Hacker News Reader

Comment by fiatpandas | Hacker News Reader

Don't really understand the implied connection to NSA or GCHQ. My theory as to why Snowden is popping up in some of the tweets is because the bots are probably using something like Markov Chains to generate real sounding (as far as Twitter spam detection is concerned) tweets sourced from trending content.

The accounts would then be sold as followers, which is an attractive service to some for a variety of reasons

View on HN

If they have so many accounts, and they can automatically set them up to say something, couldn't it be part of the public opinion manipulation program?

https://firstlook.org/theintercept/2014/02/24/jtrig-manipula...

Think if you were the agency or the leader of an intelligence group. Would you do this? Would others? The answer is yes obviously, even if bots retweet or broadcast even the better.

Tilting public opinion goes back to the beginning of the last century and the beginning of 'public relations' i.e. Edward Bernays http://en.wikipedia.org/wiki/Propaganda_(book).

Companies, brands and others do this as well influencing thought, it is a huge problem on Facebook as well with likes, ads, accounts.

It is a faction of marketing (public relations) and has elements of tabloid.

The Markov stuff seems like such a poor approach. Though you can get by with it in 140 characters better than other places.

Twitter provides a public stream of tweets. Simply pull tweets out of that, if it doesn't contain another users name or url, etc. Distribute out to one of your bots, weighted by chattiness. Get fancy and have some sort of baysian filtering for what category bucket it falls into and give your bots interests.

Poor approach or not, it's a fact that Markov chains are used by web scrapers, comment bots, and email spammers.

Wouldn't that be trivial to detect for their spam filters?

For all the janky looking URLs I get from eggs that only have like twenty tweets, half of them janky looking URLs, I don't believe anything is trivial for Twitter's spam filters.

Maybe? They never did anything with the proof of concept ones I worked up, but I never did anything with them other than basic interactions with the network.

I'd guess there are enough duplicate tweets. "Another crappy day at work", "looking forward to dinner tonight", "going to see my sister this weekend!", etc. That I doubt it'd be a very useful heuristic.

I actually did that and let it run for a week. It got flagged for spam pretty quickly. I suspect because it was a pretty similar tactic to avoid filters.

Reply on news.ycombinator.com