Closed Bug 937694 Opened 10 years ago Closed 7 years ago

[dev] include (# of app ratings) in api for E.Me results selection and rankings

Categories

(Tracking :: User Story, defect)

x86
macOS
defect
Not set
normal

Tracking

(Not tracked)

RESOLVED WONTFIX

People

(Reporter: cgalimidi, Assigned: clouserw)

References

Details

Must decide public vs. private data

- private data : downloads
- public data : ratings

options:
- # of downloads (pro: accuracy for e.me's algorithm con: private data - would need to either make this data public or create new data dump that is private and for E.me only)

- app ratings (pro: already public data, con: data must be added to data dump)

- % of total downloads in marketplace (pro: accuracy of data for e.me, con: data does not yet exist in marketplace
I think that the use of number of app ratings is a decent approximate indicator of download numbers that is publicly available. I think we can try this to start.
Assignee: wmaggs → clouserw
Summary: [Product] Determine data to use for E.Me app results selection and rankings → [dev] include (# of app ratings) in api for E.Me results selection and rankings
Blocks: 881063
No longer blocks: 873543
If I'm not mistaken, app rating's span "all time" which makes them a week signal.
Moreover, the data must be related to a term or else it's not enough data to conclude from.
I don't understand how "all time" makes ratings a weaker signal than number of downloads? That is also an accumulative number, no? 

I'm not suggesting that number of ratings is as good as number of downloads, but that it is an okay approximation of the same signal. Since we already have number of ratings built in to the API and publicly available it would be much less work on the Marketplace side to start with this number. We can of course extend later if needed.
Depends on: 938855
It's different in the sense that clicks and downloads is a statistic spanning 100% of the app users, as opposed to app rating which is substantially less (on average around 1% of the users). Also, it's inherently biased.

"All time" data is not only weak as much as it is misleading. If an app was really popular a year ago, it might outrank a more relevant and popular app that came out last month. The download data we've asked for would be relevant to the time frame of the dump.

In any case, we've come to an agreement about the data with Bill and Wil.
Ratings data is already in the API, albeit at a separate URL. It shouldn't be much hassle to ping the ratings resource for any given app when you need updated info. You probably don't need to ping it very often anyway, since rating data is unlikely to quickly shift: popular apps will continue to have large numbers of relatively stable ratings and unpopular apps are likely to not acquire many new ratings.
No worries, Ran and I have got this bug.  What comment 0 says and what we're actually doing aren't really the same thing, but that's how it goes.  Cheers
(see bug 938879 for what we're actually doing)
Blocks: 948661
No longer blocks: 881063
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.