As discussed, the insight we most want is how frequently were users seeing/engaging with duplicated images and how that affected task completion rates, etc.

Looking at Android image engagement by unique user we saw the following - this counts the image filename shown by each decision action of Accept, Reject or Skip

iOS also saw duplicated image engagement but results are not 1:1 comparable with Android because image Filename is only recorded when user submits a reject action OR is issued a warning for clicking under 10 seconds,
instrumentation for filename was not as thorough on iOS because MVP. Considering the tech is the same behind both platforms we should use the Android duplication rate when assessing impact of duplicated images.

Looking at qualitative data where users can enter text to indicate rejection reasons reports of image already being in article were 13.1% of responses (out of 213 responses).

Data

Will be adding Android/iOS Task Completion Rate by matching dates (as requested) to ticket next.

'Clean' Event Analysis for Android: Data

'Clean' Event' Analysis for iOS: Data

• Let's discuss how we want to organize this data to reduce cross-duplication, I added iOS results to Android sheet and Android results to iOS in sheets but we probably want to update decks or final outcomes.

• Also for iOS I had to do a workaround way to get clean data because we aren't tracking image filenames on iOS as explicitly as we are on Android - should we add that value to all image selection events so we can track more closely in future? To be compatible with Android we would need to add image file name to data instrumentation for actions 'suggestion_accept', ' suggestion_reject', 'suggestion_skip' and action = 'impression' AND active_interface = 'recommendedimagetoolbar'

After discussion we have decided against adding any further analytics to iOS Image Recs since we only need it for yroubleshooting problems like duplicate images which we can also get from Android data (same API).

Thanks @SNowick_WMF!

Let's discuss how we want to organize this data to reduce cross-duplication, I added iOS results to Android sheet and Android results to iOS in sheets but we probably want to update decks or final outcomes.

Thanks, could you update the decks (Android, iOS)? I'll update the iOS narrative slides & project pages.

Also for iOS I had to do a workaround way to get clean data because we aren't tracking image filenames on iOS as explicitly as we are on Android - should we add that value to all image selection events so we can track more closely in future? To be compatible with Android we would need to add image file name to data instrumentation for actions 'suggestion_accept', ' suggestion_reject', 'suggestion_skip' and action = 'impression' AND active_interface = 'recommendedimagetoolbar'

We discussed in Data sync that we don't need to add this tracking for iOS.

Image recs results Narrative have been updated.

For image suggestion accept rates, reject rates, and skip rates all look comparable for iOS & Android, and don't necessarily point to repeated images being an issue. However we are still seeing a higher bounce rate for iOS compared to Android.

In looking at the updated rejection reasons for 2024-06-12 - 2024-08-22, rejections mentioning the infobox and image already being in article have declined compared to the analysis from 2024-05-11 to 2024-06-10. This is most likely due to the expanded check added on iOS for existing images, and the refresh of the mediawiki dumps on July 25.
https://docs.google.com/presentation/d/1HwOh92Laj6RArUknSeqGtDV3-_xdn1GMt11ymjVrJkI/edit#slide=id.g2f59ed3f15e_0_3

@HNordeenWMF I have updated the Suggestion Accept Rate slides on both Android and iOS Image Recs decks and indicated data is newer, revised clean data.