ELM based imputation-boosted proactive recommender systems

Abstract

Proactive recommender systems are intelligent systems that provide (i.e. push) pertinent recommendations to the users based on their current tasks or interests. The recommendation algorithms employed in these systems usually compute similarity score or build up a model offline using training data for producing online recommendations. As training in proactive recommender systems when the availability of items changes often and rapidly is very time consuming, existing recommendation algorithms are less effective in such application domains. To address this problem, we present a proactive recommender system that generates real time recommendations using the proposed Extreme Learning Machine based Imputation-Boosted Collaborative Filtering (ELMICF) algorithm. Extreme Learning Machine (ELM) is a machine learning algorithm which considerably reduces the time required for training the system due to the very fast learning process of ELM. It has been used in literature for numerous classification, generalization and prediction applications. ELMICF first employs an imputation technique to handle data sparseness in the input rating matrix and then uses the ELM as a classifier to predict the novel ratings. A prototype of the system has been implemented for restaurant recommendations to show the feasibility of our proposed approach. The performance of ELMICF is compared with MLP/ANN and naive based classification techniques using normalized Discounted Cumulative Gain (nDCG), average precision, training time and mean prediction time metrics.