narendrakalam2001/cinematch-pro-artifacts

CineMatch Pro — trained artifacts

Trained models, indexes and the processed catalogue for CineMatch Pro — a movie recommendation + review-intelligence platform (visual search, hybrid recommender, aspect-based sentiment, CineBot).

The code loads these files automatically. In the project's .env:

HF_ARTIFACTS_REPO=narendrakalam2001/cinematch-pro-artifacts

or download them yourself from the project root:

python -c "from src.artifact_store import ensure_artifacts; ensure_artifacts(repo_id='narendrakalam2001/cinematch-pro-artifacts')"

Contents

Path	What it is
models/svd_model.pkl	Collaborative filter — scikit-surprise SVD, 100 factors, trained on 5,000,000 MovieLens-25M ratings (RMSE 0.8005)
models/tfidf_matrix.pkl, models/tfidf_vectorizer.pkl	TF-IDF content filter over 62,423 movies
models/faiss_poster.index, models/faiss_poster_metadata.pkl	FAISS IndexFlatIP over 56,029 CLIP ViT-B/32 poster embeddings (512-dim)
models/roberta_sentiment/	cardiffnlp/twitter-roberta-base-sentiment fine-tuned on 50K IMDB reviews (accuracy 94.26 %, F1 94.29 %)
chromadb_store/	ChromaDB index of 52,703 movie plots (all-MiniLM-L6-v2 embeddings)
data/processed/merged_movies.parquet	Processed movie catalogue (MovieLens ids/titles/genres merged with TMDB metadata)
models/model_card.json	Metrics and library versions of the training run

These files were built with numpy 2.1.3, scikit-learn 1.6.1, scikit-surprise 1.1.5, transformers 5.16.1 and chromadb 1.5.9 — use the project's requirements.txt so the pickles and the Chroma store load.

Data sources and licence

• MovieLens 25M (GroupLens Research, University of Minnesota) — used under GroupLens' research licence: non-commercial use, cite F. Maxwell Harper and Joseph A. Konstan. 2015. The MovieLens Datasets: History and Context. ACM TiiS 5(4), no implied endorsement by the University of Minnesota or GroupLens. The files above are derived from it and are shared for research / portfolio use under the same conditions.
• TMDB — movie metadata and credits. This product uses the TMDB API but is not endorsed or certified by TMDB.
• IMDB 50K movie reviews and movie-poster dataset (Kaggle) — used to fine-tune the sentiment model and build the poster index.

This is a non-commercial portfolio project. Model limitations (popularity bias, binary sentiment, qualitative visual search) are described in the project README.