mz256 / guardian_news_classification Goto Github PK

Write thorough description of the project, add figures.

Deploy model as web app, using as input new scraped articles.

Create new dataset with more topics to cover common topics such as sports, health, politics. If good accuracy achieved, deploy in Dash app.

Add PCA and t-SNE

Create new dataset by scraping by category, not just keyword. This way, we can have a labelled dataset which we can use for both topic modelling AND supervised NLP, such as text classification.

The one we have now is not easily labelled.

Add module for unsupervised topic modelling, perhaps using LDA.

Deploy remotely.

Created a new branch "add-analysis" to work on the analysis script.

Save text files as separate .txt files, divided by category, rather than as 5 dataframes.

More convenient for workflow and future integration with other services.

Try various modelling strategies for text classification. Run cross validation and hyperparameter tuning.