Tried and Tested Evaluating Data Entry Consistency in

  • Slides: 17
Download presentation
Tried and Tested? Evaluating Data Entry Consistency in the Component Model Prototype Brit Hopmann,

Tried and Tested? Evaluating Data Entry Consistency in the Component Model Prototype Brit Hopmann, July 1, 2008

Tried and Tested? Presentation • Background & aim consistency check • Images & data

Tried and Tested? Presentation • Background & aim consistency check • Images & data • Analysis & results (in 3 steps) • Problems & possible solutions • Conclusion & discussion data entry Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Background & aim consistency check Background • Problems with data entry

Tried and Tested? Background & aim consistency check Background • Problems with data entry ambiguity • Consequences: decreased recall and precision Aim • Check and improve data entry consistency • Solutions for existing problems? • Changes? data entry Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Images Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Images Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis Data • 5 images • 25 features (+ 2 discarded)

Tried and Tested? Analysis Data • 5 images • 25 features (+ 2 discarded) • 6 respondents Analysis in 3 steps • Categorizing the answers (per image / per feature) • Analysis per image and per feature • Problems and possible solutions Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis – step 1 example: answers for image 4 Koninklijke Bibliotheek

Tried and Tested? Analysis – step 1 example: answers for image 4 Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis – step 1 Categorizing the answers (per image / per

Tried and Tested? Analysis – step 1 Categorizing the answers (per image / per feature) 3 categories category consistency times percentage no problem 57 45, 6% indecisive 16 12, 8% problem 52 41, 6% There were many problematic descriptions. Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis – step 2 Summary per watermark (5 images) None of

Tried and Tested? Analysis – step 2 Summary per watermark (5 images) None of the images was described unanimously. image no prob ≈ problem per image 1 15 2 8 fine image 2 13 3 9 fine image 3 10 3 12 problematic image 4 11 3 11 indecisive (≈) image 5 8 5 12 problematic 57 16 total Koninklijke Bibliotheek – National Library of the Netherlands 52

Tried and Tested? Analysis – step 2 Summary per feature (25 features + 2

Tried and Tested? Analysis – step 2 Summary per feature (25 features + 2 discarded) Described consistently 5 -7 features* 3 of 5 images not consistently described 10 -8 features* * Depending on quality of definitions Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis – step 3 How problematic are specific features? 1. Is

Tried and Tested? Analysis – step 3 How problematic are specific features? 1. Is or might the feature be problematic with other watermarks? 2. What caused the problem and can it be solved easily? -> image quality, definition, subjective description 3. What is the distinctive value of a particular feature? 4. Is there a possible solution? 5. Which problems will remain problematic after having applied a specific solution? Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Analysis – step 3 What could contribute to a solution? •

Tried and Tested? Analysis – step 3 What could contribute to a solution? • • Grouping • Allow assigning multiple features • Drawback: increased decision effort Extra tool • For instance graphical indicator of angle • Drawback: exceptions such as rotated scans Clarify definitions • For instance description of additional elements • Relatively straightforward to apply Otherwise Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Crucial Question Which features will still pose a problem with a

Tried and Tested? Crucial Question Which features will still pose a problem with a particular solution applied? Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Conclusion Problems that will eventually remain • Eyes: alignment, distance, position

Tried and Tested? Conclusion Problems that will eventually remain • Eyes: alignment, distance, position • Horns: endings, orientation, inside Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Conclusion Remaining problems • Ears: endings, roundness, width • Eyes: shape

Tried and Tested? Conclusion Remaining problems • Ears: endings, roundness, width • Eyes: shape • Head: shape • In-between horns: shape • Nose: concomitance with other elements, shape Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Discussion Features that evoke subjective descriptions (shape!) cause problems Thinking about

Tried and Tested? Discussion Features that evoke subjective descriptions (shape!) cause problems Thinking about possible changes in the model • Clarify certain issues: clear definitions, instruction • Expand model: extra tool, grouping • Simplify model: adjust or take out features Reconsider role of component model within Bernstein data Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Discussion Questions and Comments? Koninklijke Bibliotheek – National Library of the

Tried and Tested? Discussion Questions and Comments? Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Koninklijke Bibliotheek – National Library of the Netherlands

Tried and Tested? Koninklijke Bibliotheek – National Library of the Netherlands