1483059 - Add snippets targeting for specific Top URLs

Status: RESOLVED WONTFIX

(Firefox :: Messaging System, enhancement, P3)

Description

[Lisa Wright]

We'd like to be able to target snippet users based on whether or not they've visited a "Top URL". e.g. Can we provide a list to you and get info back on (1) did a user visit one of these top 1000 sites, and (2) if yes, which one? (e.g. Facebook, Amazon, etc.)

If we can determine that someone visits Facebook a lot (but does not have it pinned to their newtab - per targeting request Bug 1483055), and they don't have the Facebook Container add-on for instance, we could prompt a very targeted CTA. Or, if a user visits a particular non-default search engine a lot, we could craft messages to help them set it as their default.

Per our previous conversation Kate, it sounded like this one would be a bit trickier (so this will likely be a 64 ask), but that their may be an opportunity here to look at the URL-based recommendation engine that Pocket uses and possibly leverage what they already built? 

Please let us know what else you may need on this request. Thanks!

Comment 1

[Lisa Wright]

Note: This will only be helpful for us if we're able to determine *which* Top URL is visited, as opposed to just y/n they visited one of these Top 1,000.

Comment 2

[Kate Hudson :k88hudson]

There would definitely be some privacy concerns around this, my understanding is that we wouldn't be allowed to report on which specific URL was matched, but only the particular group.

Also, there would probably be some limitations in terms of message size for how large the bucket of URLs would be (1000 seems high to me).

That being said some similar work is being done for the Contextual Feature Recommendation project around URL-specific-targeting that would provide a better solution for this.

Comment 3

[Kate Hudson :k88hudson]

I should mention also that our existing implementation for this specifically allows for matching on a domain and all its TLDs, e.g. "google" would match "google.com", "google.ca", "google.com/foo", etc. not on particular paths. Would that be sufficient for your needs, or do you think you would need the ability to match on more specific URL patterns?

Comment 4

[Lisa Wright]

Matching on a domain is totally sufficient for us, no need to target more specific URL paths. A couple of questions though:

1) When you say existing implementation, do you mean Fx 61? Because as far as I know that capability is not exposed in the admin tool for us to use right now, but I can ask Giorgos to add if it's available.

2) Is that tied to a DNS service? Would I be able to type in any domain I know and target off of it, or is there a particular subset only that we can target?

Comment 5

[Kate Hudson :k88hudson]

1) Sorry, in this case I mean what has been implemented so far for the new version.

2) There is a subset available to us via a Firefox service that's pretty extensive, although for some unusual TLDs (e.g. .club you might want to target them explicitly)

Comment 6

[Giorgos Logiotatidis [:giorgos]]

(In reply to Kate Hudson :k88hudson from comment #5)
> 1) Sorry, in this case I mean what has been implemented so far for the new
> version.

Kate can you please clarify what has been implemented -with link to docs please- and on which version will be available?

Comment 7

[Andrei Oprea [:andreio]]

We currently have topsite targeting attributes [0]
An example message `targeting` would be

>  (["www.foo.bar"] intersect topFrecentSites|mapToProperty('host'))

Check if `www.foo.bar` is in the users' topsites. The API allows for even more fine grained queries checking the exact frecency[1] of those websites.
You can't get the list of topsites you can only provide a list and check if there is a match.
This is available in 64.

[0] https://github.com/mozilla/activity-stream/blob/master/content-src/asrouter/docs/targeting-attributes.md#topfrecentsites
[1] https://developer.mozilla.org/en-US/docs/Mozilla/Tech/Places/Frecency_algorithm

Comment 8

[Jean Collings]

Marshall - please let us know if you're comfortable with Comment 7 for snippets. Thanks!

Comment 9

[Jean Collings]

Pinging Marshall again per his comment:

"For CFR, we are constructing a 'bucket' of AT LEAST 5 URLs that can be targeted against. That mitigates some of the privacy risk because it means we can't say with confidence that the user visited a particular site in the bucket.

Have you spoken to the CFR team about that approach? It looks like it was mentioned in passing in the bug."

Team just needs approval from you that you are comfortable with this approach.