Recommender systems ML block diagram

Loading...

來自 University of Washington 的課程

机器学习基础：案例研究

7821 個評分

University of Washington

机器学习基础：案例研究

7821 個評分

課程 1（共 4 門，Specialization Machine Learning）

你是否好奇数据可以告诉你什么？你是否想在关于机器学习促进商业的核心方式上有深层次的理解？你是否想能同专家们讨论关于回归，分类，深度学习以及推荐系统的一切？在这门课上，你将会通过一系列实际案例学习来获取实践经历。在这门课结束的时候，

從本節課中

Recommending Products

Ever wonder how Amazon forms its personalized product recommendations? How Netflix suggests movies to watch? How Pandora selects the next song to stream? How Facebook or LinkedIn finds people you might connect with? Underlying all of these technologies for personalized content is something called collaborative filtering. <p>You will learn how to build such a recommender system using a variety of techniques, and explore their tradeoffs.</p> One method we examine is matrix factorization, which learns features of users and products to form recommendations. In an iPython notebook, you will use these techniques to build a real song recommender system.

Recommender systems ML block diagram4:52

與講師見面

Carlos Guestrin
Amazon Professor of Machine Learning
Computer Science and Engineering
Emily Fox
Amazon Professor of Machine Learning
Statistics

[MUSIC]

In this module we talked about a notion of collaborative filtering and

a couple different recommender systems for implementing that idea where we leverage

the types of other products people purchased to recommend other products.

And we explored this notion of having some set of customers and

products and understanding relationships between these in the context of

product recommendation tasks, thinking about movie recommendation.

And in the IPython notebook you guys explore this very concretely with

an implementation of a song recommender.

Once again, let's revisit this machine learning workflow, but

in the context of recommender systems.

In this case, what's our training data?

Well it's our user,

product, ratings, table.

And what we're gonna do is we're gonna extract some set of features,

which in this case are a user ID, product ID pair

And the goal here is to predict the rating that some user would give to some product.

So user ID, product ID, rating.

So this is our predicted rating, y hat.

If we're specifically looking, and I guess I, sorry,

I should mention that, what is the model that we're looking at?

The model is, for example,

we talked about many different models but the one we're gonna talk about quite

extensively in the matrix factorization course is matrix factorization.

And matrix factorization has some set of parameters.

W hat, which are the estimated parameters, or

that notation represents estimated parameters.

And what are the parameters of matrix factorization?

Well, it's a set of features for every user.

And it's a set of features for every product.

Okay, so these are our parameters, but

we also talked about a featurized version of matrix factorization.

So in that case, in addition to our features

being the user ID and product ID, we might also consider other things,

we might have a list of other features like the age of the user,

the gender of the user, a description of the product, and so on.

And in that case, we would also add weights on these features.

So let me just add some extra set of,

I'll call them w0 to make it difference than this w,

some other set of weights here that are also parameters of this model.

Okay, again we'll get into this in a lot more detail in the course on this, but

the idea is we're gonna take our predicted rating and

we're gonna see how well our model is fitting the data.

And the way we're gonna do that is we're gonna take our actual data,

our real ratings, so these are actual

ratings that are in our training data set.

And we're gonna compare to the predicted ratings.

Okay, so one metric we talked about to measure the air between the predicted

ratings and our actual observed ratings was residual sum of squares.

Just like in regression, but there are also other metrics we could consider and

we could talk about those more later.

But the point is that from some notion of error between our predicted values and

our observed values, we're gonna have some machine learning algorithm.

Again, we're leaving this for

the actual module on matrix factorization to describe what this algorithm is.

But what it's going to do is it's going to iteratively update our features for

the user and for the product until we get good agreement between our predicted

ratings and our actual observed ratings.

Okay, so in this module you learned how to do collaborative filtering in practice, so

now you can go out actually implement a recommender system.

You can do a gift recommender for your family,

which will make holiday shopping really easy.

Or you can build a new song recommender that's even better than the one that we

demoed.

So lots and lots of cool things you can do with collaborative filtering.

[MUSIC]

探索我們的目錄

免費加入並獲得個性化推薦、更新和優惠。

Coursera 致力於普及全世界最好的教育，它與全球一流大學和機構合作提供在線課程。

© 2018 Coursera Inc. 保留所有權利。

通過 App Store 下載

通過 Google Play 獲取