Paper Reading #10

Comments:
Comment #1
Comment #2
Reference:
Title: Designing Adaptive Feedback for Improving Data Entry Accuracy
Author: Kuang Chen, Joseph M. Hellerstein, Tapan S. Parikh
Venue: UIST 2010
Summary:

This article summarizes a system name USHER, a theoretical data-driven system aimed at improving data entry quality. USHER strives to decrease time during data entry, and increase time verifying the data after entry. Thus, the overall goal is not efficiency or decreasing time, but rather increasing the correctness of the data being entered. USHER has been shown to improve error rates by 54-78% for radio buttons, but the impact for other widget types was statistically insignificant.  The system tries to ask the most unpredictable answers first, and uses the data from those questions to "guess" on the following questions.

The feedback mechanisms use two formulas:

1. t = a/D, a > 0
2. k = min(7, ceiling(logb(D)), b > 1

For defaults, or expected answer, the first equation is used with a=1.5. This produces a threshold of 75% confidence. Warnings also use the first equation, with a=0.1. This equates to less than 5%. Widgets, with the exception of text auto-complete, use the second equation. Drop downs copy the k most likely choices, while radio buttons highlight the k most likely choices. Text auto complete uses a likelihood based approach rather than alphabetical.

The model for USHER is a "Bayesian network over a form that captures the relationships between for questions based on prior data." To test this system, they created a web application with Java Servlets and Adobe Flex. All feedback mechanisms used client-side code that calculated the likelihood values based on previous answers.



Discussion:
This was another interesting article. Right now it seems like the UIST articles are more interesting than the CHI 2010 articles. I really liked the quote this article started with from Carl Malamud: “information is a form
of infrastructure; no less important to our modern life than our roads, electrical grid or water systems.” I certainly think a good case can be made for that statement, which is why we need systems like USHER to increase the validity of the information we enter into our databases. Also I liked how the article described every meaningful detail about the system, all the way down to formulas and programming techniques, while keeping the article sufficiently succinct. The use of pictures broke up the text nicely too (may seem unimportant to some, but I find it highly important). The strategy of using the most unpredictable questions first is also interesting to me. Usually it seems like people do the exact opposite and then try to predict off those. I can see how using the unpredictable questions first could be more useful. Overall, I liked this article.