classify-media-relevance

To filter site chrome from crawled release media, classify each item using URL patterns, alt text, filename, and page position to return a keep or drop decision, ensuring editorial content is retained while junk like avatars and tracking pixels is removed.

Media relevance classification handles tracking pixels and oversized files using deterministic pre-checks that evaluate content-type, size hints, and tracking domains before full signal analysis, dropping unsupported items early to avoid redundant work and byte-level image requests.

Yes, you can use alt text and page position to classify editorial images during HTML parsing by weighing these contextual heuristics alongside filename patterns and URL paths to determine if an image is meaningful editorial content or irrelevant site chrome.

The best way to structure media classification output for a release pipeline is a JSON array preserving input order, containing a keep or drop decision, a confidence level, and a brief reason, allowing callers to apply conservative fallbacks for low-confidence drops.

You should avoid byte-level image requests when classifying media relevance to save bandwidth and processing time, relying instead on metadata, URL patterns, and deterministic pre-checks to drop tracking domains and duplicate uploads efficiently.