Algorithm Variants¶
rapid_textrank ships with seven core algorithm variants plus AutoRank, a smart ensemble that fuses the full eligible keyword pool for a document. The core variants all share the same graph-and-PageRank foundation, but they differ in how they construct the graph, weight edges, and post-process results.
Choose the variant that best matches your document type and extraction goal. If you are unsure where to start, use AutoRank or see Choosing a Variant.
Variant Comparison¶
| Variant | Best For | Description |
|---|---|---|
| BaseTextRank | General text | Standard TextRank implementation |
| PositionRank | Academic papers, news | Favors words appearing early in the document |
| BiasedTextRank | Topic-focused extraction | Biases results toward specified focus terms |
| TopicRank | Multi-topic documents | Clusters similar phrases into topics and ranks the topics |
| SingleRank | Longer documents | Uses weighted co-occurrence edges and cross-sentence windowing |
| TopicalPageRank | Topic-model-guided extraction | Biases SingleRank towards topically important words via personalized PageRank |
| MultipartiteRank | Multi-topic documents | Builds a k-partite graph removing intra-topic edges; boosts first-occurring variants |
Learn More¶
- How TextRank Works -- the three-step pipeline shared by all variants.
- Choosing a Variant -- a decision flowchart and scenario table to help you pick the right one.
AutoRank-- the recommended keyword default when you do not want to choose a single variant manually.
Interactive Notebook
Compare all variants side-by-side in the Algorithm Variants Notebook.