magpie

Why we included this project

Magpie was built at CERN to tag High Energy Physics abstracts with subject categories and extract keywords from them, so it has been exercised on real scientific text. It's a self-contained training pipeline: instead of wiring together separate libraries, you point it at a corpus of UTF-8 text files paired with label files, and it builds word2vec embeddings, trains a neural network, and returns ranked label scores for arbitrary text. A batch mode handles corpora that don't fit in memory, and trained models can be saved and reloaded for reuse. Teams doing document tagging or topic classification will find the API approachable, though it's worth noting the project hasn't been updated since early 2023 and leans on older Keras-era tooling.