Initial Sampling for Automatic Interactive Data Exploration

In many real world applications, users might not know the queries to send to a database in order to retrieve data in the user-interested areas. Users can apply a trial and error method to discover the queries. However, as the data set is usually quite large, the discovery of queries will take a long time and the whole process is labor-intensive. We want to build a discovery-oriented, interactive data exploration system, that guides users to their interested data areas through interactive sample labeling process. In each iteration, the system will strategically select some sample points to present to users for feedback, as relevant or irrelevant, and finally converge to a query that is able to retrieve all the data in the user-interested area.