A Novel K Means Biclustering Fusion Based Collaborative Recommender System

Abstract

Collaborative Filtering (CF) examines a consumer's interests and delivers automatic and personalized suggestions for purchasing a variety of products. Sparsity, on the other hand, is one of the approach's primary flaws. This difficulty is inherent in the system due to an ever-increasing quantity of people and things. To address the problem of sparsity, many existing techniques have been given. The user-item rating matrix can only provide minimal information to estimate unknown evaluations in both user-based and item-based instances. In scarce data contexts, they are ineffective. In sparse data situations, many clustering-based methods are useless. In sparse situations, traditional similarity measurements like cosine, pearson correlation, and jaccard similarity are ineffective. Although the system is able to analyze similarity in this scenario, there is a chance that the similarity is unreliable due to insufficient information processed. As the accuracy of prediction drops, this has an impact on the performance of a recommender system. As a result, neighbourhood formation is an important phase in collaborative filtering. As a result, a neighbourhood approach that can perform well in a sparse environment is required. A K Means biclustering fusion based technique is proposed to mitigate the sparsity problem by fusing item-based CF with user-based CF. A Mean absolute difference measure is used to find a neighbouring bicluster that has a significant partial similarity with the active user's preferences, which supports the algorithm in locating quality neighbours. The quality of the neighbours improves the accuracy of your recommendations.