How jung-akim/netflix_app works
Project type and purpose
• Type: Data-science topic-modeling & classification pipeline with a lightweight web front-end.
File structure
• Data directories
• raw_data/ – original CSV/TSV and pickled DataFrames
• preprocessed_data/ – lemmatized corpus, TF-IDF, dictionaries, term-doc matrices
• Model artifacts
• coherence_scores/ – pickled coherence results from grid search
• lda_mallet_model/ – pickled Mallet LDA models
• prediction/ – pickled prediction outputs (LDA & BERT)
• Visualizations
• mallet_lda_vis/ and std_lda_vis/ – pyLDAvis HTML exports
• Code & notebooks
• .ipynb (grid_search, LDA/NMF/LSA, topic_labelling, classification, BERT)
• utils.py – NLP pipeline and helper functions
• predictor_api.py & predictor_app.py – Flask back-end
• templates/predictor.html – simple HTML UI
• Others
• __pycache__ (build artifacts), .ipynb_checkpoints
README for additional insights
• Pipeline: Data ingestion → preprocessing → topic modeling → human labeling → BERT training & comparison → 2D visualizations → Flask deployment.
• Tech stack: Python, JupyterLab, gensim, spaCy, NLTK, pyLDAvis, ktrain/transformers, sklearn, umap, wordcloud, Flask.