Development build. This page describes main, not a released package. The latest published Lodestar.Text is 0.6.0 — read its documentation.

Home › Text

Two ways to pull the words that matter out of a document, and neither reads the document twice to do it — the score comes out of the same graph or the same run table that finds the candidates in the first place.

Rake and TextRank disagree on what a candidate even is. Rake treats a document as runs of words separated by stop words and punctuation, and scores a run by how often its words occur and how much they co-occur with other words — no order, no neighbours outside the run itself. TextRank treats it as a graph: every word (after stop words are dropped) is a node, an edge joins words that stood near each other, and the score is that graph's dominant eigenvector — the same family of algorithm PageRank uses on links.

flowchart TD
    A["Extracting keywords from one document"] --> B{"Score candidates<br/>by co-occurrence within<br/>a stop-word-delimited run?"}
    B -->|yes, cheap, no graph| C["Rake"]
    B -->|no, rank a graph instead| D["TextRank"]

Both return IReadOnlyList<KeywordMatch>, sorted by descending score — the scale is each extractor's own and is not comparable between them. Neither downloads a stop-word list or a model: RakeOptions.StopWords and TextRankOptions.StopWords take what you supply and default to StopWords.English, already in the assembly (decision 0005).

• The keyword extraction guide — RAKE and TextRank side by side, and the KeyBERT-style composition with Lodestar.Onnx and Lodestar.Embeddings.Search.Mmr.
• Python → C# equivalence — what this replaces on the rake-nltk and summa side, and the divergences each has from its reference.