Some smart folks at Warwick have done some very large scale analysis of digitized
historical texts and more specifically to this post here, polarization in political speeches and surveys showing perhaps unsurprising things like war correlates with falls in GDP and falls in happiness, but also that (at least in the US) there's a marked diversification of political opinion in the disappearing center of politics, while the "leaders" concentrate at the poles, the behaviour by the larger population is more akin to anarchy.
what this suggests is that, over a sequence of political tweets, one might expect a bot to take a consistent position on a bag of topics, but a human to show lack of adherence to a party (pole) position. a very light (LDA is probably overkill) approach to the set of tweets should see the divergence of real people from the political machine.....it would be interesting to compare this with more complex (e.g. deep learning) approaches.
One problem for the party machine is that they cannot significantly dilute or randomize the belief tropes they exhibit for fear of losing cult followers, so one should be able to filter (unfollow/mute, whatever) them permanently, fairly effectively - certainly, anecdotally, this is what i see in the twittersphere. e.g. in recent politics in the UK, accounts that are high probability bots (low lifetime, small follower cadre, auto-timing retweets etc) also bundle together beliefs (e.g. with brexit, goes "NHS health tourism" "the UK is too crowded with immigrants" and cliche/propaganda phrases like "take back control" and "soveriegnty" etc - so need to take each of these sources over some number of successive tweets and compare with known non-bots to see if real people are more diverse - i think they will turn out to be so. if it works for speeches, should work for microblogging and social media too...
historical texts and more specifically to this post here, polarization in political speeches and surveys showing perhaps unsurprising things like war correlates with falls in GDP and falls in happiness, but also that (at least in the US) there's a marked diversification of political opinion in the disappearing center of politics, while the "leaders" concentrate at the poles, the behaviour by the larger population is more akin to anarchy.
what this suggests is that, over a sequence of political tweets, one might expect a bot to take a consistent position on a bag of topics, but a human to show lack of adherence to a party (pole) position. a very light (LDA is probably overkill) approach to the set of tweets should see the divergence of real people from the political machine.....it would be interesting to compare this with more complex (e.g. deep learning) approaches.
One problem for the party machine is that they cannot significantly dilute or randomize the belief tropes they exhibit for fear of losing cult followers, so one should be able to filter (unfollow/mute, whatever) them permanently, fairly effectively - certainly, anecdotally, this is what i see in the twittersphere. e.g. in recent politics in the UK, accounts that are high probability bots (low lifetime, small follower cadre, auto-timing retweets etc) also bundle together beliefs (e.g. with brexit, goes "NHS health tourism" "the UK is too crowded with immigrants" and cliche/propaganda phrases like "take back control" and "soveriegnty" etc - so need to take each of these sources over some number of successive tweets and compare with known non-bots to see if real people are more diverse - i think they will turn out to be so. if it works for speeches, should work for microblogging and social media too...
No comments:
Post a Comment