chronowords 0.2.0

Detect semantic shifts in word embeddings over time

chronowords

Detect semantic shifts over time in word embeddings. Train small PPMI-based language models, create topic models using NMF, and analyze semantic changes using Procrustes alignment.

Features

• Memory-efficient word embedding training using Count-Min Sketch
• Topic modeling with Non-negative Matrix Factorization
• Temporal alignment of word embeddings using Procrustes analysis
• Cython-optimized PPMI matrix computation

Installation

pip install chronowords

Quick Start

from chronowords.algebra import SVDAlgebra
from chronowords.topics import TopicModel

# Train word embeddings
model = SVDAlgebra(n_components=300)
model.train(your_corpus_iterator)

# Find similar words
similar = model.most_similar('computer')
for word in similar:
    print(f"{word.word}: {word.similarity:.3f}")

# Create topic model
topic_model = TopicModel(n_topics=10)
topic_model.fit(ppmi_matrix, vocabulary)

Documentation

Full documentation available at ReadTheDocs.

Requirements

Python ≥ 3.10 NumPy SciPy scikit-learn Cython

Contributing

Pull requests welcome. For major changes, open an issue first.

License

MIT