Data Science Project - Newspapers Text Mining

Development repo of end-to-end Data Science project for extracting insights from Newspaper articles via Text Mining toolkit

Results achieved so far

1. LinearSVC Text Classifier with 82% accuracy for predicting 6 labels (from Politics to Sports) available via containerized API
2. XGBoost Classifier with 92% accuracy for predicting unsupervised generated labels of also 6 different subjects

Current solution in action!

Scope

The main goal of this project is to build a production ready Text Classifier wrapped inside an API in order to showcase how I personally approach Data Science problems specifically when it comes to Text Mining.

Therefore I usually follow the steps below, which are explained as follows:

1. Data Extraction: extract newspaper articles using avaible API's (notebooks)
2. Data Processing: clean the data before diving in (notebooks)
3. Exploratory Data Analysis (EDA): draw insights via Sentiment Analysis (notebooks)
4. Modeling: build a supervised Text Classifier and clusterize articles using GloVe Word2Vec (notebooks)
5. Productization: deploy Text Classifier via API end-point (api)

Applied Technologies

All technologies involved in this project are described below more or less in the order of the notebooks.

• Feature Extraction: Tfidf Vectorizer and RegEx
• Supervised Learning: LinearSVC, Logistic Regression, Multinomial Naive Bayes, Random Forest and XGBoost Classifiers
• Clustering: DBSCAN and K-Means
• Transferred Learning: GloVe Word2Vec
• Dimensionality Reduction: t-SNE
• Statistical Testing: Chi Squared for words correlations to labels
• Model Selection: Train Test Split and K-Fold Cross Validation
• API: Data Extraction from NYT, Sentiment Analysis with TextBlob, Productization / Deployment with FastAPI

Main libraries and frameworks: scikit-learn, xgboost, textblob, fastapi, pandas, numpy, seaborn, matplotlib, docker, docker-compose

How to run

Requirements

• python >= 3.6
• conda: Miniconda Python 3.

Step-by-step guide

• Create conda environment based on yml file.

conda env create -f environment.yml

• Activate conda environment.

conda activate ds-env

• Run notebooks as you wish

Folder organization

├── README.md                   <- The top-level README for scientists and engineers.
│
├── api                         <- API folder with containerized Text Classifier
│
├── data                        <- Data folder (versioned in the cloud, not with git)
│   ├── external                <- Data from third party sources.
│   ├── interim                 <- Intermediate data that has been transformed.
│   ├── processed               <- The final, canonical data sets for modeling.
│   └── raw                     <- The original, immutable data dump.
│
├── models                      <- Trained and serialized models (versioned in the cloud, not with git)
│
├── notebooks                   <- Notebooks folder 
│
├── credentials                 <- Required credentials stored locally (ignored on repo for security issues)
│
├── reports                     <- Code-free, stakeholders-ready reports such as markdown files
│   ├── figures                 <- Graphics and figures to be used in reporting
│   └── data                    <- Output data generated by models or analyses
│
├── environment.yml             <- The conda env file for reproducing the environment
│
├── setup.py                    <- makes 'src' installable so it can be imported
│
└── src                         <- Reusable Python code
    ├── __init__.py             <- Makes src a Python module
    └── ...                     <- Further modules as work in progress