Recommender system for content delivery networks

Patent metadata

Field	Value
Publication number	US-9479552-B2
Application number	US-201213483329-A
Country	US
Kind code	B2
Filing date	May 30, 2012
Priority date	May 30, 2012
Publication date	Oct 25, 2016
Grant date	Oct 25, 2016

Abstract

A device includes a processor. The processor is configured to determine a hit time for each of a plurality of content items based on at least one of users' history of access to the content items on content distribution clusters in a content distribution network, the users' past ratings of the content items, and social network information associated with the users. The hit time of a content item indicates a number of times that the content item is likely to be accessed by the users. The processor is further configured to compute caching priorities of the content items based on a caching policy of the device and the determined hit times, and initiate a redistribution, over a network, of the plurality of content items over the content distribution clusters of the content distribution network based on the caching priorities.

First claim

What is claimed is:

1. A method comprising:
caching each of a plurality of content items in one of a plurality of content distribution clusters;
obtaining a user access history of the content items based on recorded information about users' past access to the content items and the users' past ratings of the content items;
determining a hit time for each of the content items based on at least one of the user access history, the users' past ratings, and social network information associated with the users, wherein the hit time of a content item indicates a number of times that the content item is likely to be accessed by user devices associated with the users;
obtaining caching priorities of the content items based on a caching policy at the content distribution clusters and the determined hit times; and
transferring one or more of the content items from a first subset of the content distribution clusters to a second subset of the content distribution clusters based on the caching priorities,
wherein each of the content distribution clusters includes one or more physical devices configured to send one or more of the content items, to one or more of the user devices, without passing the one or more of the content items through any of others of the content distribution clusters.

2. The method of claim 1 , further comprising:
selecting a user behavior model to determine the hit time for each of the content items based on at least the social network information associated with the users.

3. The method of claim 1 , wherein the caching policy is one of:
a least frequently used (LFU) policy, a least recently used (LRU) policy, or a mix of the LFU policy and the LRU policy.

4. The method of claim 3 , wherein obtaining the caching priorities includes
evaluating, for each of the content items, a function of a time since a last access of the content item, a frequency of accessing the content item, and one of the hit times; and
adding, for each of the content items, the results of evaluating to obtain one of the caching priorities.

5. The method of claim 1 , wherein the user access history includes information regarding one of:
which of the users accessed which of the content items; or
how much time each of the users spent on each of the content items.

6. The method of claim 5 , wherein obtaining the caching priorities includes:
predicting the users' ratings of the content items based on the user access history and the users' past ratings;
determining, based on the users' predicted ratings, a probability of a request for each of the content items by each of the users; and
predicting, for each of the content items, one of the hit times based on the probabilities.

7. The method of claim 6 , wherein determining the hit time comprises:
adding the probabilities of a request for the user.

8. The method of claim 6 , wherein predicting the users' ratings includes:
extracting first latent features associated with the content items and second latent features associated with the users based on the user access history and the users' past ratings; and
calculating the users' predicted ratings using the first latent features and the second latent features.

9. The method of claim 8 , wherein extracting the first latent features associated with the content items and the second latent features associated with the users includes:
selecting values, as the first latent features and second latent features, that minimize a root mean square error between the predicted ratings and the users' past ratings.

10. The method of claim 6 , wherein, when the user access history includes how much time each of the users spent on each of the content items, each of the users' past ratings for one of the content items is a ratio of a time the user spent on accessing the one of the content items to a total duration of the one of the content items.

11. The method of claim 1 , wherein determining a hit time for each of the content items includes:
constructing a model neighborhood for a user;
obtaining a predicted voting value for the content item for each of the users based on the model neighborhood; and
determining the hit time for the content item based on the predicted voting values for the users,
wherein the model neighborhood is a subset of the users, and
wherein the predicted voting value is a probability that the user will select the content item given the model neighborhood.

12. The method of claim 11 , wherein constructing the model neighborhood includes:
extracting first latent features associated with the content items and second latent features associated with the users based on the user access history and the users' past ratings; and
identifying first users, among the users, closest to the user in a space of the second latent features, and
including the identified first users in the model neighborhood.

13. The method of claim 12 , wherein identifying the first users includes:
identifying the first users closest to the user in the space of the second latent features based on Pearson correlation.

14. The method of claim 11 , wherein constructing the model neighborhood further includes:
identifying first users, among the users, closest to the user based on social network information for the user; and
including the identified first users in the model neighborhood.

15. The method of claim 11 , wherein constructing the model neighborhood further includes:
extracting first latent features associated with the content items and second latent features associated with the users based on the user access history and the users' past ratings;
identifying first users, among the users, closest to the user in a space of the second latent features;
identifying second users, among the users, closest to the user based on social network information for the user; and
including the first users and the second users in the model neighborhood.

16. The method of claim 15 , wherein the first users and the second users are tuned to obtain maximum hit times.

17. A device comprising,
one or more processors to:
determine a hit time for each of a plurality of content items based on at least one of users' history of access to the content items on content distribution clusters in a content distribution network, the users' past ratings of the content items, and social network information associated with the users, wherein the hit time of a content item indicates a number of times that the content item is likely to be accessed by user devices associated with the users;
compute caching priorities of the content items based on a caching policy of the device and the determined hit times; and
initiate a redistribution, over a network, of the plurality of content items over the content distribution clusters of the content distribution network based on the caching priorities,

wherein each of the content distribution clusters includes one or more physical devices configured to send one or more of the content items, to one or more of the user devices, without passing the one or more of the content items through any of others of the content distribution clusters.

18. The device of claim 17 , wherein when the one or more processors initiate the redistribution, the one or more processors are further configured to:
send a content item to a remote device in one of the content distribution clusters;
receive a content item from a remote device in one of the content distribution clusters;
send the computed caching priorities, to another device, for redistributing the plurality of co

Assignees

• Chen Ming
• Yang Xiwang
• Liu Yong
• Zhang Lei
• Verizon Patent & Licensing Inc

Inventors

• Chen Ming
• Yang Xiwang
• Liu Yong
• Zhang Lei

Classifications

• H04L65/4084 Primary

  Electricity · mapped topic

• H04L67/22

  Electricity · mapped topic

• H04L67/2852

  Electricity · mapped topic

• G06Q10/10 Primary

  Office automation; Time management · CPC title

• H04L67/306

  User profiles · CPC title