1 of 12

PULSE

2 of 12

Hello!

We’re Adam, Julian and Dom and together we’re Pulse

2

3 of 12

What are we trying to do?

Also, what’s the logo about?

1

4 of 12

How did we do it?

Pull twitter data via its API

Machine learning/deep learning/other buzzwords

Data visualisation with GIS

1

5 of 12

Getting the data

Twitter API access, so we can pull data

  • Filtered by hashtag
  • Filtered by area

Google Maps API to convert Twitter location information to longitude & latitude

5

6 of 12

Different deep learning methods we tried

Grid search on:

  • Support Vector Machine
  • Naive Bayes Classifier
  • Random Forest Classifier
  • Neural Network

NN worked best, SVM was almost as good

6

7 of 12

Visualising the data

FME

  • Workspace designed to process data
  • Point geometry added to each tweet
  • Tweets referenced to UK districts
  • Districts height by % total tweets in district
  • District Colour by Positivity on a colour gradient lowest to highest Positivity %

7

8 of 12

#Hackathon

8

9 of 12

#Homeless

9

10 of 12

What’s next for Pulse?

  • Changes over area vs changes over time
  • Comparing two hashtags in one visual
  • Automate the entire process from start to finish

10

11 of 12

#NationalSandwichDay

11

12 of 12

Thanks!

12