Nokore - A Collection of Scripts for Detecting Spammy, Fake or Otherwise Dangerous Communications Online
This repository applies SIMOn -- a Character-level CNN + bidirectional LSTM modeling library for text classification -- to email spam classification, and other forms of social media "dangerous/fake/spammy" communication detection.
The repository achieves this via a collection of scripts, with the eventual goal of comparing the SIMOn-based-model to emerging giants BERT and/or ELMo.
The name "Nokore" is a Twi word, by the Akan people of Ghana, for "Truth".
Architecture is described at https://arxiv.org/abs/1901.08456
Review https://github.com/algorine/simon
Also See Texas AI Summit Talk video: https://youtu.be/SmIsWF1xBeI
To get started, make sure you are using python v3.5+ and pip install via
pip3 install git+https://github.com/algorine/simon
Then, install keras-bert using pip install -q keras-bert.
Then, study the scripts and pretrained models included in the Nokore/scripts directory.
Rendered Jupyter notebooks are also provided in the Nokore/scripts directory, and they are meant to be self-explanatory.