RSATree: Distribution-Aware Data Representation of Large-Scale Tabular Datasets for Flexible Visual Query

Area of Science:

• Data Visualization
• Database Systems
• Information Retrieval

Background:

• Analysts use charts like histograms for large-scale data analysis, relying on aggregate queries.
• Predefined data cubes accelerate responses but limit query flexibility due to fixed binning schemas.
• This inflexibility hinders effective pattern discovery in large datasets.

Purpose of the Study:

• To develop a system enabling arbitrary queries and flexible binning strategies for large-scale tabular data analysis.
• To overcome the limitations of predefined binning schemas in data cubes.
• To enhance analysts' ability to investigate implicit data patterns through adjustable granularities.

Main Methods:

• Leveraging an R-tree-based space partitioning scheme for data distribution.
• Employing locality-sensitive hashing for locality-preserving random data access.
• Utilizing a summed area table scheme for interactive aggregated value queries with linear complexity.
• Implementing a web-based visual query system for specification, query, and exploration.

Main Results:

• Demonstrated efficiency and utility through experiments on real-world datasets.
• Analysis of time and space complexity confirming system performance.
• Successful implementation of a web-based system supporting user-adjustable data granularities.

Conclusions:

• RSATree provides a flexible and efficient solution for analyzing large-scale tabular data.
• The system enhances data exploration by allowing arbitrary queries and adjustable binning strategies.
• This approach significantly improves analysts' ability to uncover patterns in complex datasets.