Enhancing stability of recommender system: An ensemble based information retrieval approach

Abstract

Collaborative filtering is popularly used for providing recommendation services. These services are provided by numerous different recommendation algorithms that are proved to be effective. Methods: The Stability of recommender systems are now becoming the interesting component in the fields of research. It becomes necessary to evaluate the consistency of the predictions to retain users trust on the system. Stability is a measure of consistency level that, certain recommendation algorithms possess. We work with three ensemble techniques, Boosting, Bagging and Smoothing which helps in improving the stability and providing greatly personalized predictions as outcome. Due to the nature of ensemble techniques of filtering the outliers from data, these techniques are employed for improving accuracy also. The stability is computed in 2 phases as it is proved to be experimentally efficient. Findings: This paper analyses the stability factor of the outcome and the impacts on varying the data quantity. We also analyze the impact on stability due to the variations made to the data under evaluation. Applications: This approach makes us understand the performance metrics and quality of any recommender system being used.