SmithKettlewellRERC / DVX-YouDescribe-issues

Issue tracking repo for DVX & YouDescribe (Automatically exported from code.google.com/p/sk-dvx)
http://youdescribe.org/
0 stars 0 forks source link

We need a heuristic for estimating completeness of a description #56

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago
There are many partial descriptions currently in the database. These often 
consist of a few descriptive clips at the beginning of a video, but the 
describer is either only testing, or loses stamina. Many videos do not have 
descriptions all the way through. It would be nice to have a way of estimating 
the degree of completeness for a given description set. 

This has nothing to do with evaluating the quality of the descriptions for 
wording or recording quality, but only to evaluate the distribution of clips 
throughout a video in a simple, easy-to-understand way.

I propose something like:

DescriptionEffort = ( (LastClipStartTime-FirstClipStartTime) / 
TotalVideoDuration )  *  (NumberOfClips / VideoDurationInMinutes)

This should result in numbers around 1 or higher for reasonably thorough 
descriptions and lower numbers for the cases with only a few clips at the 
beginning. It will be zero for descriptions that include only 1 clip.

Original issue reported on code.google.com by joshua.a...@gmail.com on 8 May 2013 at 4:41

GoogleCodeExporter commented 9 years ago
I was thinking along the lines of standard deviation. 
The first measure of completeness is how much the set of clips 'covers' the 
time of the video. Making the assumption that most videos have an even 
distribution of the times when description is desirable, the more complete 
description is the one that has rather evenly-spaced clips, in other words, a 
low clustering of clips. So, the standard deviation of the times between clips 
should be a good first approximation of completeness. The computation should 
include the time from the beginning of the video to the first clip, and the 
time from the last clip to the end of the video. This will avoid the case of a 
set of evenly distributed clips that are only in the beginning of the video.

Original comment by keith...@gmail.com on 29 Apr 2014 at 1:07

GoogleCodeExporter commented 9 years ago
An implementation based on the algorithm Josh describes is in:

http://youdescribe.ski.org/testtabs/

for example:

http://youdescribe.ski.org/testtabs/player.php?v=5OLlemqd3SE

Original comment by Owen.R.E...@gmail.com on 19 Jun 2014 at 6:28