Future directions

[riccardopoiani]

• Search for precision @ low values recommendations and build an hybrid that re-ranks
• Filter predictions in some way
  • maybe subclass never bought from users of that age
  • subclass never bought from that users (if many ratings)
• To handle users with many ratings one can try the following: exclude some ratings and search for an similar users/items. Once this is done recommend, filtering the excluded ratings.
• Tune feature weighting for users with many ratings / users with age 4
• Score weighting using side information