Page MenuHomePhabricator

Hackathon: track how long people are looking at a given image
Closed, ResolvedPublic

Description

Migrated from: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/1001

Narrative

GLAMs have a long-standing request of getting reports on “views” for their content. Following a discussion with Erik Zachte we realized that the issue in Media Viewer is non-trivial. Since we allow people to browse images sideways very quickly, it means that some “views” for images last a very short time, and probably shouldn’t be counted as views per se. The idea is thus to not only measure when an “view” in Media Viewer happens, but also measure how long the user has had the image open for, and ideally substract the time they spent looking at another tab, etc. Then it will be up to the GLAMs to sort that data and remove short skip-like views from the totals if they wish to.

I don’t think we should be the ones making the determination of where the duration limit stands between a skip-like view and a proper view. We’re better off recording it all, this way everything is stored and the criteria can be applied later.

The plan is to issue HEAD requests to the thumbnail’s URL with extra tracking parameters. This is because not only we don’t want to overload EventLogging, since there won’t be any sampling (every request is recorded), but also because the way the non-Media Viewer view data will be compiled will be based on server access logs.

In terms of sending the requests in time in case of browser closure, we’ll have to wait until we see the results of #998 to see what percentage of views we’d be losing by relying on async calls made onbeforeunload.

Acceptance Criteria

Related Bugs

Related Stories

Related Changesets

Event Timeline

MingleTerminator raised the priority of this task from to Low.
MingleTerminator added a project: Multimedia.
In mingle on 2014-11-16 at 12:44:28, @Gilles wrote:

https://gerrit.wikimedia.org/r/#/c/173640/

Not really blocking, at least not for the first version or per-file view stats.