AgentArts, a content personalization technology company, has received an interesting patent...it sounds a bit like Amazon.com’s collaborative filtering patent, only more granular: U.S Patent No. 6,763,354 covers the method by which AgentArts’ technology derives relationships between related content items (such as video or music) through datamining consumer preferences and then verifying the relationships between content items.

According to the company, conventional collaborative filtering solutions have a major problem—highly popular items skew recommendations towards a comparatively small number of related items (such as songs by major music artists, or blockbuster movies). Unlike these, AgentArts’ approach seeks out “content experts” for any given item, derives relationships based on these “expert” opinions and verifies these relationships using “experts” in the related items.