Query Recommendations for Interactive Database Exploration

Abstract

Relational database systems are becoming increasingly popular in the scientific community to support the interactive exploration of large volumes of data. In this scenario, users employ a query interface (typically, a web-based client) to issue a series of SQL queries that aim to analyze the data and mine it for interesting information. First-time users, however, may not have the necessary knowledge to know where to start their exploration. Other times, users may simply overlook queries that retrieve important information. To assist users in this context, we draw inspiration from Web recommender systems and propose the use of personalized query recommendations. The idea is to track the querying behavior of each user, identify which parts of the database may be of interest for the corresponding data analysis task, and recommend queries that retrieve relevant data. We discuss the main challenges in this novel application of recommendation systems, and outline a possible solution based on collaborative filtering. Preliminary experimental results on real user traces demonstrate that our framework can generate effective query recommendations.

Matrix Factorization Based Query Recommendation

“…In this instantiation of the framework [2], the session summary S i is represented as a weighted vector, where every coordinate corresponds to a distinct database tuple. Assume that the total number of tuples in the database, and as a consequence the length of the vector, is T. The weight S i [τ] represents the importance of a given tuple τ ∈ T in session S i , and is non-zero only if τ is a witness for at least one query in the session.…”

Predicting Your Next OLAP Query Based on Recent Analytical Sessions

“…Recently, in the database community, there has been an increasing interest in leveraging past queries or query answers to assist interactive relational database exploration [3][4][5][6][7][8][9]. The approaches proposed include past query browsing and/or searching [7], query completion [6] and query recommendation [5,8,9]. Noticeably, automatic query recommendation approaches either rely on the query answer and database instance, which may lead to efficiency problem, or treat sessions as sets of queries, overlooking the intrinsic sequential nature of the exploratory process.…”

Describing Analytical Sessions Using a Multidimensional Algebra

“…More elaborated management was not needed as operational systems mainly issue canned (i.e., known in advance) queries. To our knowledge, only two works go beyond and propose a framework to manage the knowledge captured in the issued analytical queries (i.e., the DBMS log) to support query recommendation [3,4] or query completion [11] for interactive analysis of relational sources. The first approach follows the idea of recommender systems in the exploration of Web data.…”

“…This trend is rather evident for scientists, who are increasingly using relational databases and SQL for conducting analytical sessions over huge repositories of data [11]. For this reason, novel works have focused on supporting analytical tasks over relational sources [3,4,10,11,14,17]. Specifically, it has already been pointed out the necessity to come up with flexible, powerful means for analyzing the issued queries (the keystone of these systems, usually stored in the DB query log), and decompose, store and handle them in a dedicated subsystem in order to better support any decisional task with the knowledge captured in the analytical queries [10].…”

“…Specifically, it has already been pointed out the necessity to come up with flexible, powerful means for analyzing the issued queries (the keystone of these systems, usually stored in the DB query log), and decompose, store and handle them in a dedicated subsystem in order to better support any decisional task with the knowledge captured in the analytical queries [10]. As examples of such tasks, the system should be able to support the user when formulating new queries and leverage his / her knowledge with other users (query recommendation) [3,4,11] and, at the same time, database administrators should tune their databases to cope with the evolution of queries issued (physical as well as conceptual design) [14,17].…”