This is the FAQ for the Recommendz system. Recommendz attempts to provide recommendations to you based on previous information regarding what you like and dislike. It is a research project at McGill University. The project is supervised by Professor Gregory Dudek at the Center for Intelligent Machines. Core infrastructure for Recommendz was build by Matt Garden (now at Nortel) as part of his completed MSc thesis. Several other people are current and ongoing contributors. Web site design partly by Krys Dudek.

If you are interested in joining the Recommendz research group as a graduate student, please drop us a line.

• Why do I have to rate movies? I just want some information.

The way we determine what to recommend depends on building a mathematical model of your preferences. To do this, we need information on your likes and dislikes; the more information the better.
• What's an overall rating?

The overall rating is what you thought of the movie: was it bad, good or great?
• Do I have to see a movie before I rate it? What about movies I would never want to see?

You can rate a movie you have not seen, if you feel confident you can guess what you would think of it. If general, try to avoid saying much about the features of such movies since you might guess wrong (but to rate a movie you need to say something about at least one feature).
• How many movies do I need to rate?

You should be able to get recommendations if you rate as few as 3 or 4 movies. To get reliable recommendations you probably will need to rate 3 or 5 movies that you like, as well as 3 movies that you dislike, so the system can determine both what kinds of movies to recommend, and what kinds not to recommend. In general,the more movies you rate the better the system works.
• Who put in those crazy features, and how did you choose them?

The features that you see are those that have been suggested by our users. Popular features are displayed more often, while obscure ones only show up occasionally (but there are a lot of obscure ones).
• How can I make the recommendations for me more reliable?

Rate more items and rate several meaningful features per item: the more the better. If you want to see more alternative features you might use for an item, you can click the reload button, or the "submit and add more features" button. Remember, there are two ways to improve the recommendations:
- Rate more movies covering both good and bad ones.
- Provide more features for movies you rate.
• How well does it work?

We have solid data that shows that the quality of the recommendations is better than what you get by simply reading reviews or looking at the average ratings of other people.
• How do you know the recommendations are so good?

The statistical technique is called cross-validation. Simply put, we consider one of the movies you rated, and look at what the system would predict for you. We compare the prediction to your actual rating. We do this for every movie.
• Why is it so good? How can that be?

The recommendations are based on knowledge of your tastes. For example, many animated movies and many violent action movies get positive reviews on average. Many people, however, will only like one or the other. Our system tries to figure out if you're a "cartoon person" or an "action movie person". If it figures out you're not an "action movie person", for example, it can not recommend action movies even though they are liked on average. (Note: this explanation is rather over-simplified just to get the idea across.)
• Why all this fancy stuff, aren't averages good enough?

Consider this simple observation: most people love some movies and hate others, but if you look at the average ratings for almost any movie it's almost invariably a "middling" value, usually between 6 and 8. Thus, a simple average is not a very good reflection of people's individual tastes.
• How does it work?

The system uses a technique called collaborative filtering, but it also combines that with several other methods. We have some published work on our approach, such as the following paper (PDF document). It's a bit dry. Visit the mobile robotics lab publications pages for additional documents.
• How can I get more detail on a movie?

If the item info page shows a picture, you can click on that to be taken to the relevant page at the IMDB (the Internet Movie Database).