Method and system for enhanced content recommendation

1. A method, implemented on a machine having at least one processor, storage, and a communication platform connected to a network, for recommending, to a user, content items, the method comprising:

obtaining a user profile comprising features of content items that have been viewed by the user;

generating a first set of candidate content items based on the user profile by:

comparing, for each feature, a number of times the user viewed content items associated with the feature to an average number of times content items associated with the feature was viewed by other users, and

selecting one or more content items to be included in the first set of candidate content items based on the comparison;

generating a second set of candidate content items, wherein each of the candidate content items of the second set is similar to the one or more content items, wherein a similarity is computed based on a plurality of activities performed by other users with respect to the candidate content item and the one or more content items, the generating further comprising:

obtaining, for each of the plurality of activities, a candidate activity vector representing the activity of other users with respect to the candidate content item,

obtaining, for each of the plurality of activities, a content activity vector representing the activity of other users with respect to the one or more content items,

generating a first integrated vector based on the candidate activity vectors and a second integrated vector based on the content activity vectors, and

incorporating the candidate content item in the second set based on a criterion associated with the first integrated vector and the second integrated vector;

ranking each of the candidate content items in the first and second sets using a learning model, which is trained with at least one of user feature information, content feature information and user-content cross feature information; and

providing, based on the ranking, the candidate content items in the first and second sets as content recommendations to the user.

2. The method of claim 1 , further comprising estimating a likelihood that the user interacts with each of the candidate content items of the second set by:

aggregating the similarities between the candidate content item and the one or more content items, wherein the one or more content items is a plurality; and

including the candidate content item in the second set of candidate content items if the likelihood exceeds a certain threshold.

3. The method of claim 1 ,

wherein the previous interaction includes at least one of clicking, typing, forwarding, and/or scrolling.

4. The method of claim 1 , wherein the similarity is computed based on at least one of a cosine function, a mutual information algorithm, and a locality sensitive hashing (LSH) algorithm.

5. The method of claim 1 , further comprising:

assigning weights to each of the plurality of activities; and

updating, based on the assigning, the first integrated vector and the second integrated vector, a negative weight being associated with non-clicking interaction activity.

6. The method of claim 1 , wherein the user has not previously viewed any of the candidate content items of the second set.

7. A system, implemented on a machine having at least one processor, storage, and a communication platform connected to a network, for recommending, to a user, content items, the system comprising:

a user-profile based candidate content selection unit configured to:

obtain a user profile comprising features of content items that have been viewed by the user, and

generate a first set of candidate content items based on the user profile by comparing, for each feature, a number of times the user viewed content items associated with the feature to an average number of times content items associated with the feature was viewed by other users and by selecting one or more content items to be included in the first set of candidate content items based on the comparison;

a user-activity based candidate content selection unit configured to:

generate a second set of candidate content items, wherein each of the candidate content items of the second set is similar to the one or more content items, wherein a similarity is computed based on a plurality of activities performed by other users with respect to the candidate content item and the one or more content items, the user-activity based candidate content selection unit being further configured to:

obtain, for each of the plurality of activities, a candidate activity vector representing the activity of other users with respect to the candidate content item,

obtain, for each of the plurality of activities, a content activity vector representing the activity of other users with respect to the one or more content items,

generate a first integrated vector based on the candidate activity vectors and a second integrated vector based on the content activity vectors, and

incorporate the candidate content item in the second set based on a criterion associated with the first integrated vector and the second integrated vector; and

a unified ranking unit configured to

rank each of the candidate content items in the first and second sets using a learning model, which is trained with at least one of user feature information, content feature information and user-content cross feature information, and

provide, based on the ranking, the candidate content items in the first and second sets as content recommendations to the user.

8. The system of claim 7 , wherein the user-activity based candidate content selection unit is configured to estimate a likelihood that the user interacts with each of the candidate content items of the second set by:

aggregating the similarities between the candidate content item and the one or more content items, wherein the one or more content items is a plurality; and

including the candidate content item in the second set of candidate content items if the likelihood exceeds a certain threshold.

9. The system of claim 7 , wherein the previous interaction includes at least one of clicking, typing, forwarding, and/or scrolling.

10. The system of claim 7 , wherein the similarity is computed based on at least one of a cosine function, a mutual information algorithm, and a locality sensitive hashing (LSH) algorithm.

11. A machine-readable tangible and non-transitory medium having information for recommending, to a user, content items, wherein the information, when read by the machine, causes the machine to perform the following:

obtaining a user profile comprising features of content items that have been viewed by the user;

generating a first set of candidate content items based on the user profile by:

comparing, for each feature, a number of times the user viewed content items associated with the feature to an average number of times content items associated with the feature was viewed by other users, and

selecting one or more content items to be included in the first set of candidate content items based on the comparison;

generating a second set of candidate content items, wherein each of the candidate content items of the second set is similar to the one or more content items, wherein a similarity is computed based on a plurality of activities performed by other users with respect to the candidate content item and the one or more content items, the generating further comprising:

obtaining, for each of the plurality of activities, a candidate activity vector representing the activity of other users with respect to the candidate content item,

obtaining, for each of the plurality of activities, a content activity vector representing the activity of other users with respect to the one or more content items,

generating a first integrated vector based on the candidate activity vectors and a second integrated vector based on the content activity vectors, and

incorporating the candidate content item in the second set based on a criterion associated with the first integrated vector and the second integrated vector;

ranking each of the candidate content items in the first and second sets using a learning model, which is trained with at least one of user feature information, content feature information and user-content cross feature information; and

providing, based on the ranking, the candidate content items in the first and second sets as content recommendations to the user.

12. The machine-readable and non-transitory medium of claim 11 , wherein the machine is further caused to estimate a likelihood that the user interacts with each of the candidate content items of the second set by:

aggregating the similarities between the candidate content item and the one or more content items, wherein the one or more content items is a plurality; and

including the candidate content item in the second set of candidate content items if the likelihood exceeds a certain threshold.

13. The machine-readable and non-transitory medium of claim 11 , wherein the previous interaction includes at least one of clicking, typing, forwarding, and/or scrolling.