fastText

FastText is an open-source library for efficient learning of word representations and text classifiers. Developed by Facebook AI Research (FAIR), it extends the word2vec approach by incorporating subword information through character n-grams, enabling the model to generate meaningful vectors for rare and unseen words and to better represent morphologically rich languages.

In unsupervised mode, FastText trains CBOW or skip-gram models using subword information. Each word’s vector is formed from the vectors of its character n-grams in addition to the word itself, allowing the model to infer representations for out-of-vocabulary words. It supports negative sampling and hierarchical softmax for efficiency and offers options such as quantization to reduce memory usage on large corpora.

In supervised mode, FastText provides a text classifier that learns a linear model over a bag-of-words representation augmented with word-gram features. It is designed for speed and scalability on large datasets and supports multi-label classification, making it suitable for tasks such as sentiment analysis and topic categorization.

The library is implemented in C++ with bindings for Python, and includes a command-line interface for training, evaluating, and producing predictions. It is released under an open-source license (MIT), with available pre-trained vectors for multiple languages and active community support.