yake

YAKE, which stands for Yet Another Keyword Extractor, is an unsupervised keyword extraction algorithm designed to identify representative keywords and key phrases from single documents. It is intended to operate without requiring a large training corpus or supervised learning, making it suitable for quickly indexing or tagging individual texts.

The method works by first generating candidate keywords from the document, typically by extracting n-grams up to a defined maximum length and applying basic filtering to remove noise such as stop words and punctuation. For each candidate, YAKE computes a set of features that capture statistical and positional characteristics of the term within the document. Common features include the term’s frequency in the document, its dispersion or distribution across the text, the position of the first occurrence, the length of the term, and occasionally capitalization cues. These features are then combined through a heuristic scoring function to assign a single score to each candidate keyword.

YAKE is designed to be lightweight and fast, suitable for long or numerous documents without external resources. It is language-agnostic to a degree, relying on document-local features and language-specific stop word lists or similar filtering. Because it does not depend on pre-trained models or large corpora, YAKE can be used for on-the-fly indexing, content tagging, or preliminary summarization in multilingual settings, provided appropriate language resources are available.

Implementations of YAKE are available in open-source libraries and can be integrated into text processing pipelines. It is often compared with other unsupervised methods such as TextRank and RAKE, offering a competitive alternative for single-document keyword extraction.