ViewSeeker: An Interactive View Recommendation Tool

ViewSeeker: An Interactive View Recommendation Tool

BigVis 2019 ViewSeeker: An Interactive View Recommendation Tool Xiaozhong Zhang, Xiaoyu Ge, Panos K. Chrysanthis, Mohamed A. Sharaf Table of Contents Background and Motivation Problem Definition The ViewSeeker

Experimental Testbed and Results Conclusion What is a View? A view is a histogram or bar chart showing the binned aggregated values over a dataset. A target view showing the player 3-point attempt rate of a NBA team.

An Insight from Views Create a reference view showing the attempt rate of all NBA players. The comparison between the two may explain why the selected team won a championship. Views are powerful agents in helping the user explore and understand big data!

What is View Recommendation? Among all possible views, find the top-K views that are most useful according to the users interest. Existing view recommendation approaches proposed many view utility functions to estimate the usefulness of a view, e.g.

deviation accuracy usability p-value [Manasi Vartak, et al. SEEDB: ... Visual Analytics. In PVLDB 8, 13 (2015)] [Humaira Ehsan, et al. MuVE: Visual Data Exploration. In IEEE ICDE 2016.]

[Bo Tang, et al. Extracting Multi-dimensional Data. In ACM SIGMOD 2017.] However There are too many view utility functions to choose from and we dont know which one best captures the users interest. Predefined View Utility Functions Recommendation precision of some predefined view utility functions:

Predefined view utility function cannot capture the users interest well. Our Contribution The first attempt towards automatically discovering the most appropriate view utility function during interactive view recommendation that best captures the users interest. Table of Contents

Background and Motivation Problem Definition The ViewSeeker Experimental Testbed and Results Conclusion Classical View Recommendation INPUT A database D A set of results R produced by a user-specified query Q

A predefined view utility function u() OUTPUT The top-K views V = {v1, v2, ..., vk} constructed from R The views in V should have the highest utility among all possible views according to u(). Interactive View Recommendation INPUT A database D

A set of results R produced by a user-specified query Q A set of n possible utility functions U = {u1, u2, ..., un} User feedback on example views A time constraint tl Interactive View Recommendation OUTPUT A composite utility function up(), which can be any arbitrary combination of the utility functions in U. The top-K views ranked by up() should have high utility

according to the users ideal utility function u*(). The computational delay between each subsequent interactions (feedback) with the user is within tl. Table of Contents Background and Motivation Problem Definition The ViewSeeker Experimental Testbed and Results Conclusion

ViewSeeker Phase 1: OfflineInitialization Stage 1: Calculate aggregated values for each view Stage 2: Calculate utility features for each view

KL-divergence EMD L1

L2 Max. Difference Usability Accuracy p-value ViewSeeker Phase 2: Interactive View Recom. Recommend

ViewSeeker Phase 2: Interactive View Recom. Which view to select? The most informative view! Which view is the most informative? The view whose utility the system is most uncertain about! ViewSeeker

Optimizations Before interactive recommendation, calculate rough utility features using partial data During interactive recommendation, rank the views based on up() Refine utility features using all data for highly-ranked views Table of Contents Background and Motivation

Problem Definition The ViewSeeker Experimental Testbed and Results Conclusion Performance Metric Top-k Precision: = | |

V* is the top-k views recommended by u*() Vp is the top-k views recommended by up() Evaluation of Effectiveness Number of labeled views needed to reach a 100% precision Single-Component u*()

Evaluation of Effectiveness Number of labeled views needed to reach a 100% precision On average, only need 9 16 labeled views Two-Component u*() Three-Component u*() Evaluation of

Optimizations Utility Distance (UD): = (

()

() / V* is the top-k views recommended by u*() Vp is the top-k views recommended by up() u*() is the ideal utility function ) Evaluation of Optimizations

Running time needed to reach UD = 0 Three-Component u*() Evaluation of Optimizations Number of labeled views needed to reach UD = 0 43% reduction in running time 19% increase in user labeling effort

Three-Component u*() Conclusion ViewSeeker uses active-learning techniques to discover the most appropriate view utility functions during an exploration based on the users feedback. The optimization techniques significantly improve the runtime efficiency of the ViewSeeker. ViewSeeker outperforms baseline approaches by a significant margin for both synthetic and real-world datasets.

Thank You! ViewSeeker: An Interactive View Recommendation Tool Xiaozhong Zhang, Xiaoyu Ge, Panos K. Chrysanthis, Mohamed A. Sharaf http://db.cs.pitt.edu This work was supported, in part, by NIH under award U01HL137159.

Recently Viewed Presentations

  • Gas Piping Technology Committee

    Gas Piping Technology Committee

    Gas Piping Technology Committee. Future. still holds new challenges for the Committee. With its dedicated members, the Gas Piping Technology Committee will continue to prove to be equal to those challenges and continue to prosper
  • Temperature: Fahrenheit, Celsius, and Kelvin

    Temperature: Fahrenheit, Celsius, and Kelvin

    Kelvin. 1848, British scientist, William Thomson Lord Kelvin developed a scale that relies on the average kinetic energy of atoms. Advantages:-Absolute scale-Empirical -No Negatives-Still a scale of 10. T. K = T C + 273CONVERSION
  • Remains - BA English Revision

    Remains - BA English Revision

    Context. Simon Armitage's poetry is known for its colloquial style, strong rhythms and voice. He often uses the monologue form in his poetry to create immediate and moving characters. The reference to 'desert sand' suggests this poem is set in...
  • Venturing Leader Specific Training

    Venturing Leader Specific Training

    Venturing Leader Specific Training ... BSA Mission Statement The mission of the Boy Scouts of America is to prepare young people to make ethical and moral choices over their lifetimes by instilling in them the values of the Scout Oath...
  • Unit 1: Circles & volume

    Unit 1: Circles & volume

    Unit 1: Circles & volume. Test 1/24. ... Composite solid volume. Composite solids are made up of more than one standard 3D shape. Volume applications. Complete the worksheet on Volume Applications. Finish this for homework Warm up! This is Sparta!...
  • CBE 417 - webpages.sdsmt.edu

    CBE 417 - webpages.sdsmt.edu

    With Aspen Rate-Based Distillation feature, RadFrac can perform rate-based distillation calculations (packed tower configurations). Rating Mode. Design Mode. Rigorous Model. Multicomponent Methods. AspenPlus (methanol, ethanol, 1-propanol): Component. P* i
  • N O I T C I ADD x

    N O I T C I ADD x

    Dr Owen Bowden-Jones. David Stuart, Antidote. Mark Dunn. Stacey Hemmings. Declarations. CH has in the past received honoraria for travel / lecturing (not related to this work) from: Bayer, Lilly, Pfizer and Janssen . DrPepe Catalan. Flick Thorley. DrAmritSachar.
  • Strategic stadies and International Relations. Security for ...

    Strategic stadies and International Relations. Security for ...

    Security Studies and International Relations Security for whom? ... There is a presupposition of international anarchy but it is possible to overcome this anarchy. ... of the state before the military and political threats American policy is the power policy...