1. VIKRAM MADITHAM - Research Scholar, Department of CSE, JNTUA University, Ananthapuramu, Andhra Pradesh, India.
2. SUDHAKAR REDDY N - Department of CSE, SV College of Engineering, Tirupati, Andhra Pradesh, India.
3. MADHAVI KASA - Department of CSE, JNTUA University, Ananthapuramu, Andhra Pradesh, India.
Recommender systems help the users filter the preferences when large volumes of data are present. However, recommender systems fail to render quality recommendations due to data sparsity problem where users are reluctant to accord feedback. Collaborative filtering technique is one of the quintessential techniques used in recommender systems to mitigate the data sparsity problem. It predicts users’ ratings to convert sparse into dense matrix. Similarity models drive collaborative filtering to acquaint the prophecy of users’ rating. The existing similarity models work on co-rated items which is not adept to ascertain nearest neighbors in the data. In this paper, we have proposed a new similarity model based on the consideration of user ratings as vectors. The new Resultant vector similarity model (RV Sim) is incorporated in the collaborative filtering algorithm to validate the outcomes using the popular data set Movie Lens. Firstly, resultant vector similarity model is derived based on the resultant vector mathematical model in which vector directions are considered to determine the similarity. Finally, the RV Sim based collaborative filtering algorithm is applied to generate the efficient recommendations. It is imminent that proposed similarity model proved to be superior when it is compared with the existing similarity models (Rjaccard,Cosine, Pearson, Mean square distance etc.) in line with the prominent evaluation metrics i.e. Precision, Recall, F1- Score in the domain of recommender systems. Upon increasing the nearest neighbors 5NN, 20NN, 50NN and 100NN, the RV Sim model has got F1-Score with 0.726, 0.74, 0.753 and 0.763 with considerable improvement.
Recommender systems, Resultant vector, Collaborative filtering, Similarity model.