Querying Entity-relationship Graphs By Example Tuples: Experimental Evaluation And User Study

Abstract

The World Wide Web today has evolved into a rich repository of entities where many knowledge bases containing entity-related information are directly available. Such knowledge bases are often in the form of entity-relationship graphs. To query entity-relationship graphs, users need to provide input entities, attributes and relationships by complex query graphs. To improve the usability of graph database systems, we study a novel mechanism that queriesentity-relationship graphs by example tuples. It allows users to express a query in the form of one or more tuples consisting of entities. The underlying query system automatically builds aquery graph based on the example tuples and ranks matching answer tuples.The focus of this thesis is to evaluate our query system's accuracy and efficiency. To evaluate accuracy we employ two methods. In the first method we evaluate queries whose ground truths are known and calculate system's precision and recall. In the second methodwe conduct user study on ranked answer lists and calculate rank correlation co-efficient. The run time efficiency of the system is measured with respect to the size of the query graph.