← Tool directory
TOOL PROFILE

KeyBERT

Extract candidate phrases from article text to review topic coverage across a content inventory.

Open source

Community support

Upvotes come from SearchEngineAnswer members. GitHub stars and reader reviews are counted separately.

Claim this tool if you maintain it.

Best for
Publishing
Setup required
Use Python and pip in a virtual environment. The default embedding backend needs model files, which may download on first use.
Costs to check
The MIT-licensed Python library can run locally without an API account. Budget for model downloads, storage and compute.
License in catalog
MIT

GitHub repository

GitHub stars
4,226
Forks
389
Last push
2026-08-25
Primary language
Python

GitHub snapshot from . Stars count GitHub bookmarks. Reader ratings appear separately below.

How to install KeyBERT

  1. Create and activate a Python virtual environment, then run python -m pip install keybert.
    python -m pip install keybert
  2. Save the sample script as topics.py and put cleaned article text in article.txt. On first run allow the embedding model to download.

First tutorial for KeyBERT

  1. Put a short article about redirect chains in article.txt. Run python topics.py and inspect the candidate phrases and scores.
    from pathlib import Path
    from keybert import KeyBERT
    
    text = Path('article.txt').read_text(encoding='utf-8')
    model = KeyBERT()
    for phrase, score in model.extract_keywords(
        text, keyphrase_ngram_range=(1, 2), stop_words='english', top_n=8
    ):
        print(f'{phrase}: {score:.3f}')
  2. Run it again on a second article about sitemap coverage using the same parameters. Compare whether the two shortlists reflect the different topics.
  3. Repeat the first run with uncleaned text to see whether repeated navigation changes the shortlist. Use the cleaned version for your inventory and review ambiguous phrases manually.

Proposed evaluation based on the upstream documentation. Use the same model and preprocessing when comparing pages.

Official installation and usage documentation ↗

Where KeyBERT fits

Use KeyBERT to compare the main phrases in existing articles before merging or expanding pages that cover similar topics.

When to choose another tool
Use Search Console or keyword research data when you need observed demand or query performance for a specific market.
First evaluation
Extract one- and two-word phrases from two articles after removing navigation and shared boilerplate. Compare the returned phrases with the articles’ main headings.
How to interpret the result
Inspect the top phrases in context. A similarity score reflects the document and embedding model. It does not measure search demand, ranking potential or factual accuracy.

Reviews and discussion

Reader reviews

Be the first to share your experience with this tool.

Discussion

Ask a question or share a setup tip.

Write a review or join the discussion

Reviews should describe your own experience. All submissions are moderated.

Flag an outdated detail

Include the field that needs updating and a source link. Copy your draft, then use the corrections policy to contact the editorial team.