Conversation

This should be testable, and is probably more nuanced than a blanket statement. Emoji, like all comms, are signaling but wrapped in a nice data format. The 🇺🇸 is quite different from 💃🏾 or 🐞.

Quote Tweet

Claire Lehmann

@clairlemon

Apr 18, 2019

Anyone else noticed a direct correlation between the number of emojis in a Twitter handle and the stupidity of tweets emanating from that handle? Seems to afflicts "both sides" equally.

does pushshift have the ability to easily pull this data? Thinking firehose + profiles + most recent scrapes of matching profiles.

Replying to

I agree. We could grab an x% sample over a day / week and start there. It's definitely doable. My only question is how does one qualify "stupidity" in a tweet. We could start by quantitative analysis first I guess -- look at tweet length, etc. (is there a difference there).

Replying to

The premise of stupidity is not the right way to start (also probably shallow minded!). Instead we should look to see how eg. 🌞 people look different from 💀. Thinking simple word counts or word2vec (if enough data) would be very illuminating.

JSON Baumgartner

@jasonbaumgartne

Replying to

@metasemantic

Agree on all points. The biggest issue is the time it would take to gather enough tweets but I have 750 million tweets from verified accounts. Although that might skew things since they're verified and a subset ... we can grab 45% of a day's worth of tweets in a months time.

4:14 PM · Apr 19, 2019Twitter Web Client

Replying to

I think the interesting story is in the firehose not just the verified. (too much corporate interest in the blue checkmarks). How difficult is it to pull an unbiased sample from a small window? I know I can get a small percentage with the API... how much better can you do?

Replying to

and

Ah sorry! Didn't see the time estimate! Do you have a "day" pulled already? We could filter on English only tweets to make it better but I'm not sure your reconstruction allow for that.