GSQueRIE: Query Recommendation Using Matrix Factorization

Abstract

The emerging technological development has resulted in generation of huge heterogeneous data. To explore database, user approaches towards Database Management System applications by forming SQL queries. User lacks proficiency in the domain of SQL language and finds complexity in formulation of queries, which became need to dive towards Recommendation system. The existing Recommendation system faces the problem of data sparsity, cold start, finding some relative changes between the queries and hidden features to calculate the exact similarities. To trounce this problem, Matrix Factorization technique is used in synergy with Recommendation system. Matrix factorization method became foremost methodology which discovers the latent features, essential for the interactions between users and items. Hence, we have developed GSQueRIE (Generic-Scalable QueRIE) system with the goal of improving the overall scalability and flexibility of collaborative database exploration. Using this system, user will be able to explore the database interactively to get top K query recommendations generated with the help of Matrix Factorization technique. We have used Bayesian Probabilistic Matrix Factorization method along with Monte Carlo approximation for predictive distribution. We also compared Fragment based approach and Matrix factorization approach, discussing the advantages and disadvantages of each. We have shown through experiment that Matrix Factorization method results better than Fragment based approach while generating recommendations in terms of precision, recall and f-score metrics.