Automatic determination of genre-specific relevance of recommendations in a social network

• US 10,404,824 B2
• Filed: 07/13/2012
• Issued: 09/03/2019
• Est. Priority Date: 07/22/2011
• Status: Active Grant

• Current Assignee: Funke TV Guide GmbH
• Original Assignee: Funke Digital Tv Guide GmbH (Axel Springer SE)
• Inventors: Korst, Jan, Barbieri, Mauro, Pronk, Serverius Petrus Paulus
• Primary Examiner(s): Trost, IV, William G
• Assistant Examiner(s): Moorad, Imran
• Application Number:
  US14/131,510
• Publication Number:
  US 20140136621A1
• Time in Patent Office: 2,608 Days
• Field of Search: 709204, 715700
• CPC Class Codes: G06Q 30/0631 Item recommendations
• CPC Class Codes: G06Q 50/01 Social networking
• CPC Class Codes: H04L 67/306 User profiles

Abstract

The present invention relates to an operating method of operating a recommender system, a filtering apparatus (260) for a recommender system (200), a recommender system and a corresponding computer program. An idea of the invention is to automatically learn for a user A in a social network, which recommendations of contacts of user A, who are also members of the social network, are relevant with respect to a genre into which user A is interested in. A learning algorithm is used to interpret feedback from user A in response to receiving recommendations from his/her contacts. Thereby, for each combination of a contact and a genre, a relevance-taste index can be determined. The determined relevance-taste index is subjected to a filter. Only such recommendations are provided to user A, whose associated relevance-taste indices fulfill a filtering criterion. Thereby, the amount of irrelevant recommendations submitted to user A can be significantly reduced.

Claims

1. A method of operating a personal recommender system arranged for being coupled to a computer implemented social network, the computer implemented social network being a computer network that allows participation of a plurality of users, wherein each user can set up a personal list of contacts, wherein each contact of such list is also a participant of the same or another social network, wherein the method is carried out automatically by a machine and comprises the steps of:

a) detecting that a recommendation of a contact-genre tuple (CU1,g) has been filed in a user account of a user A of the social network, wherein the recommendation of the contact-genre tuple (CU1,g) relates to a content item x of a specific genre (g=g(x)) and has been initiated through a contact account of a contact CU1 of user A,

b) monitoring a recommendation related reaction behaviour of user A in response to receiving the recommendation filed in the user account,

c) repeating steps a) to b) for a plurality of recommendations of the same contact-genre tuple (CU1,g) and logging a plurality of monitored reaction behaviours for determining a user A related relevance-taste index r=r((CU1,g)) associated to the contact-genre tuple (CU1, g) in dependence of the plurality of monitored reaction behaviours, and

d) filtering a current recommendation of the same contact-genre tuple (CU1,g) sent to the user A by filing the current recommendation in the user account only when a filtering criterion is fulfilled by the relevance-taste index of the current recommendation and blocking the current recommendation when the filtering criterion is not fulfilled by the relevance-taste index, presenting filtered recommendations to user A through a display of a user interface coupled to the personal recommender system and storing blocked recommendations in a memory comprised by the recommender system,

wherein the method further comprises;

buffering current recommendations that have been sent to the user account but have been blocked and therefore not been provided to user A,

filing at least one of the buffered recommendations in the user account, if the at least one recommendation fulfils an adjusted filtering criterion,

grouping blocked recommendations being related to a common content item y,

determining an accumulated relevance-taste index by summing relevance-taste indices associated to each of the grouped blocked recommendations,

filing the grouped blocked recommendations as a single combined recommendation in the user account, if the accumulated relevance-taste index fulfils the filtering criterion,

identifying senders F(A) of the grouped blocked recommendations, wherein each of the senders F(A) has an associated contact account that is linked to the user account of user A,

for each of the identified senders F(A), determining a like-degree λ

(B,y) for the common content item y, the like-degree λ

(B,y) indicating the respective sender'"'"'s B interest or disinterest in the common content item y,

calculating a normalized accumulated relevance-taste index in dependence of the determined like-degrees and the accumulated relevance-taste index, wherein the normalized accumulated relevance-taste index corresponds to an assumed like-degree λ

(A,y) indicating user A'"'"'s interest or disinterest in the common content item y, and

filing a recommendation for the common content item y in the user account, if the normalized accumulated relevance-taste index fulfils the filtering criterion.

2. The method of claim 1, wherein the normalized accumulated relevance-taste index λ

(A, y) is calculated according to the formula

λ

(A,y)=Σ

B∈

F(A)r(B,g(y))·

λ

(B,y)/Σ

B∈

F(A)|r(B,g(y))|,

wherein Σ

B∈

F(A)r(B, g(y)) is the accumulated relevance-taste index.

3. The method of claim 2, additionally comprising the step of:

presenting filtered recommendations to user A through a display of a user interface coupled to the personal recommender system by employing a colour encoding for encoding a respective assumed like-degree of each of the presented recommendations.

4. The method of claim 1, wherein the step of determining the relevance-taste index includes applying a learning algorithm, such as applying a neural network, a Naive Bayes classifier, or a support vector machine.

5. The method of claim 4, wherein applying the learning algorithm includes taking into account a warning relating to a content item, which has been submitted through the contact account, the warning indicating a contact'"'"'s dislike with respect to the content item.

6. The method of claim 1, comprising the additional steps of:

grouping blocked recommendations being related to a common content item y,

determining an accumulated relevance-taste index by summing relevance-taste indices associated to each of the grouped, blocked recommendations, and

filing the grouped, blocked recommendations as a single combined recommendation in the user account, if the accumulated relevance-taste index fulfils the filtering criterion.

7. The method of claim 1, further comprising the step of:

e) generating a colour overlay for displaying the filtered recommendations in order to indicate a respective assumed like-degree λ

(A, x), wherein a degree of colour saturation used to overlay a given content item is a function of λ

(A, x).

8. A non-transitory storage medium storing a computer program for operating a recommender system, the computer program comprising program code means for causing the recommender system to carry out the steps of claim 1, when the computer program is run on a computer controlling the recommender system.

9. A method of operating a personal recommender system arranged for being coupled to a computer implemented social network, the computer implemented social network being a computer network that allows participation of a plurality of users, wherein each user can set up a personal list of contacts, wherein each contact of such list is also a participant of the same or another social network, wherein the method is carried out automatically by a machine and comprises the steps of:

a) detecting that a recommendation of a contact-genre tuple (CU1,g) has been filed in a user account of a user A of the social network, wherein the recommendation of the contact-genre tuple (CU1,g) relates to a content item x of a specific genre (g=g(x)) and has been initiated through a contact account of a contact CU1 of user A,

b) monitoring a recommendation related reaction behaviour of user A in response to receiving the recommendation filed in the user account,

c) repeating steps a) to b) for a plurality of recommendations of the same contact-genre tuple (CU1,g) and logging a plurality of monitored reaction behaviours for determining a user A related relevance-taste index r=r((CU1,g)) associated to the contact-genre tuple (CU1, g) in dependence of the plurality of monitored reaction behaviours, and

d) filtering a current recommendation of the same contact-genre tuple (CU1,g) sent to the user A by filing the current recommendation in the user account only when a filtering criterion is fulfilled by the relevance-taste index of the current recommendation and by blocking the current recommendation when the filtering criterion is not fulfilled, presenting filtered recommendations to user A through a display of a user interface coupled to the personal recommender system and storing blocked recommendations in a memory comprised by the recommender system;

e) grouping blocked recommendations being related to a common content item y,

f) determining an accumulated relevance-taste index by summing relevance-taste indices associated to each of the grouped blocked recommendations,

g) filing the grouped blocked recommendations as a single combined recommendation in the user account, if the accumulated relevance-taste index fulfils the filtering criterion,

h) identifying senders F(A) of the grouped blocked recommendations, wherein each of the senders F(A) has an associated contact account that is linked to the user account of user A,

i) for each of the identified senders F(A), determining a like-degree λ

(B,y) for the common content item y, the like-degree λ

(B,y) indicating the respective sender'"'"'s B interest or disinterest in the common content item y,

j) calculating a normalized accumulated relevance-taste index in dependence of the determined like-degrees and the accumulated relevance-taste index, wherein the normalized accumulated relevance-taste index corresponds to an assumed like-degree λ

(A,y) indicating user A'"'"'s interest or disinterest in the common content item y, and

k) filing a recommendation for the common content item y in the user account, if the normalized accumulated relevance-taste index fulfils the filtering criterion.

10. The method of claim 9, wherein the normalized accumulated relevance-taste index λ

(A, y) is calculated according to the formula

λ

(A,y)=Σ

B∈

F(A)r(B,g(y))·

λ

(B,y)/Σ

B∈

F(A)|r(B,g(y))|,

wherein Σ

B∈

F(A)r(B,g(y)) is the accumulated relevance-taste index.

11. The method of claim 10, further comprising:

presenting filtered recommendations to user A through a display of a user interface coupled to the personal recommender system by employing a colour encoding for encoding a respective assumed like-degree of each of the presented recommendations.

12. The method of claim 9, further comprising:

presenting filtered recommendations to user A through a display of a user interface coupled to the personal recommender system by employing a colour encoding for encoding a respective assumed like-degree of each of the presented recommendations.

13. The method of claim 9, wherein the step of determining the relevance-taste index includes applying a learning algorithm, such as applying a neural network, a Naive Bayes classifier, or a support vector machine.

14. The method of claim 13, wherein applying the learning algorithm includes taking into account a warning relating to a content item, which has been submitted through the contact account, the warning indicating a contact'"'"'s dislike with respect to the content item.

15. The method of claim 9, further comprising the step of:

generating a colour overlay for displaying the filtered recommendations in order to indicate a respective assumed like-degree λ

(A, x), wherein a degree of colour saturation used to overlay a given content item is a function of λ

(A, x).

16. A non-transitory storage medium storing a computer program for operating a recommender system, the computer program comprising program code means for causing the recommender system to carry out the steps of claim 9, when the computer program is run on a computer controlling the recommender system.