Cross media recommendation

Patent metadata

Field	Value
Publication number	US-10558682-B2
Application number	US-201715420311-A
Country	US
Kind code	B2
Filing date	Jan 31, 2017
Priority date	Mar 18, 2013
Publication date	Feb 11, 2020
Grant date	Feb 11, 2020

Abstract

Methods, systems and computer program products are provided for cross-media recommendation by store a plurality of taste profiles corresponding to a first domain and a plurality of media item vectors corresponding to a second domain. An evaluation taste profile in the first domain is applied to a plurality of models that have been generated based on relationship among the plurality of taste profiles and the plurality of media item vectors, and obtain a plurality of resulting codes corresponding to at least one of the plurality of media item vectors in the second domain.

First claim

What is claimed is:

1. A system for cross-media recommendation, comprising:
a first database operable to store a plurality of taste profiles corresponding to a first domain;
a second database operable to store a plurality of media item vectors corresponding to a second domain; and
at least one processor configured to:
generate a training set based on the plurality of taste profiles and the plurality of media item vectors, wherein at least a portion of the training set includes ground truths across different domains;
apply an evaluation taste profile in the first domain to a plurality of models generated based on a relationship among the plurality of taste profiles and the plurality of media item vectors, wherein the plurality of models are trained based on the training set; and
obtain a plurality of resulting codes corresponding to at least one of the plurality of media item vectors in the second domain, wherein a plurality of weighted term vectors are generated by multiplying, for each term in a taste profile of the plurality of taste profiles, an affinity by a probability that the term is associated with a media item.

2. The system according to claim 1 , wherein the first domain is music and the second domain is any one, or a combination, of books, movies, or games.

3. The system according to claim 1 , wherein the portion of the training set that includes the ground truths across different domains includes an even distribution between positive examples and negative examples.

4. The system according to claim 1 , wherein the at least one processor is further configured to obtain, for each of the obtained resulting codes, an output class, and
wherein each output class indicates a predetermined value if the evaluation taste profile and a media item corresponding to the respective resulting code form a ground truth included in the training set.

5. The system according to claim 4 , wherein the at least one processor is further configured to obtain, for each of the obtained resulting codes, a confidence interval, and
wherein each confidence interval indicates an amount of confidence in a relation between the evaluation taste profile and the media item vector corresponding to the respective resulting code.

6. A non-transitory computer-readable medium having stored thereon one or more sequences of instructions for causing one or more processors to perform:
storing a plurality of taste profiles corresponding to a first domain;
storing a plurality of media item vectors corresponding to a second domain;
generating a training set based on the plurality of taste profiles and the plurality of media item vectors, wherein at least a portion of the training set includes ground truths across different domains;
applying an evaluation taste profile in the first domain to a plurality of models generated based on a relationship among the plurality of taste profiles and the plurality of media item vectors, wherein the plurality of models are trained based on the training set; and
obtaining a plurality of resulting codes corresponding to at least one of the plurality of media item vectors in the second domain, wherein a plurality of weighted term vectors are generated by multiplying, for each term in a taste profile of the plurality of taste profiles, an affinity by a probability that the term is associated with a media item.

7. The computer-readable medium according to claim 6 , wherein the first domain is music and the second domain is any one, or a combination, of books, movies, or games.

8. The computer-readable medium according to claim 6 , wherein the portion of the training set that includes the ground truths across different domains includes an even distribution between positive examples and negative examples.

9. The computer-readable medium according to claim 6 , wherein the one or more sequences of instructions further cause the one or more processors to perform: obtaining, for each of the obtained resulting codes, an output class, and
wherein each output class indicates a predetermined value if the evaluation taste profile and a media item corresponding to the respective resulting code form a ground truth included in the training set.

10. The computer-readable medium according to claim 9 , wherein the one or more sequences of instructions further cause the one or more processors to perform:
obtaining, for each of the obtained resulting codes, a confidence interval,
wherein each confidence interval indicates an amount of confidence in a relation between the evaluation taste profile and the media item vector corresponding to the respective resulting code.

11. A method for cross-media recommendation, comprising the steps of:
storing a plurality of taste profiles corresponding to a first domain;
storing a plurality of media item vectors corresponding to a second domain;
generating a training set based on the plurality of taste profiles and the plurality of media item vectors, wherein at least a portion of the training set includes ground truths across different domains;
applying an evaluation taste profile in the first domain to a plurality of models generated based on a relationship among the plurality of taste profiles and the plurality of media item vectors, wherein the plurality of models are trained based on the training set; and
obtaining a plurality of resulting codes corresponding to at least one of the plurality of media item vectors in the second domain, wherein a plurality of weighted term vectors are generated by multiplying, for each term in a taste profile of the plurality of taste profiles, and affinity by a probability that the term is associated with a media item.

12. The method according to claim 11 , wherein the first domain is music and the second domain is any one, or a combination, of books, movies, or games.

13. The method according to claim 11 , wherein the portion of the training set that includes the ground truths across different domains includes an even distribution between positive examples and negative examples.

14. The method according to claim 11 , further comprising: obtaining, for each of the obtained resulting codes, an output class, and
wherein each output class indicates a predetermined value if the evaluation taste profile and a media item corresponding to the respective resulting code form a ground truth included in the training set.

15. The method according to claim 14 , further comprising:
obtaining, for each of the obtained resulting codes, a confidence interval,
wherein each confidence interval indicates an amount of confidence in a relation between the evaluation taste profile and the media item vector corresponding to the respective resulting code.

Assignees

• Spotify Ab

Inventors

• Whitman Brian

Classifications

• G06F16/435

  Filtering based on additional data, e.g. user or group profiles · CPC title

• G06F16/26 Primary

  Visual data mining; Browsing structured data · CPC title

• G06F16/24578

  using ranking · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F16/635

  Filtering based on additional data, e.g. user or group profiles · CPC title