(...) is about providing consumers with multiple ways to discover content: generic ‘you might also like’ recommendations based on content, personalized recommendations based on a profile of historic storefront events, or via social recommendations from other consumers through lists and reviews.
AgentArts' proprietary datamining technology translates various types of consumer activity (page views, previews, purchases) into content relationships. AgentArts’ patented datamining algorithm is highly effective at building accurate content relationships for a range of different content categories, and is particularly effective at building quality content relationships for lesser known content.
Sounds interesting. But what is the difference over other recommendation approaches, as used by Amazon, Pandora or Last FM?
1 comment:
The approach Pandora takes is content specific in the sense that they have a lot of humans listening to music and tagging it with very very detailed metadata about the music and use this to drive their recommendations.
Amazon, AgentArts and Last.fm all use a which is somewhat based on Collaborative Filtering. Around 1999, several companies, Amazon and AA included (last.fm was invented then) independently came up with item to item based filtering which can scale much much better than traditional col.filtering.
The details each company (and others) use differ in subtle ways and often the value is in the use case or domain for example, Last.fm is very music focused and would have little domain knowledge or tweaks for other conent types (of course they could develop this), Amazon is internal only and covers multiple data types and AgentArts' product is a white label solution which can be used across many different content forms by large ecommerce services. Thus the Agentarts solution is more focused on being an installable enterprise product than say last.fm which is product specific consumer focused.
disclosure - I'm the CTO of AgentArts
Post a Comment