Slide 1

Slide 1 text

Multichannel Marketing Platform Phishing Detection with 
 Recurrent Neural Networks João Pereira

Slide 2

Slide 2 text

The world E-goi +15 Years
 experience 500.000
 users 700.000.000
 SMS sent in 2018 100.000.000.000
 Emails sent in 2018

Slide 3

Slide 3 text

What was accomplished? F1-Score Recall Precision 98.0% 98.2% 98.4%

Slide 4

Slide 4 text

About Phishing

Slide 5

Slide 5 text

What do you look for? Sender Email doesn’t make sense Sense of urgency
 Call for immediate action Generic greeting Often poorly written ....

Slide 6

Slide 6 text

The Data issue “Data! Data! Data! I can't make bricks without clay.” - Sir Arthur Conan Doyle

Slide 7

Slide 7 text

The tools for the job Python Google Colab Keras

Slide 8

Slide 8 text

[ 4, 23, 12, 52, 65, 8, 24, 23, …, 14 ] Tokenize Making the data readable Generate Embeddings

Slide 9

Slide 9 text

How about neural networks? Feed Forward Network Recurrent Network

Slide 10

Slide 10 text

Understanding the Architecture Great explanation at: https://colah.github.io/posts/2015-08-Understanding-LSTMs/

Slide 11

Slide 11 text

Understanding the Architecture Great explanation at: https://colah.github.io/posts/2015-08-Understanding-LSTMs/

Slide 12

Slide 12 text

Forget Gate

Slide 13

Slide 13 text

Forget Gate

Slide 14

Slide 14 text

Input Gate

Slide 15

Slide 15 text

Input Gate

Slide 16

Slide 16 text

Cell State

Slide 17

Slide 17 text

Cell State

Slide 18

Slide 18 text

Output Gate

Slide 19

Slide 19 text

Output Gate

Slide 20

Slide 20 text

Thank you!