The only way to do something with a truly cold start is with rules. We support user-defined ranking of items, and promoted items with no other usage data. Content-based similarity recommendations for a particular item are also supported but you have to do a little work to make this happen by attaching the correct properties to items and putting the right properties in the query.
The first thing to start returning non-cold recommendations is popular items, which will give some results with only a few interactions, then item-based will start working since they do not rely on a particular user’s history, then personalized requires a user’s history. The UR blends all these into a single internal query as “fallbacks” so you will get the best recs over the cold-start ones, and with enough data you’ll never get the cold start ones unless you ask specifically for them. On Mar 3, 2017, at 1:14 AM, Masha Zaharchenko <[email protected]> wrote: Hi! Does the Universal Recommender solve the cold-start problem for users(any other way than recommending the most popular items) and items? Thanks, Maria
