What are the most influential tweets for different user groups?

[swood-ecology]

What tweets are most liked/retweeted by farmers? by non-profit staff? by academics?

Are there differences in the content/message of these tweets that make them more appealing to these different groups?

again, do this by theme--soil health vs rangeland health vs forest health etc

[kylemonper]

I started working on creating a word list that was informed in part by the descriptions of users from the TNC partner list and also by looking at lots of user profiles to find key words that could easily distinguish a user without having too much overlap between groups. This can definitely be refined but I think it's a solid starting point

I then used that word list as the basis for a function that calculates the proportion of each group that RTed each tweet. I started trying to visualize this information by creating word clouds of the tweets that were most liked by each group and also making Boxplots showing the spread of proportions for each group done by category (soil, rangeland, all) -- still trying to think of other interesting ways to visualize.

• I'm going to wait to add this to the master branch because there are a couple of minor bugs that I need to work out within the function. Namely there's seems to be an issue with running one of the Stringr functions on some of the tweets/user descriptions from India.

[kylemonper]

fixed bug! the issue was str_detect was reading things like 'B)' as an unclosed parenthesis. solved used fixed()

now that this is fixed, the next step will be to repeat my previous analysis to now include all tweets.
-putting this on hold while I address the tweet_content edits

[swood-ecology]

@kylemonper should we chat on zoom so you can walk me through how this works?

[kylemonper]

@swood-ecology - yeah of course! I can do 12 or 2 PT this Friday (5/17) or anytime after 10 am PT next Tuesday

[kylemonper]

I took a little time to run the grouping function on the both the top 50 Retweeted and a random set of 50 RTs, in both cases its pretty clear that the function needs refining, though I'm still not sure how best to proceed.
using the descriptions as the only source of data simply isn't going to work, so my next thought is to use each users 'x' number of recent tweets as a supplement -- this will be a good thing to talk through next Friday as well.