Route Finder Tweak: better sorting

I frequently use the Route Finder to give me a ticklist, which usually involves using "Stars" as the first sort option. Sometimes obscure routes show up high on the list because they have just a single (or small handful) of votes, and they don't really belong at the top of the list.

Fixed! Routes are now penalized if they have a small number of votes, so if you sort by "Stars", the results are not always exactly ordered by Stars. Trust me, it's better.

I also added the number of votes to the resulting list (in small gray text), so you can judge the accuracy of the star rating for yourself.

Example

I've been frustrated by this for years. One simple solution is to give EVERY route in the database, one vote of 2 stars to begin with.

0 votes = avg 2
1 @ 4* = avg 3
2 @ 4* = avg 3.33
3 @ 4* = avg 3.5
etc

there would never again be a route with a perfect 4* rating,

BUT.. the routes with the most number of high votes would be at the top.

Josh, that's interesting, but I did it this way (sql speak):

order by quality_stars * least(quality_stars_votes, 5) desc

This causes routes with less than 5 votes to be penalized proportionally to the number of votes they have. Routes with more than 5 votes are treated equally.

Nick, I like Joshua's suggestion. I know of other sites that rank things similarly (IMDB, e.g.).

How hard would it be to implement the following? A route starts with 10 (or some other fixed number) of votes at 2.5 stars (or some other rank which is close to the average rank of all routes in the database). Each time a user ranks a route, one of the old votes is replaced, until all 10 are gone.

Common problem in the local search domain. One possible solution:

Bayesian Rating is using the Bayesian Average. This is a mathematical term that calculates a rating of an item based on the “believability” of the votes. The greater the certainty based on the number of votes, the more the Bayesian rating approximates the plain, unweighted rating. When there are very few votes, the bayesian rating of an item will be closer to the average rating of all items.

Baysian rating = ( (avg_num_votes * avg_rating) + (this_num_votes * this_rating) ) / (avg_num_votes + this_num_votes)

from: the broth blog

Excellent tweak!

This tweak should also be applied to "Area Classics" I find often the "Classics" are obscure climbs that the FA thought deserved 4 stars.