Prediction of Consumer Preference through Bayesian Classification and Generating Profile

Abstract

Collaborative filtering system overlooks the fact that most consumers do not rate a preference; because of this oversight the consumer-product matrix shows great sparsity. A memory-based filtering system has storage problems and hence proves inefficient when applied on a large scale where tens of thousands of consumers and thousands of products are represented in the matrix. Clustering consumer into groups based on the web documents they have retrieved/fetched allows accurate recommendations of new web documents through solving the problem of sparsity. A variety of algorithms have previously been reported in the literature and their promising performance has been evaluated empirically. We identify the shortcomings of current algorithms for clustering consumer and propose the use of Naïve Bayes classifier to classify consumer into group. To classify consumer into group, this paper uses the association word mining method with weighted word that reflects not only the preference rating of products but also information on them. The data expressed by the mined features are not expressed as a string of data, but as an association word vector. Then, collaborative consumer’s profile is generated based on the extracted features. Naïve Bayes classifier classifies consumer into group based on association words in collaborative consumer’s profile. As a result, the dimension of the consumer-product matrix is decreased. We evaluate our method on database of consumer ratings for special computer study and show that it significantly outperforms previously proposed methods.