NetHui 2020: Social Media Election Trends (David Hood)

• So, ways you can use data to think about how social media and society work with special reference to New Zealand political Twitter. You can use sentiment analysis. You can look at replies to politicians and the sentiment expressed by the words in them, and you can find that there isn’t really any difference on average between male and female politicians in the responses they’re getting in New Zealand. This is actually different to other countries with the misnomer of “United” in their names, so the lack of difference is worthwhile mentioning in and of itself in this day and age. Also, this has shifted in a more optimistic direction since I first made the graph in 2017 with the last election. If you look at the sentiment expressed to individual parties, there is hardly any difference at all.

00:59 - So, when you read stuff about sentiment analysis and parties, be pretty skeptical of the differences. In particular, if we throw in the unknown words in relation to the identified words, those differences become even less. Now, you can do tricks like using synonyms and chopping the ends off words, and that gets that unknown column right down again, but still there’s no real difference there. Also that sentiment analysis is based on looking up predefined sentiments of predefined words, which makes this the tweet expressing the most trust in politicians in the past six months. Because green used metaphorically is associated with trust, and school used metaphorically associates trust.

01:49 - So someone chanting “green school, green school, green school” is expressing a whole lot of trust, based on simple sentiment analysis. So, we can do better by throwing in a whole bunch of other different techniques, when we’re thinking about data, we can use good old ratios, the number of replies on the horizontal, the number of likes on the vertical. New Zealand Twitter is essentially leftwing-ish, but then New Zealand is essentially leftwing-ish, compared to, for example Australia, you can see that in the New Zealand Election Study when you compare it to the Australian Election Study. What New Zealand is, is not biased in general based on gender. Once again we’re checking gender, but with ratios of tweets tweeted by female politicians, male, and parties.

02:42 - There’s not a lot of difference in there when you look at all the tweets. Even when you focus on the most ratioed tweets there’s not a whole lot of difference there either. It’s possible that maybe the female ones might possibly be getting a few more replies, but it’s a hard read. Now, this lack of polarization in New Zealand is also reflected by how many political parties, people who reply a lot to political tweets, have liked tweets of. So, most people who are very active and replying to MPs, have liked at least two political parties in their past 400 likes, some three, or four.

03:35 - New Zealand is much less polarized than most people expect. You can also look at the percentage of replies that each MP has got. What percentage is from people who have liked something from their party, and this once again no particular gender difference in there. So, we’re going to have to get to the trolls and the really nasty people by some other means. Like if we just have a look at actually counting up how many replies people make, if you’re capturing the data live like I am, of accounts that are gone from Twitter, a lot of them have sent more tweets than they claim to have sent, because they are deleting a lot of tweets.

04:27 - So that’s a point that can be used for troll identification, or at least suspicious accounts. Similarly, if you match the number of words in colon, common from the left to the right between the original tweet and the reply, non sequitur replies like the green schools one, show up as not a natural conversation. So that’s just a few building blocks one can use for a data lead analysis of what’s going on. .