Information Filtering System Based on Clustering Approach



The level of web information has been increased day by day scheduled to fast development of internet. Now-a-days people make their decision based on the available information from the internet. But the problem is how the people effectively choose or filter the useful information from the gigantic amount of information. This problem is referred as information overload.

Recommendation System is a supportive tool to solve the information overload problem. It is part of information filtering system used to recommend the user based on their own interest, neighborhood similarity and earlier background. Collaborative Filtering is one of the popular techniques trusted recommendation system.

Every advice system should ensure personal privacy for both user's neighbour and their data. To overcome the scalability and model reconstruction problem, a vitality graph established private neighborhood advice system is proposed to ensure the user's personal privacy. First, the compressed network is constructed and then your feature collection is extracted from the compressed network using transformed data. The info is transformed using hybrid transformation fuses principal part examination and rotation change to protect users' level of privacy with accurate tips. Finally the item to be recommended is expected which achieve better performance than the existing strategy. MovieLens Dataset can be used to evaluate this technique.


Recommendation System is one of the info filtering system which provides valuable information to the users by filtering the information regarding to user's interest. Traditional solutions of recommendation systems are collaborative filtering, content established filtering and cross Approach. Content Based Filtering (CBF) methodology predicts the advice predicated on the rating distributed by the user for the similar items in previous background. Collaborative Filtering (CF) recommends the user predicated on rating of this item by similar users. Hybrid approach combines both the approaches. All of the methods have their own edge and drawback.

CF mainly classified as memory founded CF and model centered CF. Memory based CF first calculate the similarities between your requested individual and all the user to get the neighbors then determine the prediction based on identified neighbors ranking pattern. Model established method first built a model predicated on the desire of the user. Main aim of the recommender system is to reduce the prediction problem. The main issues in CF recommender system are scalability, sparsity and privateness.

  • Scalability: Large number of users and items in the network resulted in the increase in the computational complexity of the system. In E-commerce, scalability plays a important issue because it contain large numbers of users.
  • Sparsity: All the users don't show their interest to rate all the items they work together private, that will lead to data sparseness in the system. This will not give exact recommendation to the seekers.
  • Cold Start: Lack of information for new items and users in suggestion system will contributes to unpredictable items in the machine.
  • Privacy: Users might provide false information inorder to protect their personal information. This causes inaccurate recommendation.

The suggested work mainly focuses on two fundamental issues in CF namely scalability and level of privacy. The first challenge is how to improve the scalability of CF, because these systems should search the entire user for locating the neighbors. The next problem is how to safeguard the average person users privateness while prediction. Both an issues lead to poor performance of the machine. So the important problem is to handle both a predicament properly for better performance.


Recommendation system helps the people to get exact information based on neighbors' pattern. Remarkable growth in e-commerce site makes the web vendors to develop their sales and profits. They use this technique which suggests product to users' by their friends and neighbors' choice about that. Scalability concern in RS due mainly to enormous progress in users tends to decline in precision of prediction on recommendation. Clustering procedure reduces scalability problem by grouping the similar