Now that you've seen how recommendation systems are used at Google and in other businesses, it's time to look into the specific scenario that we'll use in the next lab. The core pieces of a recommendation system are: data, the model, and infrastructure to train and serve recommendations to users. Our data set for this scenario will be housing rentals that we want to recommend to our users based on their preferences. We'll use a machine learning model to make these recommendations. A core tenet of machine learning is to let the model learn for itself what the relationship is between the data that you have like user preferences and housing features, and the data that you don't have, like a user's rating on a property that they haven't yet seen. What would this user rate this property if we showed it to them? You will not be writing custom logic. Logic like, if the house is a beach house and the season is summer and the user likes cozy houses, then house number 22 is what we recommend. Can you think of the issues if we used such logic instead of letting the model figure it out? What if there were many different kinds of beach houses, with all kinds of different features like the number of bedrooms, locations, amenities, you get the picture. What if you have different kinds of users. Some users would like to go to beach houses in summer and others would like the shoulder season. Hard-coding logic for all these features and different kinds of user segments isn't scalable, and it assumes that we always know the right answer for every scenario. Let's illustrate this point with an example. Take Google Search. Say you go to Google and search for "giants." What should we show you as the results to make it most relevant for you? If you're in California, should we show results for the San Francisco baseball team and the local games nearby? What about if you're based in New York? Should we tailor the results to show the New York Giants football team instead, and should we write this as a rule? Well, up until a few years ago, this is exactly how Google search worked. There were tons of rules that were part of the search engine code base to decide which sports team to show a user when they search for "giants." The query is "giants" and the user is in the Bay Area, show them results about San Francisco Giants. If the user's in New York area, show them results about New York Giants. If they're anywhere else, show them results about tall people. Just imagine how many, if-then or case statements this would be, and how hard it would be to maintain, and this is just for one query. Multiply this by the large variety of queries that people make, where they make them from, what device they're on, and what kinds of interests different people have, and you can imagine how complex a code base would become. The code base starts to get unwieldy because hand-coded rules are really hard to maintain. So this is where machine learning comes into play. Machine learning scales much better because it doesn't require hard-coded rules. It's all automated. Learning from data in an automated way, that's what machine learning is. Our data set in this case is that we know historically when people searched for "giants" and we showed them a bunch of links, which of those links did people click on? Since we know this, we have a data set of a query term and the links that people liked. Why can't we train a machine learning model to basically provide a ranking of these links? That's exactly what Google itself did internally with a deep learning model called Rank Brain. After rolling it out, the quality of our search ranking results improved dramatically, with the signal from Rank Brain becoming one of the top three for influencing how different link results are ranked. If you're interested, I'll provide a link where you can read more about it. We'll revisit machine learning in greater depth in each of the modules in this course. But for now, just remember that machine learning is this idea that we want to teach the computer using examples, not with rules. Any business application where you have those long switch or case statements or if-then logic manually coded, and you have a history of good labeled data, that is data for which you know a good answer or a bad answer, a history of good labeled data, any such application is a possible application for machine learning.
