cs235project

CS235 project for twitter sentiment analysis COVID-19 related tweets based on BERT deep neural network

How to run this project ?

To run this project, It needs some python packages to be pre-installed, It is recommended that this project should run on python 3.6 or greater version.

For training data and twitter data, please see folder to download appropriate files saved in S3.

If you want to train complete model, It is recommended you have GPU based machine with NVIDIA graphic card installed. CPU based machine will take longer time (~ 12-15 hours).

To install necessary packages, download requirement.txt file and run pip install -r requirement.txt this will install necessary packages.

afterwards, tune your model parameters from config/parameters.py file, by default it will take file's parameters as it is.

upon set up params, run main.py for training model, once model successfully trained, it will be saved as bin/<model>.bin

Document Structure.

1. app

   this folder is application for commandline quick sentiment test as well as web application.to run standalone web application

   python app.py

   after successfully application run, open browser and type http://localhost:5000, you will see text block. type any text for finding sentiment, It loads pre-trained model and draw sentiment out of it [negative, neutral, positive]

2. bin

   folder where trained model saved. some of the trained model in s3. if you want to use those model download from here: `https://cs235project.s3.amazonaws.com/models/dropout_sentiment.bin`

3. config

   project configuration file. model hyperparams, data path and all kinds of constants defined here.It has various files based on different parameters.

4. data

   model training data, note here this data is tagged data from google playstore or yelp. where there are two main field. user reviews (in text format), user sentiment (0 for negative, 1 for neutral, 2 for positive.) This is supervised learning so this data is used for model fine tuning and transfer learning. this should be in .csv file.

5. dataprocesssor

   this is folder has data preprocessing file.convert raw text data into deep learning model input object that compatible to BERT input layer.
6. evaluation

this folder has model evaluation file. after training model, It will generate confusion matrix and get test data accuracy.Note here, this is after training step,means you need a trained model for evalaution on test data.

7. model

   various model architectures abstract class, like simple model, model with dropout layer or model with two deep neural layers etc.

8. twitter

   twitter raw tweets data related to covid19, after model training and fine tuning. tag sentiment on raw covid19 related tweets.

9. main.py

   main file for start training model.

10. builder.py

    supportive file for main.py

Useful data links

• COVID19 related tweets data ( Full ).

  • https://covid-data-for-project.s3.amazonaws.com/coviddata.zip

• COVID19 tweets data samples.

  • https://cs235project.s3.amazonaws.com/twitter_sentiment_data/2020-[mm]-clean.xlsx

  replace mm to month digit. for example January replace 01. project contains data from January 2020- October 2020.

• COVID19 tagged tweet example by model.

  • https://cs235project.s3.amazonaws.com/twitter_sentiment_tag/2020-[mm]-tagged.csv