Efficient Recommendation of Aggregate Data Visualizations

Data visualization is a common and effective technique for data exploration. However, for complex data, it is infeasible for an analyst to manually generate and browse all possible visualizations for insights. This observation motivated the need for automated solutions that can effectively recommend...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on knowledge and data engineering 2018-02, Vol.30 (2), p.263-277
Main Authors: Ehsan, Humaira, Sharaf, Mohamed A., Chrysanthis, Panos K.
Format: Article
Language:English
Subjects:
aggregate views
Aggregates
Data aggregation
Data exploration
Data processing
Data visualization
Exploration
Measurement
Pruning
Scientific visualization
Search problems
Upper bounds
Utility functions
view recommendation
visual analytics
Visual databases
Visualization
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Data visualization is a common and effective technique for data exploration. However, for complex data, it is infeasible for an analyst to manually generate and browse all possible visualizations for insights. This observation motivated the need for automated solutions that can effectively recommend such visualizations. The main idea underlying those solutions is to evaluate the utility of all possible visualizations and then recommend the top-k visualizations. This process incurs high data processing cost, that is further aggravated by the presence of numerical dimensional attributes. To address that challenge, we propose novel view recommendation schemes, which incorporate a hybrid multi-objective utility function that captures the impact of numerical dimension attributes. Our first scheme, Multi-Objective View Recommendation for Data Exploration (MuVE), adopts an incremental evaluation of our multi-objective utility function, which allows pruning of a large number of low-utility views and avoids unnecessary objective evaluations. Our second scheme, upper MuVE (uMuVE), further improves the pruning power by setting the upper bounds on the utility of views and allowing interleaved processing of views, at the expense of increased memory usage. Finally, our third scheme, Memory-aware uMuVE (MuMuVE), provides pruning power close to that of uMuVE, while keeping memory usage within a pre-specified limit.
ISSN:1041-4347
1558-2191
DOI:10.1109/TKDE.2017.2765634