Saturday, September 7, 2013

The AstroBin Voting System

Like many web sites that allow people to vote on contributions, AstroBin has run afoul of troll voters.

Currently AstroBin voters use a 5-point scale. The text descriptions of each vote are:

  1. "It's a start, but it needs better data"
  2. "It looks good, but it needs better processing"
  3. "Good data, good processing"
  4. "Great Result!"
  5. "Top Quality!"

Each vote is weighted by a scheme that reduces the effects of outliers, and people aren't allowed to vote for their own images. Salvatore Siovene gives of his own time to comb the votes for those cast by dummy accounts. I have only the highest praise for Salvatore and his efforts regarding AstroBin, which is a labor of love on his part. If some parts of it become commercially successful, more power to him!

In a perfect world, everyone votes on an image based on the image's technical and artistic merit. But we all know how difficult that can be from watching judged sporting events such as skating and gymnastics. In a hobby where egos can run rampant there's little hope of ever doing away with votes cast with dishonest intent. Troll voters adapt quickly to remedies; after it was announced that 1-votes would likely bring scrutiny, the trolls began voting 2s, often for images that were clearly excellent.

Another web site I visit has been plagued by troll voting, and many people have suggested alternative systems for legitimizing voting:

  • Change to a 1 to N scale, usually with N being a number like 10 (This is much like electronic zoom, it accomplishes nothing)
  • Clip the distribution of votes more strongly (Like median stacking, this works well only when there are many votes--usually there aren't)
  • Allow only selected members to vote (Aside from not being very democratic, this leads to other problems, such as favoritism)

I gave the problem quite a bit of thought a few years ago and came to the conclusion that the voting system that would work best was a binary one: a 1 vote indicates that you think the image is above average, and not casting a vote suggests you think it's below average. Note that there is no 0-vote. If you don't think an image is very good, you simply cast no vote.

A binary system is imperfect. Ghost accounts can still pollute the voting. While their effect can reduced by tracking IP numbers, they can never be entirely eliminated .

Another failing of binary voting is that the number of votes depends in part on the interest in the target object. For example, photogenic nebulae may get more attention and more votes. If one image gets 100 1-votes out of 500 "looks," does that mean it's a better image than one that gets 5 votes out of 25 looks? Not really. But 100 votes looks a lot better than 5.

At the moment AstroBin is polling its members to see if it should switch to a simple "like" system. This is basically the binary system that I've been talking about, so I voted for the change. Personally, I wish that the poll had included the option to do away with voting entirely.

All of the above, I suppose, explains why I just opted out of the voting system entirely.

No comments:

Post a Comment