Suggesting tiles by region with finer granularity than country

[Ed Lee]

[ For context around Suggested Tiles, please read http://ed.agadak.net/2015/04/whys-and-hows-of-suggested-tiles ]

So far Tiles have been served to users at a country level. This is good enough for content that should reach a broad audience, but this also prevents us from providing highly relevant content to smaller audiences.

For example, if we want to suggest a Mozilla recruiting tile to developers for our San Francisco / Mountain View offices, right now we would have to show it to all developers in the US. This would probably lead to disappointment for people interested in working remotely from other states to only find out the position is local.

We're planning on using finer granularity, and we're curious to hear people's thoughts on what's the appropriate level as there's potential privacy concerns.

One proposal is state level for USA. The smallest state by population is around 600k people for Wyoming [1]. There's 50 states, but that might not be the right granularity either as large states like California could be split into northern and southern audiences. Another issue is metropolitan areas near borders such as New York can span several portions of multiple states.

Another proposal is Designated Market Area (DMA), which I believe was created to address the above states concerns and has already been widely used. There's 210 of these with sizes that vary from very large nearing 20m for New York [2] down to Glendive, MT with 10k.

For either proposal, we don't need to use all states or DMAs. For example, if we determine a minimum audience size of 1m, we can use some of the larger ones and skip the smaller ones.

Do people have suggestions on what's the appropriate granularity?

Ed Lee

[1] http://en.wikipedia.org/wiki/List_of_U.S._states_and_territories_by_population
[2] http://www.tvb.org/markets_stations#!id=137&type=market

[Chris Hofmann]

I think its hard to answer the question about appropriate granularity
without out looking at the larger context.

You've made the assertion that "Tracking is complex and overloaded term"
and that what tiles is doing "shouldn't be considered tracking."

I think those the assumptions to be looking at and examining.

The current ad network mechanism use a variety of techniques to target
content to specific users.

It would be great to outline those, then show places where tiles matches or
differs from those techniques.

What's the level of location targeting that are used in a sample of Ad
network targeting? I'm guessing the ad networks ad tracking and targeting
systems are trying to get as specific and precise as possible on location
since that delivers the highest value for them. Maybe it delivers the
highest value for users as well, but that value may, or may not, be
delivered based on the they kind of tile content they are looking at or
interested in.

If one of the things we were trying to do is to put users in more control
over what data gets shared with advertizing networks and what does not then
we ought to be looking at ways to put user in decision path for the
location questions that you are asking. It gets complicated in the UI
but have we thought about ways the user can share their choice about what
level of location precision gets shared with the tiles system, or maybe
inffering this from other location precision selections they might have
made? We are doing that in other places in the UI with the location API.
Maybe if a user always answers yes to the "this page wants to know you
location?" question then we can assume tiles can get very precise.

If the new tile page wants to comply with the location API then maybe it
should be asking the "this page wants to know you location?" question.

-chofmann

> _______________________________________________
> dev-planning mailing list
> dev-pl...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-planning
>

[Ed Lee]

On Friday, April 24, 2015 at 10:58:46 AM UTC-7, Chris Hofmann wrote:
> If the new tile page wants to comply with the location API then maybe it
> should be asking the "this page wants to know you location?" question.

Firefox is not sending geolocation data to our servers, so that's part of the reason why there isn't a location prompt. The servers are guessing a location based on IP addresses.

Using the geolocation API would provide much more detailed location information (lat/long). Whether we should use the API and if Firefox needs to ask the user's permission from a chrome context is a whole separate conversation that I don't think we need to get into now. (Although using the API would address some issues like VPN IP addresses confusing our servers.)

> have we thought about ways the user can share their choice about what
> level of location precision gets shared with the tiles system

That's a fair question as one of the main reasons why we're getting into the advertising space is to give better controls to users. Even though technically, IP addresses are automatically sent to our servers (i.e., no real choice to not send other than turning things off), we can indeed provide choice to users that we would honor from the server. But we also don't want to confuse users with too many choices, so we'll need to prioritize which finer grained choices should be made available to users vs providing coarse "turn off all."

> What's the level of location targeting that are used in a sample of Ad
> network targeting?

I would guess existing ad tech will try to use as much data as possible from various sources. For example, Verizon as an advertiser could know all the locations you've been to with your smartphone, and Google as a publisher wanting to show an ad could use its data about previous maps searches, and this is all combined with the current request that includes your IP address.

Tiles can't do that because we don't use unique identifiers, so data can't be linked up to those other sources.

But even if we strip away all that additional data and focus on just IP address mapping to location, I'm pretty sure ad networks use as detailed resolution as possible they're confident of the data. If there's data (even if inaccurate), someone will use it.

Here's a tool to check what data various geolocation providers think of your current IP address: http://www.iplocation.net/ A quick glance shows Country, Region/State, City, ISP, Latitude, Longitude, Time Zone, Postal Code, Area Code.

[Patrick Finch]

On 4/26/2015 4:43 AM, Ed Lee wrote:
> On Friday, April 24, 2015 at 10:58:46 AM UTC-7, Chris Hofmann wrote:
>> have we thought about ways the user can share their choice about what
>> level of location precision gets shared with the tiles system
> That's a fair question as one of the main reasons why we're getting into the advertising space is to give better controls to users. Even though technically, IP addresses are automatically sent to our servers (i.e., no real choice to not send other than turning things off), we can indeed provide choice to users that we would honor from the server. But we also don't want to confuse users with too many choices, so we'll need to prioritize which finer grained choices should be made available to users vs providing coarse "turn off all."

I believe that mobile has created a mindset where users expect to be
asked if their location is being used. And irrespective of what I
believe, Brett Gaylor's recent online documentary used the ability to
geolocate the user by IP address as one of the first, unnerving signals
about tracking. For those reasons, I believe this is an important
control for a user to have.

Patrick

[Gervase Markham]

On 26/04/15 03:43, Ed Lee wrote:
> Using the geolocation API would provide much more detailed location
> information (lat/long). Whether we should use the API and if Firefox
> needs to ask the user's permission from a chrome context is a whole
> separate conversation that I don't think we need to get into now.
> (Although using the API would address some issues like VPN IP
> addresses confusing our servers.)

One way of dealing with that might be to call the API but not send the
location data to our servers. Instead, you could add an "uncertain
location" flag to the request where the GeoIP data didn't match the
geolocation data, so the server knew it might not want to use geotargetting.

Gerv