Strava'a algorithm for hidden segments has issues

[PiotrekNL]

Hello,

I would like to comment on Strava's algorithm for hiding segments. From what I understand it's based, among other things, on how often the segment is viewed. This unfortunately leads to bad climbing segments being promoted over good ones for two reasons:

1)Longer segments are more likely to be clicked on because users want to see "the whole thing". Unfortunately this means that segments that starts too early (before crossroads etc.) or finish too late (after the summit) are more likely to be clicked on that better ones (containing the whole climb).

2)Segments that start too early and end too late are more likely to generate top10 places because of two reasons: fewer people do them and a lot of people stop at the top before the descent. This means those bad segments are going to be clicked on more often because people get top 10 places on them.

This is frustrating because perfect segments (those that start at the beginning of the climb after last crossroads and end at/just before the summit) are hidden over time in favor of not so great segments that require you to descend after the climb is done, sometimes for quite a bit for them to count).

The solution could be to promote segments that don't have gradient change from negative to positive (and vice-versa) at the start and finish. That would make segments like "the whole climb X" visible.  Another could be to let people vote on segment asking them if they are made according to the standards. Thank you for your attention!

[PiotrekNL]

Guys, I understand it's technical and requires some research but the algorithm is seriously messed up. It seems it has some bonus for longer segments and promotes bad segments because of it (those that start in random places before crossroads, 2 streets before the climb etc.). Segments are one of the main features people use Strava for. It will seriously hurt you if nothing is done about it. I've done some work already demonstrating the issue. It would be nice if you could think about it as well.

[PiotrekNL]

>>Of course a bad designed segment with i.e. only 10 efforts ever has a 100% top 10 probability and is maybe not hidden but why this should be a problem for anybody I can't see.

The problem is not bad segments being shown by default but good ones being hidden because bad ones are chosen instead.

>> It's very unlikely that segments with fewer efforts are more often clicked than segments with many efforts.

Two examples from the area I am recently in:

https://www.strava.com/segments/1572764?filter=overall

This segment has two problems: 1)it starts before the round about where you often need to slow down, this is a relatively minor issue though. Worse is that: 2)it ends 250-300m beyond the summit and that 250-300m is quite not so mild downhill section. People are stopping at the top to put the jacket on for example. While this is shown the are multiple very good segments that are hidden (which start at the round about and end at the actual top).

Here are some better segments with more people matching it:
https://www.strava.com/segments/17173280?filter=overall (ends in correct place, starts a bit later but still not ideal)
https://www.strava.com/segments/34652545?filter=overall (starts after the round about, ends on the top - according to Strava guidelines)
Multiple random short sections of the climb are shown but for the long parts Strava chooses to show the worst possible one while hiding decent segments.

https://www.strava.com/segments/13499654?filter=overall

This ends in a random spot in the middle of a climb just after crossroads where you can turn to the town. This means that people who turn right to the town won't have it matched but also why finish the segment if the climb continues? Again, there are multiple segments containing the whole climb or ending just before the cross road to the town which are hidden. They often have way more people matched on them but they are still hidden. Instead what is not hidden are tens of random segments containing pieces of the climb, often 200-400m long. All this and the segments made according to the guidelines are hidden.

Once the segment is hidden people will stop clicking on it (you can't even seen them on mobile for example). I've noticed this segment quality degradation for some time now and it's getting worse. Very often I am wondering "why there is not a segment containing the whole climb) or "why does it end 300 meters beyond the summit in the middle of the a downhill section" or "why does it start before the major crossroads (contradictory to Strava guidelines for segment creation). It's not a problem to show bad segments but quality segments being hidden is an issue and will lead to significant degradation of the major feature of Strava.

As a side note Strava counts number of efforts incorrectly. For example in the first segments I linked to it says there are 3959 times in the overall standings but at the top it says "14,655 Attempts By 2,944 People". Both can't be correct. Maybe there is a some problem counting efforts/matches which contributes to the wrong selection of hidden/shown segments.

[Jan_Mantau]

I made opposite experiences, in my area the hidden segments are usually the bad oder redundant ones. And of course you can hide for you personally whatever you want in the Strava website.

Of course a bad designed segment with i.e. only 10 efforts ever has a 100% top 10 probability and is maybe not hidden but why this should be a problem for anybody I can't see. It's very unlikely that segments with fewer efforts are more often clicked than segments with many efforts.