Meta & Children's Online SafetyChat: Is it possible for us to infer which high school they go to from geoAPI
Thread View
████ name redacted. Different bars may be the same person or different people.
quoted in an excerpt in the archive.
Wednesday 4 October 2017
16:07████hey art, the Instagram teen's team is currently doing some data analysis around local for teens.
Specifically we are interested to study teen penetration and engagement on school level in US. So our question on our end is, let's say we have location collection enabled (via geoAPI) and we know someone is a teen. Is it possible for us to infer which high school they go to from geoAPI, purely for analysis purposes, so it's fine if the data is anonymous
16:14████you can map the location to a polygon if there as an entity for the school with a polygon associated to it
16:15████this is just tech feasability commentary i dont know what the privacy implications here would be on person-to-place
16:16████thanks.) 11 and @ thoughts?
16:17████do you have anyone on your team with experience around this
16:17████or a team we could talk to a
16:17████cur tous how FB did their school level analysis
16:20████a good starting point can be to post on fburl.com/LOCQA this can be places data team and/or PVD i can recommend some folks to cc if you dont get a reponse on the post <
16:43████wants to take a lead on that?
16:56████Yah sure I'll take a look
16:57████Just clarification: so there is capability for us to match a location to school then so long as that school is an entity matched to the polygon?
16:57████you also have to run the analysis during school hours haha
16:58████otherwise the focation will update out of school
17:02████Want you mean the data is real time and not stored for offline purposes?
17:03████yup
17:06████Do we not store the location of teens throughout the day across multiple days?
17:06████nope, we get one location and then we delete it
17:06████when we get a new one
17:07████Oh changes so basically if they open the app somewhere that is not the school most recently, then this
17:07████I.e. When they're at home or on weekends
17:07████yeah...
17:08████Is the reasoning for this because of privacy concerns?
17:08████yup, Storage costs + battery drain
17:08████Also they have to actually open the app AT school for us to hope to get the school location
17:08████that too
17:10████Hmh interesting, let me do some more investigation as to how fb does their school analysis to see if we can replicate something on our end with less constraints
17:11████they probably use their self identified school
17:11████" Do we hope to eventually be at a place where we might machine learn a teens important" locations?
17:11████Well from what I last checked that they use explicit naming of schools, but that only gets 40% of coverage
17:11████check out E GEDAPI
17:11████https://oumicern.facebook.com/intern/wiki/GeoAPI/
17:11████But maybe they have other means of proxying from elsewhere?
17:12████maybe ke Tirescage
17:12████assigned school + social connections (friends of friends in that network)
17:13████Cool thanks Do we have the same info on IG users? Or are we hoping to map IG users to FB users to pull this location data?
17:14████so Us Teens aren't on FB
17:14████so we are hoping that the location data would enable us to build our own assignments
17:14████1.e. Structur my this data, since we don't have the structured data of self labeling
17:14████Got it - that's what I meant, so we don't have the same system built in place yet
17:14████yeah, not yet!
17:15████KK - I think this is a conversation worth having in person as well just to flush out the details of how we plan to map teens/non- FB to a location
17:15████Shared an image
17:18████https://our.intern.facebook.com/intern/dex/location-data/work-school-location/?hc_location=ufi
17:18████might be useful
17:55████thanks!