Transcript
Page 1: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Reinventing the Contingency Wheel: Scalable Visual Analytics of Large Categorical Data

TVCG 2012Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M.

Eduard Grïoller

Page 2: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Introduction Related work

◦ Contingency Wheel Contingency Wheel++ User Study Conclusion

Outline

Page 3: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency table◦ a common way to summarize categorical data

as a first step of analysis◦ an nxm matrix that records the frequency of

observations ƒij for each combination of categories of two categorical variables

Data◦ about 1million user ratings on 3706 movies

a 3706x21 table which counts for each movie, how many times it was rated from users of each occupation

a 604017 table which counts for each user, how many times he/she rated movies from each genre

Introduction

Page 4: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Introduction

Page 5: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

the column categories are visualized as sectors of a ring chart

the table cells are depicted as dots in these sectors

the dot for cell (i,j ) is placed in sector i at a radial distance from the ring’s inner circle proportional to the strength of association rij between row i and column j

to reduce the large number of dots ◦ by filtering out cells (i,j ) with rij ≦Tr (where Tr is the

association threshold) ◦ by filtering out entire rows with ƒi+<Ts (where Ts is the

support threshold)

Contingency Wheel

Page 6: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel

Page 7: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Data mapping◦ to visualize association values rij

Visual mapping◦ it difficult to accurately interpret the meaning of

these dots at the beginning Interaction

◦ dots closer to the center were often too small and overlapping

Limitations Of The Contingency Wheel

Page 8: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel++

Page 9: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel++

where cte is a constant computed from the table to ensure -1≦ ri j ≦1

Page 10: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel++

Page 11: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel++

Distributions of (a) a numerical attribute (release date) or, (b) a categorical attribute (genre) of the movies in the histograms. (c) The global distributions of release date and genre among all movies.

Page 12: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Visualizing column similarities

Page 13: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

User Study-User occupations

Page 14: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

User Study-Movie genres

Page 15: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

User Study-Movie genres

Page 16: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

User Study

Details about selected genres.

Page 17: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

a novel visual analytics methods◦ visualize and discover patterns in large

categorical data improve Contingency Wheel offer a multi-level overview+detail interface

Conclusion

Page 18: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Introduction

Page 19: TVCG 2012 Bilal Alsallakh, Wolfgang Aigner, Silvia Miksch, and M. Eduard Grïoller

Contingency Wheel


Recommended