Selecting content items based on received term using topic model

• US 10,452,710 B2
• Filed: 09/30/2015
• Issued: 10/22/2019
• Est. Priority Date: 09/30/2015
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Agrawal, Varun
• Primary Examiner(s): Hoang, Son T
• Application Number:
  US14/870,245
• Publication Number:
  US 20170091322A1
• Time in Patent Office: 1,483 Days
• Field of Search: None
• CPC Class Codes: G06F 16/3338 Query expansion
• CPC Class Codes: G06F 16/3346 using probabilistic model
• CPC Class Codes: G06F 16/638 Presentation of query results
• CPC Class Codes: G06F 16/68 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/685 using automatically derived transcript of audio data, e.g. lyrics speech recognition G10L15/00
• CPC Class Codes: G06F 16/90324 using system suggestions
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9538 Presentation of query results

Abstract

Lyrics associated with songs are processed to generate a probabilistic topic model that includes probabilities for terms of the lyrics with respect to one or more predetermined topics. At a later time, a user may desire to hear songs that are associated with a particular term, and may submit the term using a user interface. When the term is received, the probabilities of the probabilistic model are used to identify a topic of the predetermined topics that is most likely associated with the received term. The probabilistic model is used to identify songs that are associated with the identified topic, and some or all of the identified songs are presented as being related to the received term in the user interface.

Claims

1. A system for recommending songs based on terms associated with a plurality of songs, comprising:

at least one computing device; and

a recommendation engine adapted to;

for each song of the plurality of songs, determine a plurality of terms associated with the song;

determine a plurality of topics, each topic including at least one of the plurality of terms;

for each term of the plurality of terms associated with each song, determine a matching probability for the term with respect to each topic of the plurality of topics;

for each song of the plurality of songs, determine a matching probability for the song with respect to each topic of the plurality of topics;

using a user interface, receive a term of the plurality of terms associated with each song of the plurality of songs from a particular user;

receive, using the user interface, user preference data associated with the particular user, the user preference data comprising an indication of at least one of a type or a category of songs that are liked or disliked by the particular user;

based on the determined matching probability for the received term with respect to each topic of the plurality of topics, select a topic of the plurality of topics;

based on the selected topic, the user preference data, and the determined matching probability for each song with respect to the selected topic, select one or more songs of the plurality of songs; and

provide the selected one or more songs in response to the received term.

2. The system of claim 1, wherein the recommendation engine is further adapted to normalize the plurality of terms associated with each song.

3. The system of claim 1, wherein the recommendation engine is further adapted to generate a term-frequency matrix from the plurality of terms associated with each song.

4. The system of claim 3, wherein the matching probability for each term with respect to each topic of the plurality of topics is determined based on the term-frequency matrix using a probabilistic graphical model.

5. The system of claim 4, wherein the probabilistic graphical model is a latent Dirichlet allocation model.

6. A method, implemented by at least one computing device, for recommending songs based on terms associated with a plurality of songs, comprising:

for each song of the plurality of songs, determine a plurality of terms associated with the song;

determining a plurality of topics, each topic including at least one of the plurality of terms;

for each term of the Plurality of terms associated with each song, determining a matching probability for the term with respect to each topic of the plurality of topics;

for each song of the plurality of songs, determining a matching probability for the song with respect to each topic of the plurality of topics;

using a user interface, receiving a term of the plurality of terms associated with each song of the plurality of songs from a particular user;

receiving, using the user interface, user preference data associated with the particular user, the user preference data comprising an indication of at least one of a type or a category of songs that are liked or disliked by the particular user;

based on the determined matching probability for the received term with respect to each topic of the plurality of topics, determining a topic of the plurality of topics;

based on the determined topic, the user preference data, and the determined matching probability for each song with respect to the determined topic, selecting one or more songs of the plurality of songs by the computing device; and

providing indicators of the selected one or more songs in response to the received term via the user interface.

7. The method of claim 6, wherein the topic model is a probabilistic topical model.

8. The method of claim 7, wherein the probabilistic topical model is a latent Dirichlet allocation model.

9. The method of claim 6, wherein the indicators of the selected one or more songs comprise a playlist.

10. The method of claim 6, further comprising generating the topic model by:

generating a term-frequency matrix from the plurality of terms associated with each song of the plurality of songs;

for each term of the plurality of terms, determining the matching probability for the term with respect to each topic of the plurality of topics using the term-frequency matrix; and

for each song of the plurality of songs, determining the probability for the song with respect to each topic of the plurality of topics using the term-frequency matrix.

11. The method of claim 10, further comprising normalizing the plurality of terms associated with each song.

12. The method of claim 10, wherein determining the topic of the plurality of topics that is associated with the received term comprises determining g topic with a highest determined probability for the received term.

13. The method of claim 10, wherein selecting the one or more songs that are associated with the determined topic from the plurality of songs comprises ranking the plurality of songs of the plurality of songs based on the matching probabilities of the plurality of terms associated with each song for the determined topic, and selecting the one or more songs based on the ranking.

14. A method for recommending content items based on terms associated with a plurality of content items, comprising:

receiving the plurality of content items by a computing device;

for each content item, determining a plurality of terms associated with the content item by the computing device;

determining a plurality of topics by the computing device, each topic including at least one of the plurality of terms;

for each term of the plurality of terms associated with each content item, determining a matching probability for the term with respect to each topic of the plurality of topics by the computing device;

for each content item of the plurality of content items, determining a matching probability for the content item with respect to each topic of the plurality of topics by the computing device;

receiving, using a user interface, a term of the plurality of terms associated with each of the plurality of content items by the computing device from a particular user;

receiving, using the user interface, user preference data associated with the particular user, the user preference data comprising an indication of at least one of a type or a category of content items that are liked or disliked by the particular user;

based on the determined matching probabilities for the received term with respect to each topic of the plurality of topics, selecting a topic of the plurality of topics by the computing device;

based on the selected topic, the user preference data, and the determined matching probability for each content item with respect to the selected topic, selecting one or more content items by the computing device; and

providing the selected one or more content items in response to the received term by the computing device.

15. The method of claim 14, wherein the content items are songs, and the plurality of terms associated with each content item are lyrics.

16. The method of claim 14, further comprising normalizing the plurality of terms associated with each content item.

17. The method of claim 14, wherein the content items are documents, and the plurality of terms associated with each document are terms used in the document.

18. The method of claim 14, further comprising generating a term-frequency matrix from the plurality of terms associated with each content item.

19. The method of claim 18, wherein the matching probability for each term with respect to each of the plurality of topics is determined based on the term-frequency matrix using a probabilistic graphical model.

20. The method of claim 19, wherein the probabilistic graphical model is a latent Dirichlet allocation model.