Skip to content

doodyparizada/word2vec-spam-filter

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

43 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

word2vec-spam-filter

This is a project done during the Kik hackathon 2017.

In this project we demonstrate a way to classify spam messages on the client while protecting user privacy.

A client generates a "hash" from the message sending it to the server. The server then compares the "hash" to a bank of known reported messages.

The bank of known reported messages is created from spam reports. The server compares a given reported message to the previous bank of reported messages. If the message is similar to a previously reported message, a report count is incremented. Otherwise the message is added to the bank with a count of 1.

A message in the bank of reported messages is considered a spam message once it was reported more than 3 times.

Preview

Corpus downloads

We used 2 datasets for creating sentence vectors:

  1. word vectors taken from: https://github.com/stanfordnlp/GloVe
  2. word frequencies from: https://github.com/IlyaSemenov/wikipedia-word-frequency/blob/master/results/enwiki-20150602-words-frequency.txt

Configurable parameters (Hyper-Parameters)

We played around with a few configurations to get the best results for short user messages:

  • Confidence Threshold - a number between 0.0 - 1.0 to determine when 2 messages are considered the same
  • Distance Function - we used vector dot product
  • Normalization - how to deal with words we don't have in our corpus, punctuation marks, non english words
  • Vector Size - the longer the vector the higher the accuracy but heavier in memory
  • Weight Function - given a word frequency, how to create the vector weights (the should weigh less than camera)
  • Custom Corpus - creating the word vectors and frequencies from real user message data might yield better results
  • Random Indices - how many random indices should the client send to the server to mask the original message indices

Running the code

This project includes a single makefile to help with the initialization, dependency installation and corpus download. You can invoke a help message by running:

make

Or you can manually run the server and client apps:

server

In the server directory install the pip dependencies in a virtualenv:

pip install -r requirements.txt

and run the server:

python app.py

web client

To use the web client go into the webclient directory in your terminal and then:

npm install
npm run dev

That should install all dependencies and kick start the project, if it all works you should see something like:

Project is running at http://localhost:3333/ webpack output is served from /

Now load http://localhost:3333/ in your browser

There are 3 different "view modes" which can be switched using the select box at the top right corner of the page.
The 3 views are:

  • Standalone Tester: A textarea in which one can input a message and then either report it as spam or check whether it is classified as spam.
  • IM Sender: A textarea in which the user can input a message (or select a message from a bunch of existing ones) and then "send" the message to another client.
  • IM Receiver: A view which displays a list of received messages (using the IM Sender) and the ability to report each message as spam.