Development of artificial intelligence based collaborative filtering recommender systems for web applications

Abstract

Electronic commerce means buying and selling of items or services over electronic systems, such as the Internet and other computer networks. It allows businesses to open up their items and services to a huge user base. This thesis proposed several models for collaborative filtering to reduce the complexity for the users by sifting through very large sets of information and selecting the relevant information. In this thesis, a shopping website system on ASP.NET platform is developed, this makes intelligent searching and personalized recommendation for laptop purchasing using the dataset from laptoplogic website. This thesis proposed data clustering methods to handle the large number of data in recommender systems. In this thesis, an intelligent CF recommender system (ICRS) is proposed. During offline, the system uses cluster based smoothing approach, where clustering is performed by using k-means clustering and smoothing is performed by using average deviations rating of all the users, who rated that item. During online, CF based recommendation is provided. This thesis proposed two different approaches using Radial Basis Function Network (RBFN) namely RBFN_CF and RBFN_KFCM. During offline, both RBFN_CF and RBFN_KFCM use RBFN for smoothing and for clustering, RBFN_CF uses Pearson correlation clustering method and RBFN_KFCM uses kernel fuzzy c-means (KFCM) method. During online recommendation, RBFN_CF uses CF based approach and RBFN_KFCM uses KFCM based approach. In this thesis, a novel approach using probabilistic neural network (PNN) is proposed. Probabilistic neural network (PNN) is used to calculate the trust between users based on rating matrix. A comparison is made with Pearson based CF system and cosine based CF system. Also comparison is made with SVD and PNN based prediction methods. For comparison the system uses accuracy, decision-support measures and computational time as performance measure. Based on the comparisons, it is noted that the made with the proposed systems outperforms existing methods. newline newline