data baffles internet users when they are searching for something, so this data is called Big data. Big data has 4 V model where 4 V stands for Volume, Velocity, Variety and Veracity. Recommender system deals with the Veracity aspect of Big data. Veracity denotes data has noises, abnormality and irrelevant patterns. Recommender system finds relevant data of user’s interest from irrelevant patterns. Recommender systems provide personalized and non personalized recommendations to interested users. Recommender systems can be categorized into three ways according to the evolution of web. In Web 1.0 applications, traditional or rating based recommender systems came into existence. In web 2.0 application, social tagging information has been incorporated into recommender systems to improve the performance of traditional recommender systems, called tag based recommender systems. In web 3.0 applications, internet of things has been used for generating recommendations. In this thesis, Pareto principle has been applied on social tagging dataset for providing good quality item and tag recommendations to users. A pre filtering step based on Pareto principle has been added to the traditional collaborative tagging system for removing irrelevant tags from k-Nearest Neighbour (kNN) selection process. Experiments have been performed on Movie lens datasets and experimental results show good quality item and tag recommendations.