Granted Patent B2

US 8,666,979 · App. 12/757,929 · Granted Mar 4, 2014

Abstract

Claims (79)

1. A method for recommending interesting content, comprising:

obtaining a content profile comprising a vector of user words extracted from messages associated with a user;

identifying one or more Uniform Resource Locators (URLs) in a plurality of messages generated by other users comprising followees that the user follows and further followees that are followed by the followees;

generating one or more content pointer profiles each comprising a vector of words associated with one of the URLs located within at least one of the plurality of messages by extracting the words from the at least one message from the other users that include that URL and combining the extracted words as the vector of words;

comparing the content profile with each of the content pointer profiles and determining a relevance score for the URLs associated with the content pointer profiles by calculating a similarity between the vector of user words and the vector for each URL;

determining a vote score for each of the URLs, comprising:

assigning a composition power to each of the other users associated with the plurality of messages that include one of the URLs based on a frequency of messages generated by that other user;

assigning a trust power to each of the other users associated with the plurality of messages that include the URL based on a trustworthiness of that other user;

determining a vote power for each of the other users associated with the plurality of messages that include the URL by combining the composition power and the trust power for that other user; and

generating a vote score for each of the URLs by summing the vote power for the other users associated with the plurality of messages that include that URL;

combining the vote score with the relevance score as a rank score for each of the URLs;

ranking the URLs based on the rank score; and

applying a threshold to the URLs and selecting the URLs that satisfy the threshold as the most relevant URLs to the user.

2. The method according to claim 1 , wherein the content profile comprises one of a self-profile and a network profile.

3. The method according to claim 2 , further comprising:

generating the self-profile, comprising:

extracting words from the messages associated with the user and generating a list of the extracted words;

selecting one or more of the extracted words for inclusion in the self-profile and determining a strength of each selected word; and

placing the selected words and associated strengths along the vector as the self-profile.

4. The method according to claim 2 , further comprising:

generating the network profile, comprising:

identifying profiles of individuals associated with the user, wherein each profile comprises words and associated strengths for that individual;

applying a threshold to each profile and selecting the words that satisfy the threshold;

combining the selected words from each individual profile and determining a strength for each selected word; and

placing the selected words and associated strengths along a vector as the network profile.

5. The method according to claim 1 , further comprising:

generating the content pointer profile for at least one of the URLs, comprising:

selecting one or more of the words for inclusion in the content pointer profile and determining a strength of each selected word; and

placing the selected words and associated strengths along a vector as the content pointer profile.

6. The method according to claim 1 , further comprising:

determining the trust power based on at least one of a number of messages exchanged between the user and that other user, a number of other users shared between the user and that other user, and a similarity of content composed by the user and that other user.

7. A method for recommending interesting content, comprising:

receiving, from a user, a request for recommendations comprising interesting messages;

identifying messages from a neighborhood of individuals associated with the user, each message comprising a Uniform Resource Locator (URL), wherein the neighborhood comprises followees that the user follows and further followees that are followed by the followees;

assigning a composition power to each of the individuals based on a frequency of messages generated by that individual;

assigning a trust power to each of the individuals based on a trustworthiness of that individual;

determining a vote power for each individual by combining the composition power and the trust power;

generating a vote score for each of the URLs by summing the vote power for those individuals associated with the messages that include that URL;

applying a threshold to the vote scores and selecting the URLs that satisfy the threshold as the recommendations; and

presenting the recommendations to the user.

8. The method according to claim 7 , further comprising:

determining the trust power for each individual based on at least one of a number of messages exchanged between the user and that individual, a number of other individuals shared between the user and that individual, and a similarity of content composed by the user and that individual.

9. The method according to claim 7 , further comprising:

assigning a high composition power to those individuals that generate a low number of messages and assigning a low composition power to those individuals that generate a high number of messages.

10. A method for identifying interesting content, comprising:

generating a set of candidate Uniform Resource Locators (URLs) selected from messages within one or more information streams;

determining a relevance score for each of the candidate URLs based on a set of messages generated by individuals associated with a user comprising followees that the user follows and further followees that are followed by the followees;

determining a vote score for each of the candidate URLs based on two or more of the individuals associated with the user, comprising:

identifying those messages comprising one such candidate URL from a neighborhood of the messages by the two or more individuals;

assigning a composition power to each of the two or more individuals based on a frequency of the messages generated by that individual;

assigning a trust power to each of the two or more individuals based on a trustworthiness of that individual;

determining a vote power for each of the two or more individuals by combining the composition power and the trust power; and

generating the vote score for each of the candidate URLs by summing the vote powers for the two or more individuals associated with the messages that include that candidate URL;

combining the relevance scores and the vote scores for each of the candidate URLs as a combined score and ranking the candidate URLs based on the combined scores; and

applying a threshold to the combined scores and selecting the candidate URLs that satisfy the threshold as recommendations.

11. The method according to claim 10 , further comprising:

calculating the relevance score, comprising:

obtaining a content profile comprising a vector of words associated with the user;

obtaining one or more content pointer profiles each comprising a vector of words associated with one of the candidates URLs; and

comparing the content profile with each of the content pointer profiles and determining the relevance score by calculating a similarity between the vector of words associated with the user and the vector of words associated with the URL for that content pointer profile.

12. The method according to claim 11 , further comprising:

generating the content pointer profile for one of the URLs, comprising:

extracting words from the messages within which the URL is located;

selecting one or more of the extracted words for inclusion in the content pointer profile and determining a strength of each selected word; and

placing the selected words and associated strengths along a vector as the content pointer profile.

13. The method according to claim 11 , wherein the content profile comprises one of a self-profile and a network profile.

14. The method according to claim 13 , further comprising:

generating the self-profile, comprising:

extracting words from the messages associated with the user and generating a list of the words;

selecting one or more of the extracted words for inclusion in the self-profile and determining a strength of each selected word; and

placing the selected words and associated strengths along the vector as the self-profile.

15. The method according to claim 13 , further comprising:

generating the network profile, comprising:

identifying profiles of the individuals associated with the user, wherein each profile comprises words and associated strengths for that individual;

applying a threshold to each profile and selecting the words that satisfy the threshold;

combining the selected words from each individual profile and determining a strength for each selected word; and

placing the selected words and associated strengths along a vector as the network profile.

16. The method according to claim 10 , further comprising:

determining the trust power for each of the two or more individuals based on at least one of a number of messages exchanged between the user and one such individual, a number of other individuals shared between the user and the one such individual, and a similarity of content composed by the user and the one such individual.

Assignments (10)