Skip to main content

Tag Acceptance Models

When integrating Geneea into a Content Management System (CMS), one of the most important decisions is how journalists will interact with the suggested tags.

We categorize these workflows into three models: Opt-in, Opt-out, and Mixed.

General recommendations for all models:

  • Tags should be suggested automatically (e.g., every 60 seconds, when the article changes, when a paragraph is finished, etc.). We do not advise relying on journalists to request the tags manually because they sometimes forget to do so.
  • No tags returned by the Media API should be hidden from the journalist. Filtering tags by relevance should be handled on the Geneea service side, especially when the feedback loop is implemented.

1. Opt-in Model​

In this mode, tags are suggested, but none are applied to the article until a journalist manually selects them.

  • How it works: The CMS displays a list of suggested tags. The journalist selects the ones they want to add to the article. They might add additional tags manually.
  • Filtering: The Media API is configured to return tags above a certain threshold.
  • Pros: 100% human-verified; virtually no incorrect tags are displayed on the public site.
  • Cons: Slow; tags may be forgotten if the journalist is in a hurry.
  • Best for: High-stakes editorial environments where tagging accuracy is more important than speed.

2. Opt-out Model​

In this mode, all returned tags are automatically applied to the article by default; the journalist might reject them.

  • How it works: The CMS displays a list of suggested tags. The journalist inspects them and removes those they consider incorrect or irrelevant. They might add additional tags manually.
  • Pros: Fast; ensures every article is tagged even if the journalist does nothing.
  • Cons: Higher risk of incorrect or irrelevant tags reaching the public site.
  • Best for: High-volume newsrooms, breaking news, or SEO-heavy workflows where maximum coverage is the priority.

3. Mixed Model (Recommended)​

This is our recommended approach, combining the best of both models above.

  • How it works: The CMS displays all suggested tags split into two groups based on their relevance scores: accepted by default and rejected by default. When the journalist disagrees with the Media API on the relevance of a tag, it is easy to correct: they can move a tag from the accepted to the rejected bin and vice versa.
  • Best for: Most professional newsrooms.
  • Pros: Automates the "obvious" work while leaving the nuanced decisions to the journalist. It drastically reduces both the number of clicks required and the manual entry of tags, while maintaining a high standard of quality.

In this model, we recommend:

  • to suggest tags with relevance above 10–20 (this is configured on the Media API side).
  • to mark tags with relevance above 50–70 as accepted (this is configured on the CMS side). You can tune this to match your editorial strategy:
    • Threshold ~70: Best for highly curated tagging and strict precision. Only the most relevant tags are accepted automatically, relying on journalists to manually select broader topics.
    • Threshold ~50: Best for high-volume tagging. Accepts a wider variety of tags by default, ensuring strong discoverability and SEO benefits even with minimal manual review.
  • for the CMS to remember when a journalist accepts or rejects any of the tags and maintain that state even if the Media API returns updated results. We recommend indicating such "pinned" status visually, distinguishing among: explicitly accepted (✓), accepted by default, rejected by default, and explicitly rejected (✗).

Comparison Summary​

ModelDefault tag statusSpeedEditorial Control
Opt-inAll tags are "Suggestions"LowMaximum
Opt-outAll tags are "Applied"HighMinimum
MixedHigh relevance = Applied; Medium and low = SuggestionHighBalanced