Abstract:
Recommender systems are intellectual systems of information filtering that makes relevant suggestions for customers. Main purpose by generating relevant information for customers to improve their experience. Recommendations systems mostly operate in challenging environment. A Recommender system which gives fast, accurate recommendations and good experience to users is more attractive and develop more interests in users. Because as the strength of internet users and items are growing day by day many challenges are faced by recommender systems. One of the challenges is data sparsity. Highly sparse data gives decreased accuracy of predictions. High quality predictions are dependent upon how well the recommender system address its challenges. Thus there is this need to develop a model which should have improved performance under relatively highly sparse data. This thesis work purposes CF based one such technique. Out of all the obligatory steps of collaborative filtering one is making neighbours which are determined using dimensionality reduction based Spectral Co-clustering technique. Memory based algorithm of CF are divided into user based and item based, Individually these two technique provide fragmented information to make predictions for new ratings. In this research Confidence based weighted fusion method (CBWF) is proposed to merge the rating predictions from both memory based filtering techniques that is User based CF (UbCF) and Item based (IbCF).Along with confidence a parameter is used. Previously was varied based on the datasets. In this thesis parameter is controlled and made dependent upon the correlation characteristics of individual users and items based on their state. Spectral-co-clustering overcomes the sparseness of dataset and limitation of scalability. While fusion of CbCF and IbCF in confidence based weighted sum improves the prediction accuracy of system. Finally predictive accuracy metrics (RMSE, MAE) and decision support metrics( Recall, Precision, F1-score) are used to compare results for proposed technique with conventional techniques, one dimensional clustering ,two dimensional clustering and HCF techniques to show the improvement that our proposal has made.
