Get Strategic insights into your business!
We develop and deploy solutions focused on your business needs and objectives
As a company, Aelm is somewhat pleased with the result of yesterday’s UK General Election.
Just to clarify, our buoyancy is nothing to do with any gains, losses or new colours on the political map. Our contentedness lies in the fact that, for the third time in as many elections, we predicted an election result with a high level of accuracy – long before the polling stations opened.
Where the nation’s premier psephologists deployed labyrinthine forecasting tables and techniques, Aelm ran a series of sentiment analyses on Twitter to trend and gauge the nation’s political pulse ahead of Vote 2017.
As we saw in our correct predictions in the Brexit referendum and the Trump/ Clinton election, a correctly compiled, cleaned and weighted analysis of Twitter absolutely reflects public sentiment. Our approach can be used for accurate forecasting whenever public opinion is in the spotlight.
As of Friday 9th June, 2017, Aelm’s projections in three consecutive elections have come to pass with barely a margin of error.
0. From May 31st to June 8th, a total of 1,726,276 baseline Tweets were collected between 0100 and 1900 each day. That number was balanced and refined to subtract erroneous and general messages. A final total of 833,915 Tweets was used in the analysis.
1. Tweets were collected through Twitter’s streaming API, using a Python script, and converting data into CSV files.
2. At the end of each day, a different Python script was run to assign weight and value to those Tweets in a three stage process:
a. Extracting hashtags, @ tags and nouns.
b. Evaluating positive or negative sentiment
c. Batching Tweets into rightful political buckets
Resulting data was sorted and compiled in a new CSV file.
3. Next, BI tools Tableau and QlikView were used to read the converted data and offer the following analysis:
a. Party Leaders’ Reach (how many users ‘see’ conversations and discussions about party leaders).
b. Impact (party leaders’ follow numbers and the likelihood of messages appearing in Twitter users’ feeds).
c. Hashtags (the hashtag count for each party leader combined with a metric relating to contemporaneous discussion topics).
d. Sentiment (the number of Tweets categorised as positive or negative and the weight of sentiment in discussion topics).
e. Win Chance (a weighted average calculation was compiled via three parameters: Tweet count, follower count and polarity score – positive or negative).
And our projection came to pass.
It should be said that our conclusions were similar to those calculated by chief BBC psephologist, Professor John Curtice. The difference is that Aelm’s forecast was made much earlier – and it was compiled with only a deep excavation of Twitter sentiment.
If anyone would like further information on our methodology, or to discuss the deployment of sentiment analysis in other areas of business, then we welcome any and all enquiries.