New "Recommended" tag on Chaturbate

[Mila_]

I just noticed this "recommended" tag for the first time ever and I am wondering what it is and what criteria CB uses to recommend cams...



Is this the CB version of "models you may like"? Is it paid spots?
Up until now I had only ever seen "HD", "Chaturbating" and "New" tags. Does this mean the top row will no longer be the people with most traffic?

 

[WebcamStartup]

Interesting. I'd be interested in knowing as well.

 

[RubyClaire27]

It's supposed to help you find more rooms similar to the ones you visit. There's a blog post about it, but I don't find it to be terribly informative.

 

[Mila_]

Thank you @RubyClaire27 I am going to go ahead and post the info here for people to see it easily:



I wonder what they mean with "highlight your special talent in your profile and fans find you faster"... Is there a special field for this or should I simply throw keywords on my bio?

 

[RubyClaire27]

Yeah, I'm curious about the keyword thing too! Is this different than using hashtags in your topic, I wonder? Like I said, not a terribly informative blog post on CB's part.

 

[AmberPrincess]

I still don't have it on my page. Anyone else not have it?

 

[AmberPrincess]

I see it now, a I was looking for a clickable tab. It is overlayed on the first 7 or so thumbnails, where it usually says New, Chaturbating, or HD, it now says Recommended. For me, every group except male has a recommended list, not sure why that is, I enjoy a good cock just as much as a beautiful pussy.

 

[AmberPrincess]

So has anyone broadcasted yet with the new Recommended label? Did you notice a difference in viewers, for better or worse?

 

[LuxVendetta]

Oooooh. I'm excited about this. It's a chance for lesser known models to get some exposure finally.

 

[LuxVendetta]

Thank you @RubyClaire27
I wonder what they mean with "highlight your special talent in your profile and fans find you faster"... Is there a special field for this or should I simply throw keywords on my bio?

I wonder if this means the models with image-based bios won't get many (if any) recommendations. If someone finds out the answer, definitely let us know! We might have to make a list of keywords at the bottom of our profiles from now on.

It could also be based on the models you follow.

 

[WebcamStartup]

This'll be interesting. Just shot an email to CB support asking for more information on the algorithm used to calculate suggestions. Who wants to bet they just shoot me a link to the blog post and then never respond to my follow-up email asking for elaboration on the actual algo itself?

 

[WebcamStartup]

If anyone's curious as to the questions I asked:

Hello,

Just seen the blog post about the new suggested feature Chaturbate implemented. great new feature and I'm excited to see how it works. had a couple questions:

1) What criteria / algorithm is used to determine which broadcasters are suggested to a user?

2) What suggestions would you give broadcasters as far as how they should setup their profiles, ect. In order to better be matched / suggested to qualified customers?

3) Outside of profile information, ect. Are there any actions (tipping, following, entering a room, joining fanclub, ect) that are used to determine suggestions?

4) Does location, language or any demographics such as that get used in determining suggestions?

Any additional information you'd be able to provide would be awesome!

Thanks!

​

 

[WebcamStartup]

I wonder if this means the models with image-based bios won't get many (if any) recommendations. If someone finds out the answer, definitely let us know! We might have to make a list of keywords at the bottom of our profiles from now on.

It could also be based on the models you follow.

I'm hoping a lot of the suggestions will be based off user actions. For example:

Customer A, B, C, D and E all have similar tastes in models. Customer F joins the site and is tipping / following a couple models in the "pool" of broadcasters that A-E also like. CB recognizes that Customer F shares similar tastes as A-E so suggests some similar models that A-E all follow, tip, enter the rooms of. Therefore CB starts recognizing the trend and suggesting models accordingly.

Combine that with a couple demographic-specific signals; age, ethnicity, language, body-type, hair color, ect.

And you'd have more signals than just profile keywords going into the algo.

I don't expect CB to be as complex as Google, but hopefully they've got some interaction signals in there, and it's not just keyword-based.

 

[RubyClaire27]

I wonder if this means the models with image-based bios won't get many (if any) recommendations.

FWIW, Eeeveee shows up as a recommended model for me and her bio is all image based. Now, please explain to me why a model I already follow is in my recommended group. I think the algorithm is still a little glitchy.

I'm hoping a lot of the suggestions will be based off user actions

This is also what I am hoping for! User actions + keywords, as you suggested, would be awesome.

 

[WebcamStartup]

CC: @LuxVendetta @RubyClaire27 @AmberPrincess @supermila ~ You all posted, so assuming you're all interesting in the response

Official response from CB:

Hi and thank you for your questions:

1) The algorithm used is internal and not to be disclosed---however there is a combination of multiple usage factors---so not just one factor.
2) No difference in setup as far as a broadcaster is concerned. This feature is driven by behavior patterns by a viewer/customer.
3) All of the above factor into the "recommended" feature
4) No demographics as far as Geo location or language factors in.


We hope that we have answered all of your questions. Please feel free to follow up if there is any confusion on your end as we are here to assist. Hope you have a great experience on Chaturbate!

Thanks,
Support​

I love how they comment that nothing on the algo is to be disclosed, then go on to confirm that action-signals such as "tipping, following, entering a room, joining fanclub" are all included in the algo. If only getting Google leaks was this easy!

Surprised they're not using age and ethnicity in the algo. Seems like if a customer keeps visiting Asian teens or Latina MILFs they're probably into those ethnicity / age-groups. If I was one of their engineers on the project that would have totally been part of it. That's something real easy to benchmark, too. Same with countries. If a customer keeps visiting Brazilian models, he's probably into South American models.

Well there you go, you got some insight into the algo. It'd be interesting to compare different responses if other models reach out.

 

[AmberPrincess]

I love how they comment that nothing on the algo is to be disclosed, then go on to confirm that action-signals such as "tipping, following, entering a room, joining fanclub" are all included in the algo. If only getting Google leaks was this easy!

Thanks for reposting, I am definitely interested. I didn't see where they said or hinted at any of the above, did I miss something?

 

[WebcamStartup]

Thanks for reposting, I am definitely interested. I didn't see where they said or hinted at any of the above, did I miss something?

My original email:

Hello,

Just seen the blog post about the new suggested feature Chaturbate implemented. great new feature and I'm excited to see how it works. had a couple questions:

1) What criteria / algorithm is used to determine which broadcasters are suggested to a user?

2) What suggestions would you give broadcasters as far as how they should setup their profiles, ect. In order to better be matched / suggested to qualified customers?

3) Outside of profile information, ect. Are there any actions (tipping, following, entering a room, joining fanclub, ect) that are used to determine suggestions?

4) Does location, language or any demographics such as that get used in determining suggestions?

Any additional information you'd be able to provide would be awesome!

Thanks!

​

They confirmed that everything under #3 was included in the algo.

 

[AmberPrincess]

Lol, derp. Thanks

 

[WebcamStartup]

Well, if anyone wants to figure out how to test the different signals, I am more than down to help in the experimenting. I'm sure we could develop some sort of controlled system. Wouldn't be 100% perfect, as there would always be a form of cross-over, but it is possible to test the different signals.

Example: There's the age ranges in the footer. Simply by taking an account that doesn't have any recommendations, performing actions in rooms under a specific age-bracket and benchmarking the new suggestions can help determine whether CB is looking at age.

By letting that account sit for a couple months to see if the suggestions go away, we can determine whether they're looking at lifetime interactions or interactions within a specific timeframe.

 

[papichulo80]

I turned it off and gave feedback that it was annoying. In the two days it was on it only recommended models i'd already visited and knew. It was interesting to see how some models go out of vogue though. There were some guys I thought had quit but they had just dropped so far down I didn't see them anymore.
P

 

[WebcamStartup]

I turned it off and gave feedback that it was annoying. In the two days it was on it only recommended models i'd already visited and knew. It was interesting to see how some models go out of vogue though. There were some guys I thought had quit but they had just dropped so far down I didn't see them anymore.
P

Out of curiosity, did it suggest any models you were already following? You mentioned that it suggested some guys you thought had quit. That kind of implies you weren't following them (or you'd see them online) I'm just curious if whether once followed, they'd be no longer suggested.

By the way people, this is how you reverse-engineer an algo This is basically what I do as an SEO.

 

[AmberCutie]

Sounds like it works nearly identical to MFC's "models you may like" even to the point of directing you to people you already follow/friended/etc.

 

[papichulo80]

It was actually universally people I had visited and not followed.
That's why it was annoying actually. If I had wanted to follow those people I would have. There was a reason I didn't.
Everyone I chat with on there has also turned it off as they had a similar experience. No one wants a model they had a bad experience with shoved in their face every time they log in lol.

 

[TacoBelle]

Question: I am given to understand that "twinks" usually refers to young, slim gay boys, but ... I've almost always heard it used negatively. You know, almost like a slur? Is it an acceptable term to use in the LGBT community?

(Don't judge me for my ignorance... or the semi-off-topic question.)

 

[papichulo80]

Haha,
Yeah it's fine. the only people who get offended are guys who feel wrongly categorized. Its like any label. It saves time but it's potentially unfair and maybe carries negative connotations for some. Its pretty universally used tho and I don't think anyone would go so far as to consider it a slur. Just out of curiosity what prompted the question?
p

 

[Mila_]

I now officially hate this feature I am sad because I really wanted to like it, I was hoping it would give less known yet interesting girls a shot, making CB more democratic and helping me discover cool models. But the suggestions suck.

This is what it recommended to me:

* 2 slots for super popular girls that everyone knows like siswet and chroniclove. They are already on top almost every day so its pretty redundant.

* 4 slots for unknown models with a feature similar to a feature other models I like have. Most of them were uninteresting studio models so there was a reason they werent on top by themselves. One of them was texting on her phone, another one staring at her screen blankly.

* 1 spot for a model I used to follow or a model I visit often already.

This only makes my job harder. Truth is CBs layout is really easy to navigate with the big preview thumbnails. People already find what they are looking for fast because they get to see what is going in the room unlike MFC which is a sea of tiny avatars.

So it makes sense for MFC to recommend models because you may never find them otherwise, on CB it is simply annoying and pushes the best shows 1 row down.

 

[NudeTherapy]

Truth is CBs layout is really easy to navigate with the big preview thumbnails. People already find what they are looking for fast because they get to see what is going in the room unlike MFC which is a sea of tiny avatars.

So it makes sense for MFC to recommend models because you may never find them otherwise, on CB it is simply annoying and pushes the best shows 1 row down.

I think that MFC has an option where viewers can choose the thumbnail previews instead of the avatars, if I'm not mistaken.

 

[NudeTherapy]

I think it can potentially be a good thing and bring more exposure to less known but great models. So far, I'm not impressed by the feature, because the recommended models are mostly top rooms already as well as some boring ones. I think with time they'll be able to improve it if we give them feedback and suggestions.

 

[WebcamStartup]

Any time there's a launch of something like this, there's two things to consider:

A) The results might improve over time, as it requires usage metrics that CB might not have been leveraging prior to this. If they just started collecting and leveraging this data, the more it gets collected, [in theory] the better the results.

B) I'm assuming this feature is still very much in beta and will be tweaked as things move along. The best thing models and customers can do to help this is share their feedback with CB, and hopefully it gets forwarded to the correct parties within the company.

 

[billymac]

As a user and not a broadcaster I find the recommended tag very confusing
A) It does recommend people I follow not needed
B) It shows me girls who are into Anal which I totally hate if I am in a room and the girl starts anal I leave
C) I get recommended Asian Girls whilst I have nothing against them I do not recall ever visiting an Asian room
D) I prefer more Mature Models and lesser known ones but they are never recommended to me.
I suspect CB see the models who make the most money for them and slap a "Recommended" tag on.Nearly 99% of the ones recommended to me have well over a 100 thousand followers plus