ina-foss/twembeddings

Implements multiple sentence embedding models (TF-IDF, Word2Vec, ELMo, BERT, Universal Sentence Encoder, Sentence-BERT) for unsupervised First Story Detection clustering on Twitter streams. The toolkit performs temporal event detection by computing similarity thresholds between tweet embeddings and evaluates performance across English and French corpora using standard NLP frameworks (scikit-learn, transformers, TensorFlow). Supports custom datasets via TSV input with optional ground-truth labels and outputs clustering predictions with detailed evaluation metrics.

No commits in the last 6 months. Available on PyPI.