Adaptive, efficient and effective graph data integration and search framework

Abstract

This paper proposes an idea for a new graph data integration and search system leveraging techniques of provenance, visualization and association rules. The graph data model, along with techniques of web data integration and database usability are central to the development of the web which provides motivation for this work. The overall proposed system consists of three components:a) Quality-aware graph visualization leveraging user feedback: we propose a solution to utilize user feedback on relevant portions of schema of graph database in graph sample selections so as to provide an adaptive way to choose graph samples via schemas in order to provide relevant quality-aware visualization later on and a solution based on statistical techniques of averages and heatmaps visualization in assigning color intensity to spatial positions of vertices and edges based on quality to allow the user to gauge the quality of graph datasets.b) Visualization-enabled graph alignment leveraging query logs: we propose a solution to utilize graph query logs along with run-time information of graph neighborhoods that led to graph alignment decisions in task of graph alignment so as to improve its effectiveness as well as a solution based on assigning spatial positions to graph vertices and edges based on graph similarity measures to allow the user to visually compare graph datasets during integration.c) Efficient batch generation of multiple graph query autocompletion suggestions: Techniques in literature typically make use of data clustering, data reduction, utilizing query logs to provide feedback and autocompletion suggestions to users during query formulation. As another solution with a focus on efficiency in context of graph datasets, we propose to leverage the fact that multiple queries may have many autocompletion suggestions in common and therefore propose an efficient algorithm to generate them together efficiently and show the performance improvement over the current sequential approach that involves repeated scans.