I've been entertaining this idea for a while.

Train a NN/markov chain on enough of my own and publicly available facebook posts and comments that I can make it spew out realistic-looking posts and comments on demand.

Make another script that likes and unlikes pages, joins and leaves groups.

Make another that uploads random photos (or randomly generated image content) and tags me in them.

Stitch them all together into something that runs all of those on my account, generate so much noise that it becomes a tedious exercise to identify real comments and posts simply by virtue of volume.

For bonus points, run it from my home computer.

Would it be perfect? Probably not, there's probably way more you could do to simulate a real user, but it'd make a fun project if nothing else.

A lazy implementation of this idea might be to just have it post content from https://www.reddit.com/r/SubredditSimulator/

You say lazy, I say making use of existing resources :P (or not re-inventing the wheel).

Seriously though, Subreddit Simulator is probably actually a pretty good idea: a lot of them come with a handy image/gif attached and occasionally you get results that could pass as being written by a human.

That sounds human, albeit confusing. Very cool!

I was thinking the same thing ... but I was going to steal random content from the Twitter firehose. You'll want a bit of smarts behind it so that it's not obviously you (e.g. don't post much when you'd normally be sleeping). It's relatively easy to automate a browser to do this these days (giving you a realistic user-agent and behavior). You'll also want to avoid registering key-presses at an unrealistic (or constant) rate. Apparently you can even be identified by your typing. I guess you could train it on your actual typing but that might be overkill.

You will probably always lose.

It is someones daily job to try rand enrich the data they are getting from you, you are working on your bot(s) on your spare time. You just purely can not spend as much resources to keep ahead of the algorithm.

EvilCorp™ will always gain more benefit out of adapting the algorithm to even include/exclude your bot since they get profit gains across the board.

More over you need to spend a lot of time on your bot(s) to make them blend in. I'm not saying; "don't do it", I'm saying; "it is going to be hard to get real benefits"

It's asymmetric; it is significantly easier to make a clear image harder to read than it is making a fuzzy image clearer, and the same applies to data sets.

There might be swaths of teams all focused on enriching the data, but you eventually can't derive more information out of a deliberately flat information graph, and all it takes is a few extreme data points to distort the average.

It certainly would I dont know what a Markov chain is but it sounds cooler than me dot rand'ing my way through random twitter posts.

Yeah, Oauth and events is the only thing keeping me on FB. Although, I believe they opened up their event system to non-fb users, probably so they could build shadow profiles.

Once Facebook algorithms detect that your account is managed by bots, they can simply discard whatever you did since you started the bots, rendering your efforts useless. I am guessing it wouldn't be that hard to detect account anomaly for Facebook - any abnormal activity, such as liking an unliking pages several times a day, sending random people friend requests etc.

Great idea, and hilarious, BUT: you're going to end up spamming the crap out of your REAL friends with all the nonsense you post, so I think step one would be to block or de-friend anyone you care about in your social graph, first.

That’s perfect! Because then they would start removing me as a friend and the whole process wouldn’t be one-sided :-)

Someone went and did this :D

https://gitlab.com/danso/derivejoy

Even that doesn't guarantee your previous data is gone; nothing is preventing them from maintaining a history of changes, meaning they would have both your new, munged, data AND your older data, pre-munge.

Not sure what you are gaining with this.

My “data” would be a cluster fuck, technically speaking.