R.I.P. Movie Rating Prediction
There is no longer any reason to bother researching new ways of predicting the ratings users will give to movies. It’s time to move on to more interesting things. But seriously, given the fact that the last few miles of the Netflix competition were hard-fought by combining hundreds of different algorithms, is there much value in trying to improve recommender systems in this way, anymore?
I expect that the Netflix Prize data set, if left open to the public, will still be useful for a number of machine learning tasks where the goal is not necessarily improving recommender systems. So predicting movie ratings may never be really dead. But it is my hope that that as a goal for research will diminish and the focus will start moving towards other aspects of recommender systems still greatly lacking. Like building systems that facilitate discovery of new items.
Factoring in the temporal dimension was a big deal in the latter part of the competition. Sometimes you’re just in the mood for something gloomy. Or something funny. Or something ridiculous. The same movie may totally turn you off a week later. No machine (biological or mechanical) can predict these swings of emotions in the near future, so why bother? Flip that around and let’s find ways of improving the search for items matching our mood at the time.
A system that interactively elicits your mood and guides you to matching items would be incredibly useful, don’t you think?
Agreed. It will become a second tier task within recommender systems rather than the prize driving interest in R.S. I think a primary reason to have stong commercial apps to retail, crm, advertising etc. Movie listings is too niche in comparison.
I like the idea of having a side effect of the RS being better indexes and an interactive app, worked toward such a system at a CRM vendor for customer service portals. Sounds niche as well… Yet the market opportunity is measured in billions.
I do wish that amazons RS was a bit more interactive and could be directly searched against.
Not sure I understand what that RS you were working on for customer service portals is supposed to do.. Can you elaborate or is it an industrial secret kind of thing? Maybe my missing ingredient is that I don’t know what customer service portals are. :)
Might be going on too much of a tangent here… But here goes.
Customers with questions about products, accessories etc often use some kind of ‘tech support’ link off the corp website or techsupport email address to seek assistance.
IMO the best versions of the software systems responding to these customers act like an RS with a nice browsable UI. Why? A search box let’s people filter info, and Google has that down.. Yet in this setting the questions are highly repetative and very self similar w.r.t. language/vocab… Simple search falls down. Add an RS to learn from the many past interactions with customers and a faceting or othe adv UI and the ingredients are there for a system that predicts common customer questions and short cuts them to a solution.
This is mostly unsolved and plenty hard. See any tech support link on the web for examples of how badly it works now and what the possibilities are ..
Ah, I see — cool. Sounds like it’d be fun to work on!
I like the idea of selection by mood. I don’t know about plain Netflix since your mood will change by the time the thing arrives, but it would be cool for Tivo and the Netlix Instant Queue. There could be a top-level menu on the Tivo for “Help Me Decide” where it randomly selects from a bank of questions to ask you about what mood you’re in, and each answer narrows the list of possibilities from your current Tivo list, shows starting within the next 15 minutes, and the Netflix Instant Queue. That would be pretty sweet.