DataScope A Database Content Visualization Tool based on Ranking Queries CS511 Course Project Tianyi Wu Dec 08 2006 DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work Motivation Existing database systems SQL query based Form based Limited user interface Inconvenient to browse data Existing database visualization Polaris Stanford DIVE ON U of Alberta Stotle et al Query analysis and visualization of hierarchically structured data using Polaris KDD 2002 Ammoura et al Towards a novel OLAP interface for distributed data warehouses DaWaK 2001 Maryland Various projects and tools http www cs umd edu hcil research visualization shtml Some screenshots Polaris Some screenshots Some screenshots Maryland Limitations of Existing Work Particular domains Spatial temporal Time series Predefined schemas Fixed visual representation Statistical charts e g scatterplots DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work Goal Visualize databases like Google Maps Content based Explorative easy to use Dragging pan drilling zoom Domain independent Web based interfac Fast query processing Challenges Layout User preferences on what to see Not well defined and well understood Maps longitude latitude How to position objects on screen Different users have different preferences Even the same user may have different preferences based on the context of the query Data is often associated with multiple hierarchies or semantic links Powerful query engine Contribution Interface design Principles which can address the above challenges A system prototype Efficient implementation Ranking Cube Xin et al VLDB 06 Ranking Aggregation DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work About the Demo Not fully functional yet Demonstrate important concepts Ajax vs PHP Ranking Efficiency Customization Datasets Real DBLP extracted 20387 database related entries Synthetic store database DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work System Architecture DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work DataScope Overview DataScope Overview DataScope Overview Design principles Structured dimensions Ordering of attribute values Selection of comprehensive layout Quick selection Display rich information Easy customization Implementation Linear ranking functions with arbitrary selections Ranking on aggregation Design principles Structured dimensions View data in multi resolution Roll up and drill down Can be automatically generated Numeric attributes Age salary price Categorical attributes Milk dairy products food Design principles Ordering of attribute values How to order values along X Y axis Ascending Descending Alphabetic order e g AAAI CIKM Numeric order e g 2001 2002 Independent of any ranking function The order of a value is not determined by its score Design principles Selection of comprehensive layout Initial layout High level familiar to most users Map US map DBLP AI Theory System 80s 90s 00s Subsequent layout Can be changed according to different data Customizable Design principles Quick selection Dragging Scrolling the mouse wheel to roll up drilldown or zoom in out Push constraint easily Context menu Design principles Display rich information Top k as in Google Maps K representative items Outliers Display primitives Color size e g big cities have big font etc Searching Design principles Easy customization Users can freely define their own layout Adjust the resolution X location Y year More less objects on screen Customize ranking function e g rank houses by 0 7 Price 0 3 size Selection database conferences and 2003 2006 Implementation Ranking Cube Xin Et al Answering top k queries with multidimensional selections VLDB 06 Linear ranking functions Arbitrary selections Methods Partition the data space and store blocks Progressively retrieve the most promising blocks for each query Data fragments Partial materialization to deal with high dimensionality Ranking on Aggregation Example Given a relation conference year author paper Query SELECT top k COUNT author FROM R GROUP BY conference year Ranking on Aggregation Method Materialization for all possible cuboids Algorithm Input aggregation dimension D ranking dimensions R concept hierarchies H Output a set of ranking fragments S 1 For each possible group by of R and H 2 Compute aggregation for each value in D 3 Compute ranking fragments for D 4 S S D DataScope Motivation Contributions Demonstration Architecture Design Implementation Future Work Conclusion DataScope Extend the current prototype to support mapping operations and multiple sessions Improve design principles which can lead to a more effective interface Support various ranking queries efficiently Future work Interface Improve the initial system prototype Support the full set of operations Support easy customization Implementation Rich research issues Ranking objects based on user feedbacks Retrieve most relevant objects in keyword searching Multiple types of ranking queries Thank you Any questions