covid-news-analyzer

Analyze news articles related to covid-19 based on different metrics:

1. Sentiment Analysis
2. Emotion Detection
3. News Article Categorization
4. Fakeness

You can access the tool online at: https://covid-news-analyzer.herokuapp.com/

(this doesn't currently support fakeness detection due to memory constraints on Heroku's free tier)

---

How To Run Locally

Install requirements from requirements.txt using pip install -r requirements.txt and run:

Training

python -u main.py --dataset <dataset_name> --models <models> --feats <transformations> --save_path <model_path> --save_results

• <dataset_name> can be one of emo_aff, stan_sent, news_cate, fake_news representing the standard datasets used to train the models
• is a space separated list of models to train - mnb, svm, xgb, ada, rf, lr
• is a space separated list of feature transformations to use - bow, tfidf, ngram

This would train the specified list of models on all the specified feature transformations for the given dataset, and save the results in a csv file, as well as the performing model on the test portion of the dataset.

Evaluation

Run python evaluate.py to obtain the predictions on the COVID-19 article test set, and use python analysis/analyze.py to obtain the evaluation scores.

Web Portal

Run python wsgi.py to launch the web portal. This lets you analyze a specific article against trained models placed in output/model_dump directory.

Report

You can find more details about the project here.