Instagram explains how it uses AI to choose content for your Explore tab

https://ai.facebook.com/blog/powered-by-ai-instagrams-explore-recommender-system/

Trying to make sense of this, what’s everyone’s thoughts on it?

9 Likes

What has been observed for some time already is that your account will be categorized. And the more consistent the machine learning model identifies that your content matches up with what it should be, the better you are getting ranked towards your competitors.

If you post many different pictures it’s gonna be hard for the machine learning model to categorize you. And categorized content is “safe” content (normally). At least based on the article you posted we do have that confirmed now.

3 Likes

This is interesting. Especially when it mentions that if a user likes similar images within the same session it will essentially affect your categorisation. This brings me to one conclusion that since IG is a visual platform, for a niche based travel account you should basically imitate/somewhat replicate the first glimpse of an image - I.e a landscape photo, having the same colour palette/edit that you mainly see on the explore page, but a different landscape… for a user who is browsing the explore page and liking these “high quality” images on the explore page, it’s likely that if you have the same colour palette - the eye will most likely be attracted to it and therefore they click on it. Right? This could work well for photographer/content based accounts of their cover image is “mainstream”

Thoughts?

1 Like

Good find. Read through it all, it’s interesting to see that our evaluation of the explore page ranking system match that of this report. I felt like they didn’t want to actually go too in depth though, perhaps they didn’t want anyone abusing any loopholes ect. But yeah, good report, the fresh content point hit home.

Every time the same piece of content is re-used the possibility of ranking on explore with it decreases.

1 Like

So this means that the F/UF method reduces the possibilities to have the content in the explorer page?

1 Like

it does not mean that. It basis this upon each user – not on insta as a whole. person A might have seen it while person B did not. The second time it is posted, A will not see it while B will.

2 Likes

It’s like that since forever.
I have accounts that never hit explore,but they grew with f/uf.
Once i stoped automation they hit explore way often and they blew up.

2 Likes

lol i was just reading thanks for share anything about the feed ?

also i want to ask you guys if i mimic a specific account(bio,captiom,similar content,numer of posting,tiiming of posting) will i have more chance to figure on explore page of his followers

but on the article they didn’t talked about f/uf no ?

can you explain why because on the article they said that if a content is similar to the one a user liked or saved before he will potentially figure on his explore

1 Like

yes they don’t talk directly about that but they talk about the content that you like it etc…so if your page is about food and you are liking photos about toys I think with that behavior will difficult to show your content on the explorer page

1 Like

you mean that follow an account have a tiny impact in comparison to like and saves right ?

1 Like

Interesting to see how this relates to my personal feed. I really have a mix of everything there, and some of the pics are nothing related to pics I’ve liked before. There are many that are similar to ones I’ve viewed but not liked so I guess that is taken to account, too with the AI

1 Like

Saves are playing a huge role in how the explore page is built for user X as well. When you compare the posts you saved and your explore page you see how much weight it has. For good reason, save is a 100% private action, nobody else can see who saved which post… except the Instagram algorithm. It’s the same psychology like watching someones behavior instead of asking the person about it. Even if someone wants to answer 100% the truth, the real behavior “when nobody is watching/caring” is more exact.

3 Likes

i wonder does IG demote ( make it harder ) if one used hastags and then took it off – how it effects viralbility. ( is there such a word? )

1 Like

This isn’t currently happening. I repost my best posts monthly and they do consistent numbers every time.
If you want to “game” the explore, just look at your own explore page.

1 Like

the whole article is very good – one area caught my eye the most.
“There are many different ways people can engage with accounts and media on Instagram (e.g., follow, like, comment, save, and share).”

follow the account –
save ranks higher than likes.

This leads me think for ER pushing – follow the account – save the post, liking is optional. ( saves you from 24 hour like blocks no?)

2 Likes

Saves are basically categorized super likes.

Users save different things to different collections, which gives the algorithm a lot more context as to what that kind of content actually is.

They then use this data to deliver better categorized content to all the users on their feed.
The better they curate this content and deliver exactly what you want, the more likely you are to spend time on the platform.

Saves are lot more valuable for the algorithm learning process, but it’s also programmed to value as of higher importance when picking the best content because it’s a more meaningful action from the user.

Pinterest is basically built on top of these saves & collections system, to deliver the most relevant content to the user, and for a long time they had a much better “delivery system” than Instagram.

6 Likes

Key point in the ‘possibility of ranking’. Doesn’t mean that the content can’t go as viral as the original post given the right circumstances (size and engagement of posting account ect.), but it means that the more the content is circulated it becomes increasingly harder to reach as many people (disregarding follower base engagement).

1 Like

no i don’t think so i think reposting content if done well (changing md5 …) can be good in the short and long term too

1 Like