×

System and method for estimating user ratings from user behavior and providing recommendations

  • US 7,756,879 B2
  • Filed: 07/22/2005
  • Issued: 07/13/2010
  • Est. Priority Date: 07/23/2004
  • Status: Expired due to Fees
First Claim
Patent Images

1. A recommendation system for providing a recommendation set of recommended items from an item data store to a user, the recommendation system comprising:

  • a memory storing the item data store;

    an interface adapted to allow the user to browse an item set made from items in the item data store;

    a processor connected to the memory and the interface, the processor comprising;

    a ratings estimation module for;

    receiving item property data comprising at least one item attribute for each item in the item set, wherein the at least one item attribute is a property of the item;

    receiving training set data including user-specific information, wherein the user-specific information comprises at least one user-specific property that relates only to the user, and wherein the at least one user-specific property includes viewing time of at least one item browsed by the user, wherein the viewing time is an individual viewing time the user viewed the specific item; and

    generating a training set of items comprising the at least one item browsed by the user; and

    generating an item rating matrix by estimating user ratings for the items in the training set based on at least one rating predictor variable;

    wherein each rating predictor variable comprises one of the at least one user-specific property and the at least one item attribute;

    wherein an entry in the item rating matrix for a training set item e is generated according to;


    Item Ratinge=f(x1, x2, . . . , xn) 

    wherein each xi is a value of the ith rating predictor variable, n is the number of rating predictor variables, and f is one of a non-linear combination and a linear combination of the rating predictor variables;

    wherein at least one of the rating predictor variables is the viewing time; and

    ,a recommendation module connected to the ratings estimation module for receiving the item rating matrix and the item set, wherein the recommendation module (i) estimates user ratings for unseen items in the item set based on the item rating matrix without using information about any other user and (ii) generates the recommendation set based on the estimated ratings.

View all claims
  • 2 Assignments
Timeline View
Assignment View
    ×
    ×