Release Notes¶

v0.1.4¶

Bug Fixes¶

• Fixed missing MultipartiteRank Python export -- the class was registered in the Rust PyO3 module but not re-exported from python/rapid_textrank/__init__.py, causing ImportError: cannot import name 'MultipartiteRank' when installing from PyPI.

v0.1.3¶

New Algorithm Variants¶

SingleRank¶

• New SingleRank variant with weighted edges based on co-occurrence frequency and cross-sentence windowing -- words in adjacent sentences can co-occur within the same sliding window, unlike classic TextRank which treats each sentence independently.
• Full Python bindings: native class, JSON dispatch, and spaCy pipeline component.
• Unit tests, integration tests, and documentation.

TopicalPageRank¶

• New TopicalPageRank variant that biases PageRank toward user-supplied topic weights, enabling domain-focused keyword extraction via a personalization vector.
• Includes PersonalizedPageRank engine in src/pagerank/personalized.rs with automatic normalization of the personalization vector.
• 7 unit tests and an integration pipeline test.
• Python topic_weights_from_lda helper (python/rapid_textrank/topic_utils.py) for deriving topic weights from a Gensim LDA model, with an optional Jaccard pre-filter heuristic.
• Full Python bindings and README examples.

MultipartiteRank¶

• New MultipartiteRank variant that performs Hierarchical Agglomerative Clustering (HAC) on candidate keyphrases using Jaccard distance, then applies a multipartite graph ranking.
• Shared src/clustering.rs module extracted from the former TopicRank-specific clustering code, now reused by both TopicRank and MultipartiteRank.
• Disjoint-set fast-path optimization for Jaccard distance computation.
• Full Python bindings, README documentation, and notebook examples.

Core Engine Improvements¶

• Cross-sentence windowing added to GraphBuilder -- configurable window that spans sentence boundaries, improving recall for long documents.
• Shared clustering module (src/clustering.rs) -- HAC with Jaccard distance, reusable across variants.
• TopicRank edge weighting aligned with the PKE reference implementation for correctness.

Bug Fixes¶

• Fixed use_pos_in_nodes serde default -- changed from false to true, fixing 2 POS-filtering tests (test_json_include_pos_filtering, test_json_include_pos_multiple_tags).
• Fixed cargo fmt violations that were failing CI checks across multiple files.

Documentation & Examples¶

• README updated with usage examples for SingleRank, TopicalPageRank, and MultipartiteRank.
• Notebooks 02_algorithm_variants and 04_benchmarks updated with examples for all new variants.
• CLAUDE.md and AGENTS.md updated with PyO3 + Python 3.14 build caveats.
• topic_weights_from_lda usage example added to README.

Benchmarks¶

• Added SingleRank, TopicalPageRank, and MultipartiteRank to the Criterion benchmark suite.

Stats¶

• 20 commits since v0.1.2
• 27 files changed, 3,409 insertions, 271 deletions

v0.1.2¶

Highlights¶

• Added TopicRank support via JSON interface (variant="topic_rank") with spaCy-token examples.
• TopicRank behavior aligned more closely with pytextrank.
• Docs and notebooks updated to include TopicRank usage and comparisons.

Details¶

• New JSON config fields: topic_similarity_threshold, topic_edge_weight, focus_terms, bias_weight.
• README updated with TopicRank section + citation.
• Benchmarks notebook now compares rapid_textrank TopicRank vs pytextrank TopicRank using spaCy tokens.

v0.1.1¶

Highlights¶

• Closer alignment with pytextrank defaults: window size now 3, POS defaults include verbs, and scrubbed-text grouping available by default.
• Improved phrase quality: stopword-aware chunking reduces noisy phrases.
• Config parity across APIs: new options surfaced in Python, JSON, and spaCy component interfaces.

New¶

• use_pos_in_nodes to treat nodes as lemma|POS.
• phrase_grouping with lemma or scrubbed_text.
• get_stopwords(language) helper to inspect built-in stopwords.
• JSON config now supports language, use_pos_in_nodes, phrase_grouping, and additional stopwords (extends built-ins).
• spaCy component supports include_pos, use_pos_in_nodes, phrase_grouping, language, and stopwords.

Changed Defaults¶

• window_size: 4 to 3
• include_pos: NOUN + ADJ + PROPN to NOUN + ADJ + PROPN + VERB
• use_pos_in_nodes: false to true
• phrase_grouping: lemma to scrubbed_text

Notebook Updates¶

• Benchmarks now compare rapid_textrank vs pytextrank with and without spaCy tokens.
• Config blocks simplified to use new defaults.
• Stopword list printing added to algorithm explanation notebook.

Bug Fixes¶

• Stopwords now act as chunk boundaries, preventing phrases like "of NLP is to".
• Heuristic POS tagging recognizes common function words to reduce false content tokens.

Compatibility Notes¶

If you relied on old defaults, set them explicitly to preserve behavior: - window_size=4 - include_pos=["NOUN","ADJ","PROPN"] - use_pos_in_nodes=False - phrase_grouping="lemma"

v0.1.0¶

Release Date: February 5, 2026

This is the initial public release of rapid_textrank, a high-performance TextRank implementation in Rust with Python bindings.

Highlights¶

• 10-100x faster than pure Python TextRank implementations (depending on document size)
• Three algorithm variants: BaseTextRank, PositionRank, and BiasedTextRank
• 18 languages supported for stopword filtering
• Dual API: Native Python classes + JSON interface for batch processing
• spaCy integration: Drop-in pipeline component

Features¶

Algorithm Variants¶

Performance Optimizations¶

• CSR (Compressed Sparse Row) graph format for cache-friendly PageRank iteration
• String interning via StringPool reducing memory 10-100x for typical documents
• Parallel graph construction with Rayon
• Link-time optimization (LTO) with single codegen unit
• FxHash for fast internal hash maps

Supported Platforms¶

• Python 3.9, 3.10, 3.11, 3.12
• Linux (manylinux)
• macOS (x86_64, arm64)
• Windows (x86_64)

Supported Languages¶

en, de, fr, es, it, pt, nl, ru, sv, no, da, fi, hu, tr, pl, ar, zh, ja