- Sentiment give context to content. This is a useful tool for a analysis everything from response to a product to predicting political election outcomes.
- Data is readily available(reviews, news, ratings)
- Practical and useful information for various applications
- find data that can be uniformly labeled, while varying in content type and size and build a model that can take large and small test strings and produce sentiment percentage of the string being either negative, neutral, or positive.
- clone the repo to your colab notebook or local machine demo notebook sample
- since we have large files in repo, install git-lfs before cloning
import os
import getpass
!curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
!sudo apt-get install git-lfs
!git lfs install
- install all requirement package
!make install - go to the scripts directory
cd scripts - train the model and save the model under models
!python main.py - if you work on local machine, jump to the final step; if you work on colab, you need to sign up a pyngrok account to get the authtoken and replace the token below by yours
- run the demo to check the result(load pretrained1 under models, which is a pretrained XLNET with 80% accuracy)
!wget https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip
!unzip ngrok-stable-linux-amd64.zip
!get_ipython().system_raw('./ngrok http 8501 &')
!ngrok authtoken 26WJyNXXUSY34VVdvJeXnkGDO3g_xX5cnoALV1vAwq6K12F8
- get the demo link(in a new cell)
!curl -s http://localhost:4040/api/tunnels | python3 -c \
'import sys, json; print("Execute the next cell and the go to the following URL: " +json.load(sys.stdin)["tunnels"][0]["public_url"])'
- run the demo(in a new cell)
!streamlit run app.py
- use 70 thousands reviews to train XLNET for 2 epochs
