To evaluate political sentiment, I took tweets posted in 2019 by 10 media titles (in alphabetical order): BBC, Daily Express, Daily Mail, Daily Mirror, (The) Economist, Financial Times, (The) Guardian,(The) Independent, (Daily) Telegraph and (The) Times.
From the historical left-right point of view the political sympathies of British media look as follows:
Right: Daily Express, Daily Mail, Telegraph and The Times, The Economist, Financial Times
Left: The Guardian, Mirror
Centre: BBC, The Independent
I found it difficult to judge the titles when it comes to liberal/authoritarian split. However, I would expect Daily Express, Daily Mail and Telegraph to be rather on the authoritarian side while The Economist, Financial Times and The Independent would be on the liberal side. The other titles usually present a mixture of the liberal/authoritarian views.
To extract the tweets I have used GetOldTweets3 library. It seems that you can’t pull out too many at a time, therefore, for each username (media title), I have extracted tweets for each month separately and merged them at the end of the process. Below, an example for BBC for September 2019.
The tweets once saved into the data frame looks like this (BBC example).
Finally, I have merged all the media together into single file. I have done a tiny bit of cleansing by removing http links from the text body. Also, I have deduplicated tweets with the same text. I did not do any extra text cleansing, as VADER works better on text with exclamation marks, capital letters, emoticons, etc. You can find the final file here.
VADER produces positive, neutral and negative scores which represent the proportion of the text that fall into each of the categories. It also produces compound score which is a normalised sum of all lexicon ratings of the tweeted words. Compound evaluates the overall tone of the text and ranges from -1 (negative) to 1 (positive), with values between -0.05 and 0.05 classified as neutral.
What is great about VADER is the fact that the scores not only tell you if the text is positive or negative but also how positive or negative the text is. I have used compound score to evaluate the sentiment, as the most comprehensive one of the four.
Now the initial scale of the compound was between -1 and 1. As I have mentioned before, to calculate left-right sentiment I have deducted Corbyn’s score from Johnson’s score. By doing so, I have changed the overall range of the sentiment score to -2 to 2. However, I wanted to compare the left-right sentiment with the authoritarian-libertarian sentiment (which is on scale -1 to 1). To make both scales directly comparable, I have rescaled the left-right sentiment back to values between -1 to 1, using a tanh transformation.
I have also adjusted the final score for EU sentiment by dividing it by -1. This way I reversed positives to negatives and vice versa. I did that so my chart was compatible with other political compass charts available on the internet, which have authoritarian at the top and libertarian at the bottom of the Y axis.
We can now create a final chart…
VADER predicted accurately the direction of the political sentiment. We can argue about the relative scale (for example, is one title more right wing than the other one). Nevertheless, overall, left wing titles are on the left and right are on the right. The only 2 potential oddities (in my opinion) are: The Independent which seems to be more left wing than The Guardian, and in addition, on the authoritarian side. The other oddity is The Economist, which should be perhaps closer to the centre.
- The names shown on the chart are the names of the Twitter accounts.
The 4 most important findings are:
- The British media are biased towards authoritarian views- the majority sit in the top part of the chart.
- They also are more right wing rather than left wing
- The tabloids (Daily Express, Daily Mail and Daily Mirror) tend to present less balanced views — they sit on the outskirts of the chart
- Financial Times is the most balanced title — almost a perfect centre!
Using quite simple criteria (‘Johnson’, ‘Corbyn’ and ‘EU’) I could create a map of political sympathies, which is probably not far off from reality. With some additional work on right-left and authoritarian-libertarian split, the VADER method can be a powerful tool in evaluating political sentiment.
 Clayton J. Hutto, Eric Gilbert, VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text (July 2014), International AAAI Conference on Weblogs and Social Media, Ann Arbor, USA
Parul Pandey, Simplifying Sentiment Analysis using VADER in Python (on Social Media Text), (September 2018) Analytics Vidhya
 Martin Beck, How to Scrape Tweets From Twitter (January 2019), Towards Data Science
Twitter accounts used in the analysis:
@TheEconomist, @TheTimes, @guardian, @FT, @BBCpolitics, @DailyMailUK, @Daily_Express, @Independent, @DailyMirror, @Telegraph