A Proposed Movie Recommender System to Solve Sparsity, Cold Start and Diversity Problems using Clustering Algorithms

Abstract

Internet growth has triggered massive information. Thus, the need to help users solve the
information overload issue is by offering customized suggestions of services, products, information,
and products. Recommender systems (RSs) that use collaborative filtering (CF) methods are
commonly used because of their capability to construe user expectations and direct them towards
linked tools that be acceptable to their interests. These methods face some issues such as sparsity of
data, diversity and cold start. To overcome the issues faced by the RS, this paper has four proposals
to better RS achievement to make proper predictions and recommend appropriate movies. First, use of
K-Means algorithm to cluster the users in many groups according to each genre to cope with sparsity
of data and to reduce the effect of popular movies. Second, use of K-Mediods algorithm to find the
cluster that the user returned to according to all genres for increasing the diversity. Third, use of KNearest Neighbors algorithm based CF model to find neighbors in each cluster for the user in K-Means
and K-Medoids. Fourth, demographic questions have been built for handing the problem of cold start.
The evaluating of the proposed system implemented on movielens data set in two testing using some
measures and the results of these measures after implementing the algorithms show that the proposed
system has a good organization. such as precision equal to 0.92, recall equal to 0.91, MAE equal to
0.39, RMSE equal to 1.16 , diversity score equal to 8, DCG, and NDCG, for users are decreased as the
recommendation list progressed.