Users Each user is represented as a d dimensional feature vector x i R d This

Users each user is represented as a d dimensional

This preview shows page 3 - 4 out of 7 pages.

Users. Each user is represented as a d dimensional feature vector x i R d . This feature vector is typically very sparse i.e. only ˆ d d of these coordinates are non-zero for an average user (some users may of course have more or less coordinates set to non-zero values in their representation). Nothing particularly specific is known about these d features except that they will always take non-negative values. These features were possibly arrived at by looking at normalized bag-of-“words” features where “words” may have been items these users viewed or bought in the past or their profile history, but we are not sure since the engineers who developed the feature extraction routines have left the company. Items. There are a total of L items in the inventory out of which we wish to recommend items to the users. Each item is called a label . For each user i , we are told which all items do they like by giving us a label-vector y i ∈ { 0 , 1 } L for that user. For any item j [ L ], y i j = 1 indicates that user i definitely likes item j . Note that a user may like multiple items in which case all those coordinates would get turned on in the label vector for that user. This illustrates the difference between multi-class classification and multi-label classification. In multi-class settings, a data point can belong to exactly one of the possible classes but in multi-label settings, a data point may be associated with one or more of the labels. Going by what is mentioned about the items above, the reader may be tempted to think that just as y i j = 1 indicates that user i definitely likes item j , then y i j 0 = 0 should indicate that user i dislikes item j 0 . However, this is not true: y i j 0 = 0 indicates that either user did not express an interest when shown the item j 0 or else the user was never shown the item j 0 at all! In an e-commerce website with 10 million items, it is unreasonable to expect a user to have interacted with, and to have told us, all the items in which they are interested – all recommendation systems have to be careful about this fact. Your Task. Your task is to learn a recommendation system model which, when given a user feature vector, can tell us the top 5 items (labels) in which that user would be most interested. Given a test user, your method should return a vector pred of length 5 with each entry between 0 and L - 1. The first entry of the vector i.e. pred [0] (note the zero-based indexing) will be interpreted by us as the item your method thinks is most likely to be liked by the user, the second entry of the vector i.e. pred [1] is the next most likely item to be liked by the user, and so on. Note that the first item is indexed as the 0 th label and the last item is indexed as the ( L - 1) th label (again in zero-based indexing). Thus, if you think the most suitable item for the user is the ninth item, you must set pred [0] = 8. Thus, we expect a ranking of the 5 most suitable items for that user. Note that you cannot cheat by repeating items in your recommendation vector e.g. by setting pred [0] = pred [1] = . . . . Our evaluation code will automatically remove duplicates from your recommendations.
Image of page 3
Image of page 4

You've reached the end of your free preview.

Want to read all 7 pages?

  • Fall '16
  • Piyush Rai
  • Computer file

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask You can ask ( soon) You can ask (will expire )
Answers in as fast as 15 minutes