This (collaborative filtering) is a better approach, but it depends very much on the algorithm used, and how many movies you have already rated. I have not su

This (collaborative filtering) is a better approach, but it depends very much on the algorithm used, and how many movies you have already rated. I have not subscribed to MovieLens, and don't know what algorithms it uses. I developed an online recommendation system (see http://web.onetel.com/~hibou/morse/MORSE.html http://web.onetel.com/~hibou/morse/MORSE.html for details) about 20 years ago. For a user who had entered sufficient ratings for other movies, the algorithm I used (described in the paper) performed better than the more widely used average of M nearest neighbours, which in turn is better than simply averaging everyone's ratings. All it used were the individual ratings given by its users. It was entirely agnostic as to genre (which is quite fuzzy), cast, or crew. I was aware of a tendency for people to rate movies more highly if they watched them recently, but never asked users for this data. Surprisingly, using the ratings of only the one person whose tastes are closest to your own (i.e. M=1) results in very inaccurate predicted ratings.