navatar.blog

Here is an idea to further improve recommendations for movies, products, services, places etc..

So far, we are treating each individual as always same – and trying to recommend either with collaboration filtering or content filtering – It goes only so far.

For “n” number of reasons, individual is not always behaving in same way. Some of them are hard to gauge but other are easy to get based on several signals. Some of the examples –
* What one likes in morning is not same as the one in evening.
* What one likes in spring is not same as the one in summer.
* What one likes in home is not same as the one in work.
* What one likes in regular is not same as the one in vacation.
* What one likes with alone is not same as the one with family.
* What one watches on mobile is not same as the one in TV.

You get the idea.

Stop treating individual as some unbroken piece of rock. Treat each individual with some degree of multiple personalities and apply clustering to each persona.