Accessing Your Private Knowledge. The Intensive and Typically Stunning Knowledge… | by Jeff Braun | Aug, 2023

I take advantage of an Apple Watch to trace well being and health knowledge. That knowledge is accessed individually from all different Apple knowledge that you simply request from the overall Apple web site. Due to this, I present two separate Apple entries within the above tables and talk about the Apple knowledge in two matters under.

The quantity and kind of information you obtain will rely on how extensively you have interaction with a selected firm. For instance, I take advantage of social media sometimes. The moderately modest quantity of information I acquired from Fb is due to this fact not stunning. In distinction, I take advantage of Apple services rather a lot. I received a broad vary and enormous quantity of information from Apple.

Understand that when you’ve got a number of identities with an organization, you’ll have to request the information for every id. For instance, if Google is aware of you by one e-mail tackle on your Google Play account and a unique e-mail tackle on your gmail account, you’ll have to do an information request for every tackle with a purpose to get a full image of the information Google shops about you.

Within the desk above I present hyperlinks that I used to request knowledge from my goal firms. The hyperlinks are present as of the publishing of this text however might change over time. On the whole, yow will discover directions for requesting your knowledge on the “Privateness”, “Privateness Rights” or comparable sounding hyperlinks on an organization’s house web page. These hyperlinks continuously seem on the very backside of the house web page.

You normally must learn via documentation describing your privateness rights and seek for the “Accessing Your Knowledge”, “Exporting Your Knowledge”, “Knowledge Portability” or comparable subject to get a hyperlink to the precise web page for requesting your knowledge.

Lastly, the method for requesting your knowledge, the timeliness of the response and the standard of documentation you obtain explaining the information varies vastly from one firm to the subsequent. Be affected person and persevere. You’ll be rewarded with a wealth of information and information very quickly.

My Knowledge Insights

Here’s a assessment of the information information that I acquired from every firm together with just a few observations after analyzing the extra attention-grabbing information. I additionally level out some alternatives to do extra in-depth knowledge evaluation and modelling with the information from these firms.

Fb

My obtain from Fb included 51 .json information, excluding the quite a few .json information containing particular person message threads from my Fb Messenger account. Fb offers some high-level documentation for its information on the obtain web site.

Knowledge on my Fb login exercise, units that I used to login, estimated geographic location of my logins, and comparable administrative-type knowledge about my account actions seem throughout a number of information. Nothing in these information is especially attention-grabbing, although I’ll say that the situation knowledge appeared surprisingly correct, given it was was usually inferred from my IP tackle on the time of the recorded exercise.

The really attention-grabbing knowledge began to look in a file that tracked my off-Fb app and internet exercise. I can see how the information in that file, coupled with the information that Fb already has from my Fb profile, paint a demographic image that lead to me being chosen as a goal by specific Fb advertisers. The off-Fb file begins to present you a way for the way the profiling and promoting course of works at Fb.

Let’s check out the file. It’s named:

“/apps_and_websites_off_of_facebook/your_off-facebook_activity.json”

It comprises 1,860 data of actions I took on 441 totally different non-Fb web sites over the previous two years. Right here is an edited pattern of the web sites and motion varieties it data:

A number of know-how and journey associated websites rise to the highest of my off-Fb exercise listing. Now let’s have a look at my demographic profile.

The file named:

“ads_information/other_categories_used_to_reach_you.json”

comprises a listing of demographic classes that Fb has assigned to me primarily based, I assume, on my Fb profile knowledge, my Fb buddies, my exercise on Fb, and my off-Fb app and internet exercise. Right here is an edited pattern of the demographic classes:

A lot of the classes above are primarily based on my profile, my system utilization sample, and my buddies. The “Frequent Vacationers” and “Frequent Worldwide Vacationers” classes come, I assume from my off-Fb internet exercise. To date, this all checks out.

Lastly, there’s a file named:

“ads_information/advertisers_using_your_activity_or_information.json”

The “advertisers_using_your_activity_or_information” within the file title leads me to imagine that Fb makes my knowledge accessible to its advertisers who in flip use it to focus on me with adverts via Fb. This file, then, lists these advertisers who displayed an advert to me, or who at the least thought of doing so primarily based on my knowledge.

The file contained 1,366 totally different advertisers. Here’s a small pattern of these advertisers:

Journey websites, retailers, tech firms, health facilities, automotive restore firms, healthcare insurers, media firms (who symbolize advertisers), and different companies seem within the listing. It’s all kinds of organizations, however in lots of cases, I can see how they relate to me, my preferences and my habits.

Different information within the Fb obtain embody Fb search historical past, search timestamps, and browser cookie knowledge.

Google

Google’s export facility is cleverly named “Takeout”. The Takeout internet web page lists all the assorted Google providers for which you’ll be able to request your knowledge (gmail, YouTube, search, Nest, and many others.) It additionally exhibits the information accessible for every service, and the export format for every file (json, HTML, or csv). More often than not, Google doesn’t provide you with a selection of export format for particular person information.

Google does an honest job of offering a high-level overview of the aim of every file. There may be, nonetheless, no documentation for particular person fields.

I acquired 94 information in my extract. As with Fb, there have been the traditional administrative information associated to system info, account attributes, preferences, and login/entry knowledge historical past.

One attention-grabbing file is the one titled ‘…/Adverts/MyActivity.json’. It comprises a historical past of adverts offered to me because of searches.

Some entries within the Adverts/MyActivity file have URLs containing a clickserve area for instance:

Per Google’s 360 ads website, these are adverts from an advert marketing campaign being performed by one among Google’s advertisers, served to me because of some click on exercise I did. The file doesn’t give any info on which motion I took that prompted the advert to be served.

The ‘title’ column within the file distinguishes between websites “Visited” and matters “Searched”. The “Visited” data all have “From Google Adverts” within the ‘particulars’ column (see instance above), main me to imagine that Google served an advert to me in response to me having visited a selected web site.

The “Searched” data present websites I visited instantly (macys.com, yelp.com, and many others.) The ‘particulars’ column exhibits these websites whereas the ‘title’ column apparently exhibits what I looked for on these separate websites. For instance,

One different file I discovered attention-grabbing is named ‘…/My Exercise/Uncover/MyActivity.json’. It’s a historical past of the subject strategies that Google offered to me via its “Uncover” function on the Google app (previously the Google Feed function — extra on Uncover here.) Uncover matters are chosen primarily based in your internet and app exercise, assuming you give Google permission to make use of your exercise to information Uncover matters.

Although I don’t enable Uncover to make use of my internet and app exercise, Uncover nonetheless offered some subject strategies related to me. Right here is an edited pattern of the matters offered most continuously over a number of days:

We see right here the recurring themes of know-how and journey, together with a brand new theme we will even see within the Apple information — music!

Google contains in its obtain a number of information monitoring exercise historical past throughout Google’s services. For instance, I acquired historical past for my visits to the builders.google.com and cloud.google.com websites for coaching and documentation sources. No compelling insights got here from this knowledge, nevertheless it did remind me of matters I needed to revisit and research additional.

Different historic knowledge within the extract included searches and actions carried out inside my gmail account; search requests for pictures; locations searched, instructions requested, and maps considered via the Google Maps app; searches carried out for movies on the internet (exterior of YouTube); searches performed on and watch historical past for YouTube; and contacts I retailer with Google, presumably in gmail.

Not like Fb, Google doesn’t present any info on a demographic profile that Google has constructed for me.

Notice you could view your Google exercise knowledge throughout its merchandise and apps by visiting myactivity.google.com:

When you can’t export the information from this web site, you may browse the information, permitting you to get a way for the kind of knowledge it’s possible you’ll need to export via the Google Takeout web site.

Microsoft

Microsoft allows you to export a few of your knowledge via the Microsoft Privacy Dashboard. For particular person Microsoft providers not accessible on the Dashboard (for instance, MSDN, OneDrive, Microsoft 365, or Skype knowledge) you need to use hyperlinks within the “Easy methods to entry and management your private knowledge” part of Microsoft’s privacy statement page. The identical web page directs you to an online kind you may submit if you’re searching for knowledge that’s not accessible by any of the above strategies.

I selected to export all knowledge accessible via the Privateness Dashboard. This included looking historical past, search historical past, location exercise, music, TV and films historical past, and apps and repair utilization knowledge. I additionally requested for an export of my Skype knowledge. My export included 4 csv information, six json information, and 6 jpeg information.

No file documentation was included within the export and none was discovered on the Microsoft web site. The sector names within the information are, nonetheless, pretty intuitive.

A couple of attention-grabbing observations from the Microsoft information:

The file ‘…MicrosoftSearchRequestsAndQuery.csv’ comprises knowledge from searches I carried out over the past 18 months together with search phrases and, apparently, the location that I clicked on, if any, from the search outcomes. It appears to be like like the information was just for searches that I did via Bing or Home windows Search.

Primarily based on the information, it seems I clicked on a hyperlink within the search outcomes solely 40% of the time (347 out of 870 searches carried out.) From this, I assume that the searches for which I didn’t click on on a hyperlink have been both poorly crafted, returning off-topic outcomes, or I’ll have been in a position to get the reply I needed simply by studying the hyperlink previews within the search outcomes. I don’t recall having to continuously redo search phrases, and I do know I usually see the reply I would like proper in a hyperlink preview, since lots of my searches are for reminders on coding syntax. Both method, I used to be a bit shocked on the 40% click-through charge. I’d have anticipated it to be a lot greater.

Not a lot attention-grabbing was is within the Skype knowledge. It contained the historical past of in-app message threads between me and different Skype assembly contributors. Additionally included have been .jpeg information with pictures of contributors from just a few of my calls.

Apple Health

I needed to entry my Apple well being and health knowledge individually from the opposite knowledge that I exported from Apple. The well being and health knowledge are accessed from the Well being app on the iPhone. You merely click on in your icon within the higher right-hand nook of the Well being app display. It takes you to a profile display and also you then the press on the Export All Well being Knowledge hyperlink on the backside of the display:

My well being export included slightly below 500 .gpx files totaling 102 meg. They include route info from my recorded exercises over the past a number of years. One other 48 information contained 5.3 meg of electrocardiogram knowledge from self-tests that I carried out on my Apple Watch.

The file named ‘…/Apple/apple_health_export/export.xml’ comprises the actual attention-grabbing knowledge. For me, it’s 770 meg with 1,956,838 data overlaying a number of totally different well being and train measurements for roughly seven years. A number of the exercise varieties measured are as follows:

Notice that the frequency at which Apple data knowledge varies by exercise kind. For instance, Lively Power Burned is recorded hourly whereas Stair Ascent Velocity is recorded solely when going up stairs, resulting in the big distinction in remark counts between these two exercise varieties.

The info recorded for every remark embody the date/time on which the remark was recorded, the beginning and finish dates/instances of the exercise being measured, and the system that recorded the exercise (iPhone or Apple Watch).

In his glorious Medium article “Analyse Your Well being with Python and Apple Well being”, Alejandro Rodríguez offers the code that I used to parse the xml within the export.xml file and create a Pandas knowledge body. (Thanks Alejandro!) After deciding on a one 12 months subset of the information and grouping and aggregating it at day and exercise kind ranges, I found some attention-grabbing issues.

As I suspected. my common exercise ranges have been totally different for days after I was travelling in comparison with days after I was in one of many cities I name house (Austin or Chicago). To see this, I had to make use of the latitude and longitude knowledge from the .gpx train route information talked about earlier. That allowed me to find out which of the routes have been in a house metropolis and which occurred whereas I used to be travelling. I then merged that location knowledge with my exercise abstract knowledge. This was then additional summarized by exercise kind and site (house metropolis or travelling). Right here is the sample that merged:

Whereas in Chicago, I’m in an condominium constructing with an elevator, so the large decline in common flights climbed was not a shock. What was stunning was the rise in exercise ranges for Chicago versus Austin. My train routine could be very comparable in each places, but I do extra work in Chicago. I feel I can attribute this to the truth that I stroll to extra places in Chicago, moderately than driving more often than not. Clearly, I have to up the quantity that I train in Austin.

Recognizing traits just like the one above, which you can not see in the usual charts of the Apple Well being app, are a fantastic use for the well being knowledge.

The info can also be nice for modeling, given it is vitally full and usually clear. Right here, for instance, is a time sequence forecast of my train minutes primarily based on a one 12 months interval utilizing Fb’s Prophet mannequin:

Right here is identical forecast, however with annual seasonality enabled and weekly seasonality added manually primarily based on my location (Austin, Chicago or travelling):

The default weekly seasonality mannequin above (first plot) does a worse job of becoming the coaching knowledge than the mannequin with customized seasonality phrases added (second plot). Nonetheless the default seasonality mannequin is much better (although nonetheless not nice) at predicting future values of train minutes. Evidently, hyperparameter tuning would assist enhance these outcomes.

That is only a pattern of the kind of modeling you may experiment with utilizing your well being knowledge. Do you need to strive utilizing very granular time-series knowledge? Take a look at the exercise routes information. They’ve observations for every second of your recorded exercises with latitude, longitude, elevation and velocity fields.

Apple — Non-Health/Well being

You request a obtain of all of your non-fitness/well being knowledge from Apple’s principal web site. For me, that amounted to 84 information, principally .csv and .json information together with just a few .xml information. I additionally acquired lots of of .vcf information, one for every of the contacts I’ve on my Apple units, In complete, I downloaded 68meg of information, excluding the .vcf information.

Apple stands out in that it offers complete documentation for every of the information information. It contains explanations of every subject, although some definitions are extra useful than others. The documentation helped me interpret just a few knowledge information that regarded intriguing.

As with most different exports, Apple’s information included the traditional administrative knowledge, together with issues corresponding to my preferences for numerous apps, login info and system info. I didn’t discover something exceptional in these information.

There are a number of information associated to Apple Music, one of many providers to which I subscribe. Information with titles like:

• “…/Media_Services/Apple Music — Play Historical past Every day Tracks.csv”;
• “…/Media_Services/Apple Music — Just lately Performed Tracks.csv’’; and,
• “…/Media_Services/Apple Music Play Exercise.csv”

include info corresponding to:

• date and time a track was performed;
• play period in milliseconds;
• how every play was ended (for instance, it reached the top of the monitor, or I skipped previous the track);
• the variety of instances the track has been performed;
• the variety of instances the track was skipped;
• the track title;
• the album title, if any;
• the track’s style; and,
• the place the track was performed from — my library, a playlist, or one among Apple’s radio channels.

My information contained between 13,900 and 20,700 data relying on the aim of the file. The info lined almost seven years of track performs.

Apple captures a spread knowledge on how track performs are ended, most likely for functions of recommending different songs to me. Tune play termination causes embody:

For functions of the analyses I present under, I centered on the ‘NATURAL_END_OF_TRACK’, ‘TRACK_SKIPPED_FORWARDS’, and ‘MANUALLY_SELECTED_PLAYBACK_OF_A_DIFF_ITEM’ finish causes.

Typically I’ll repeat a track that I like. One query I had was “Do I play favourite songs obsessively, over and over?”. I answered that query utilizing the Apple knowledge:

The desk above summarizes the variety of instances I’ve performed some favourite songs (‘Play Rely’) and the quantity days over which I performed the songs (‘Performed on Variety of Days’). It appears to be like like I usually play a track solely as soon as per day. Additionally, provided that the play depend is lower than the day depend for some songs, I have to skip some favorites if I’ve heard them too many instances lately or if the track doesn’t match my temper on the time. So, no obsessive enjoying right here!

I additionally questioned if I favor sure varieties of songs on totally different days of the week, totally different instances of the day, and even totally different months of the 12 months. My instinct says that I do. With the Apple knowledge, it was straightforward to visualise the genres I performed at totally different instances. Right here, for instance, are the genres I performed most continuously throughout every month of the 12 months:

I clearly favor rock songs, with various and pop music added for some occasional selection. July and August appear to be the months after I want the variability.

That stated, I used to be shocked at simply how a lot rock I appear to play. Admittedly I like it. However I additionally imagine I’ve fairly broad style in music.

So, I questioned the accuracy of the style assigned to the songs in Apple’s knowledge. For one factor, 10,083 of the 22,313 track performs in my file had no style assigned to them. Additionally, there seems to be lots of overlap within the genres assigned. For instance, “R&B/Soul”, “Soul and R&B”, “Soul”, and “R&B / Soul” are all genres assigned to totally different songs in my knowledge. The totals within the chart above will surely be totally different if I recast the genres of all songs to make use of a constant style naming scheme.

Fairly than make investments the time to replace the genres, I made a decision on one other take a look at to find out if the traits within the chart really symbolize my enjoying patterns. Since Apple contains track play ending causes within the knowledge, I regarded to see if I are likely to skip previous rock songs extra continuously than different genres, indicating that I attempt to play different genres when too many rock songs are being performed.

Because it seems, I don’t skip previous rock songs considerably greater than I skip previous different genres that I take heed to continuously. I’ll must face it — I’m a die-hard rock fan.

One other attention-grabbing file is named “…/Media_Services/Shops Exercise/Different Exercise/App Retailer Click on Exercise.csv”. Whereas I don’t analyze it right here, I like to recommend it to anybody who desires to get a way for the kind of knowledge a retailer might need to monitor for exercise on their web site. For me, it included 4,900+ data with detailed historical past of my exercise whereas within the app retailer and, apparently, in Apple music. Forms of actions I took, dates/instances, A/B take a look at flag, search phrases, and knowledge offered to me (“impressed” is the time period used) are among the many gadgets included within the file.

One final probably attention-grabbing file for evaluation is named Media_ServicesShops ExerciseDifferent ExerciseApple Music Click on Exercise V3.csv. It contains the town and longitude/latitude of the IP tackle the place, I assume, I used to be utilizing Apple Music. For me, the file had 10,000 data.

Verizon

After an extended 80+ day wait, Verizon notified me I may obtain my knowledge. It included 17 csv information for a complete of 1.4 meg of information. A lot of the information lined account administrative info (cell line descriptions, system info, billing historical past, order historical past, and many others.), the historical past of notifications that Verizon despatched to me, and my latest texting historical past (however with out textual content contents). Although Name Historical past and Knowledge Utilization information have been supplied, they have been empty apart from a notation that the information was “Masked for safety”.

Verizon supplied two documentation information. One contained the names and basic descriptions of 34 attainable information that may very well be included in a obtain. The information included rely on the Verizon providers you utilize. The second documentation file contained an outline of three,091 knowledge fields that might seem within the information. Whereas the information subject descriptions are useful, they lack some element. For instance, lots of fields are described as containing codes for numerous functions, nonetheless the codes themselves and their meanings will not be described.

One file that was extraordinarily attention-grabbing is named “…/Verizon/Common Inferences.csv”. It comprises a spectacular quantity of demographic details about me and about different folks in my family. Right here is how Verizon’s documentation describes the file:

“The Common Inferences file offers info basic assumptions and inferences to ship extra relatable and related content material throughout our platforms. This will likely embody info like Attributes, Preferences, or Opinions.”

Primarily based on the character of the demographic options, I assume most of it was acquired by Verizon from exterior knowledge aggregators and never gathered by Verizon instantly from me. The quantity and scope of demographic options far exceed any info that I ever supplied on to Verizon.

In reality, the Verizon documentation speaks about one other file referred to as the “Common” info file (not included in my obtain). The documentation says the “Common” file contains knowledge that got here from exterior info sources. My guess is the data within the “Common Inferences” file additionally comes from these exterior sources. A number of the monetary knowledge within the “Common Inferences” file may have come from the credit score report that Verizon requires its clients to offer.

A complete of 332 demographic options have been included in my Common Inferences knowledge. Right here is an abridged listing together with a number of the extra stunning options:

The entire Common Inferences options are apparently utilized by Verizon to market to me and retain me as a buyer. As you may see within the above listing, options about my partner and our kids are additionally included. You may see the whole listing of 332 options here.

A couple of of the options that I discovered to be really uncommon embody:

One has to marvel if these varieties of knowledge parts are actually wanted by Verizon to assist it present service to me and, if that’s the case, how Verizon makes use of them.

Amazon

Amazon supplied 214 information containing 4.93 meg of information. A number of of the information lined:

• Account preferences;
• Order historical past;
• Success and returns historical past;
• Viewing and listening historical past (Amazon Prime Video and Amazon Music);
• Kindle purchases and studying exercise,
• and search historical past together with search phrases.

If I used to be an Alexa buyer or a Ring buyer, I assume I’d have acquired knowledge for my exercise on these providers as properly.

Six .txt information contained high-level descriptions of some of the downloaded knowledge information. A number of .pdf information include documentation for fields within the downloaded information (the “Digital.PrimeVideo.Viewinghistory.Description.pdf” file, for instance).

Essentially the most attention-grabbing information from Amazon pertain to the advertising and marketing audiences related to me by Amazon, it advertisers, or “third events”. I presume the third events are knowledge distributors from whom Amazon purchases knowledge.

The “…/Amazon/Promoting.1/Promoting.AmazonAudiences.csv” file comprises the audiences that Amazon itself assigned me to. Here’s a pattern of the 21 audiences:

Amazon’s personal viewers assignments are largely correct after I take into account merchandise that I bought or looked for, both for myself or on behalf of others.

The “…/Amazon/Promoting.1/Promoting.AdvertiserAudiences.csv” file apparently comprises a listing of Amazon advertisers who introduced their very own audiences to Amazon and whose viewers lists included me. The file comprises 50 advertisers. Here’s a pattern:

I do enterprise with or personal merchandise from a number of the advertisers within the listing (for instance, Delta, Intuit, Zipcar) so I perceive how I ended up on their viewers lists. I’ve no reference to others on the listing (for instance, AT&T, Pink Bull, Royal Financial institution of Canada) so I’m not certain how I received of their viewers lists.

In accordance with Amazon, the file

“…/Amazon/Promoting.1/Promoting.3PAudiences.csv”

comprises a listing of

“Audiences by which you might be included by third events”.

Its accuracy is poor. A complete of 33 audiences are listed, 28 of which centered on vehicle possession. The remaining 4 lined gender, schooling stage, marital standing and dependents. A pattern of the automobile-related audiences:

Whereas the gender/schooling stage/marital standing -type assignments within the file are correct, just a few of the automobile-related assignments in it are appropriate. Most will not be. And, I’m simply not that interested by cars to warrant 28 of 33 profile assignments. Mercifully, Amazon appears to disregard this knowledge when it presents product or video suggestions to me.

Parting Ideas

On this article, I hoped to indicate you the big variety of information you will get from firms with whom you do enterprise. The info lets you study what these firms take into consideration you whereas additionally studying some stunning issues about your self!

We’ve seen that some firms appropriately establish my pursuits in know-how and travelling, whereas one firm incorrectly sees me as an avid vehicle fanatic. In an eye-opening and considerably unnerving second, I spotted one other firm has in depth demographic details about my household.

I discovered I would like to extend my exercise regime in one of many two locations I name house, although I assumed my exercises have been equal in each locations. I discovered that some firms (fb, Google) wouldn’t have a powerful view of my profile. But the demographic image that Verizon has of me is shockingly correct.

The info the assorted firms provide you with provide a wealthy supply of uncooked materials for experimentation. It’s knowledge that’s vulnerable to deep evaluation, modelling and visualization actions. For instance, geographic coordinates and timestamps can be found for a lot of observations, permitting you to visualise or mannequin your actions.

I hope you discover your personal set of attention-grabbing insights by downloading your private knowledge. Please let me know when you’ve got noteworthy experiences in working with firms aside from these I cowl right here.

It’s your knowledge — Now go for it!