I did this for a user study for around 100~ people but only using 1 recommender as opposed to 10. It was a user based CF algorithm with my own tweaks. Overall the recommendations were quite quick. The main bottleneck was the actual loading and updating of the data model (no surprises there). My use case was for an experiment linked to my research, some people did no complete the study. If you are using it in a commercial environment you should probably avoid such a configuration. You could take a nightly batch type approach.
On 29 Jan 2012, at 19:26, Bill Liang wrote: > > Hi all, > I'm considering dynamically instantiating multiple recommenders on the fly. > If I want to create 10 recommenders from previously trained models, and then > get recommendation from each of them, is that a practical thing to do? in > other words, is it going to take too long beyond what average user is willing > to wait for? I know it can be easily tested, I just want to hear from other > people's experience. > Thanks,Bill
