Recomendação de consultas de banco de dados utilizando agrupamento de usuários.

Abstract

Database systems are becoming increasingly popular in the scientific community to support the exploration of scientific data, m this scenario, users may not have the necessary knowledge about the domain of the database or not knowing formulate SQL queries for data analysis. To solve this problem has been emerged many studies about queries recommendation techniques. The recommendation methods of query in database has been emphasis in maximize the accuracy of the recommendations, but other aspects such as novelty and diversity may be important for recommendations. In this context, this research aimed to improve the recommendations of SQL queries regarding the metrics: relevance, novelty, diversity, an amount of new tables, precision and recall. This goal was achieved through an approach to the recommendation of queries using clusters of database users. The results of experiments using real historical queries of the SkyServer project shows that through the proposed approach we can generate recommendations for appropriate queries for each user of the database used. Furthermore, the proposed approach was evaluate comparing with techniques described in related work. The analysis shows that the values of the metrics studied in this research are 64,6% higher in the proposed approach than the techniques compared. These results potentially provide better conditions for studies and future work using groups of users to generate recommendations for database queries. The proposed approach also enabled the design of user behavior with database, this information contributes to better understanding the interaction of users with management systems database.