Autocomplete for @-mentions should prioritize users I interact with

[gnprice]

For example, if on chat.zulip.org I type @tim, I get 5 suggestions for users I've never heard of -- more than enough to fill the visible list -- before I see Tim Abbott. Similarly if I type @tab (the start of the localpart/username of his email address), or @abb.

We should follow the same logic the web app has for this. See typeahead_helper.js, particularly sort_for_at_mentioning and its subroutines.

[jainkuniya]

Hmm, I have worked on this logic few months back. That seems be only useful only for smaller realm.

@zulipbot claim

[gnprice]

Great, thanks!

That seems be only useful only for smaller realm.

I think this may be what you mean already, but just to be explicit: I think the logic we need and don't yet have, which the web app has, gets more and more necessary as a Zulip org (realm) grows. Our existing logic is potentially enough for small realms, but users of chat.zulip.org, and I think a lot of mid-size realms much smaller than that one, will be glad to see this implemented 😄

[timabbott]

One thing worth mentioning is that the logic of sort_for_at_mentioning is based on the recent_senders data structure, which comes from data for the most recent 1200 messages that we fetch in the webapp.

We don't do the same fetch in mobile, so while we can probably use a similar algorithm, I'm not sure we have the data stored effectively in the app currently to do this that same way.

[gnprice]

Sure. There are a lot of elements of that algorithm that should be useful independently of each other -- so I'd like to first merge all the ones we can do without additional data.

After that, we can problem-solve how best to do the rest. (For example, there might be appropriate summaries the server could provide us -- potentially even some which would improve ranking quality for the web app beyond what it currently does.)

[borisyankov]

We only fetch the last 100 PM/group messages which can be used for some prioritization, but to a much smaller extent than the web app.
Getting aggregates of last conversations was discussed several times before. If we have this we can use that data.

[chrisbobbe]

I believe this will be fixed with a fix for #3902.