So in this lecture, we're going to try and motivate the problem of recommender systems as an important Data Product which we could develop later on. And we'll describe several examples of recommender systems on the Web. Okay, so what is recommendation and why [INAUDIBLE] do it? Well, recommender systems are very common. We're using them all the time whenever we get recommended content by services on the web. So a classic example of what a recommender system might be doing is to help us discover new content. And this is an example from Amazon, some recommendations of instant video for me on the basis of a movie I have watched previously. These movies are somehow similar or they're movies I'm likely to rate highly if I enjoyed the previous movie. Maybe you could try and guess what that is. Another set of example of a recommender systems would be, and following again from Amazon, in this case I'm watching Harry Potter 1. And it says you might like Harry Potter 2, Harry Potter 3, Harry Potter 4, Harry Potter 5, Harry Potter 6, Harry Potter 7 part 2, Harry Potter 7 part 3, you get the idea. So well, are these good or bad recommendations? This may be not so obvious. They're not really helping me to discover new content, rather this recommender system would be helping me to identify content I probably already know about. And it's just surfacing it to a more relevant or to a more prominent part of the user interface. It can also be useful. So it's just helping us find or helping us access content that we were already looking for. Recommender systems could also hep us to discover just which things go together. This is an example again from Amazon of people who bought x also bought y. So if you're looking at some pair of jeans and it says you might like this shirt, somehow think those items go together or be mutually compatible, perhaps on the basis of people who would co-purchase both items. A recommender system should also try and personalize user experiences in response to user feedback is an example from a streaming music service where you can click thumbs up or thumbs down on a particular track or you can skip the track. And based on what you do, the future tracks that are surfaced should be personalized to better match your preferences. So if you click thumbs up, maybe you'll get more tracks that are similar to this one. If you click thumbs down, maybe you'll get tracks that is somehow different. Also the advertisement that's recommended here is really the operative recommender system. And it's somehow recommending a product that is contextually relevant to the music you are listening to. This is another example of a recommender system from Netflix. In this case, they're just trying to surface their estimate of how I would rate a movie. I think this feature's non-interactive on Netflix. What it was doing is trying to estimate in a personalized way what would I rate this movie. The important thing to notice here is this rating of 3.6 stars is not like the global average rating from Mad Max rather it's saying, how would I rate this movie? And that could help me to know whether this movie is good for my personal interests. Or it could just give me additional contextual information that says this is how you'll rate this movie. Maybe tonight I just happen to be in the mood for a movie that I would rate 3.6 stars. Okay, outside of things like e-commerce like the previous examples I was showing which may be the most obvious. We see recommender systems all the time, for example, in priority inbox, which ranks emails according to whether I'm likely to respond to them. Broadly speaking, that's an example of a recommender system. Or in things like personalized health, if I'd like to estimate how an individual might respond to a specific regime of treatment or things like friend recommendation on Facebook to People You May Know feature is really an example of a recommender system. It's not recommending things to people, it's recommending people to other people. So whenever we're trying to estimate this personalized reaction of an individual to some stimulus or item or some product, broadly speaking that's an example of recommendation technology being applied. Okay, so there are all these different versions of recommender systems, content discovery, building better user interfaces, finding things that go together, personalizing experiences or just estimating what we like. Really, recommender systems are just using machine learning to build models of people's preferences, opinions, and behavior. Okay, so in this lecture, all we've really done is to describe some of the most common use cases of recommender systems that we might observe on the web to try and motivate the importance of this problem as a data product that's worthy of study.