Multi-stage filtering for recommended user connections on online social networks

1. A method comprising:

accessing a social graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each of the edges between two of the nodes representing a single degree of separation between them, the nodes comprising:

a first node corresponding to a first user associated with an online social network; and

a plurality of second nodes corresponding to a plurality of second users of the online social network, respectively;

identifying, for the first user, a first set of candidate-users from the plurality of second users, wherein each candidate-user corresponds to a second user within a threshold degree of separation of the first user within the social graph;

filtering, using a first-stage filtering model, the first set of candidate-users to generate a second set of candidate-users, wherein the first-stage filtering model filters the candidate-users based on one or more edge-features;

filtering, using a second-stage filtering model, the second set of candidate-users to generate a third-set of candidate-users, wherein the second-stage filtering model filters the candidate-users based on the one or more edge-features and one or more user-features;

ranking, using a user-engagement model, the third set of candidate-users based on a feature superset, wherein the user-engagement model ranks the candidate-users based on a calculated probability of the first user connecting with the respective candidate-user; and

sending, to a client system of the first user, instructions for generating a suggested-friends interface for display to the first user, wherein the suggested-friends interface comprises one or more candidate-users from the third set of candidate-users having a ranking greater than a threshold rank.

2. The method of claim 1 , wherein the threshold degree of separation is two, and wherein each candidate-user in the first set of candidate-users corresponds to a second node connected by a friend-type edge to an intervening node that it connected by a friend-type edge to the first node.

3. The method of claim 1 , wherein one or more edges of the plurality of edges is an inferred connection, each inferred connection being derived based on one or more of:

contact information imported from an external system associated with one or more users of the online social network;

communication information associated with one or more users of the online social network;

login information associated with one or more users of the online social network; or

social-graph information associated with one or more users of the online social network.

4. The method of claim 1 , wherein one or more of the first-stage filtering model, the second-stage filtering model, or the user-engagement model is a machine-learning model generated based on an analysis of user interactions and connections with prior suggested friends.

5. The method of claim 4 , wherein the machine-learning model is a gradient boosted decision tree model.

6. The method of claim 1 , wherein the one or more edge-features comprise information associated with each particular pair of users, the information comprising one or more of:

a social-graph affinity of each user in the pair of users with respect to the other user;

an age of one or more edges connecting nodes corresponding to each user in the pair of users; or

a level of engagement of each user in the pair of users with respect to the other user.

7. The method of claim 1 , wherein the feature superset comprises:

edge-features and user-features used during the first-stage filtering or the second-stage filtering; and

a plurality of additional features not used during the first-stage filtering or the second-stage filtering.

8. The method of claim 7 , wherein the pair of users includes:

the first user; and

a user corresponding to an intervening node.

9. The method of claim 7 , wherein the pair of users includes:

a user corresponding to an intervening node; and

a candidate-user.

10. The method of claim 1 , wherein the one or more user-features comprise information associated with a particular user, wherein the information comprises:

demographic information associated with the particular user;

an age of the particular user's account on the online social network;

an amount of time since the particular user last accessed the online social network;

a number of friend requests sent by the particular user;

a number of friend requests received by the particular user;

a number of friend requests rejected by the particular user;

a number of pending friend requests associated with the particular user;

a friend request acceptance rate associated with the particular user;

a friend request rejection rate associated with the particular user; or

an average pending time of friend requests associated with the particular user.

11. The method of claim 10 , wherein the second-stage filtering model filters the candidate-users based on one or more user-features by comparing information associated with the first user to information associated with each of the candidate-users in the second set of candidate-users.

12. The method of claim 7 , wherein the pair of users includes:

the first user; and

a candidate-user.

13. The method of claim 1 , wherein the instructions for generating the suggested-friends interface are sent to a native application associated with the online social network on the client system of the first user.

14. The method of claim 1 , wherein the instructions for generating the suggested-friends interface are sent to a browser client on the client system of the first user.

15. The method of claim 1 , further comprising receiving a request from the first user to generate the suggested-friends interface, wherein the instructions for generating the suggested-friends interface are sent responsive to the request.

16. One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

access a social graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each of the edges between two of the nodes representing a single degree of separation between them, the nodes comprising:

a first node corresponding to a first user associated with an online social network; and

a plurality of second nodes corresponding to a plurality of second users of the online social network, respectively;

identify, for the first user, a first set of candidate-users from the plurality of second users, wherein each candidate-user corresponds to a second user within a threshold degree of separation of the first user within the social graph;

filter, using a first-stage filtering model, the first set of candidate-users to generate a second set of candidate-users, wherein the first-stage filtering model filters the candidate-users based on one or more edge-features;

filter, using a second-stage filtering model, the second set of candidate-users to generate a third-set of candidate-users, wherein the second-stage filtering model filters the candidate-users based on the one or more edge-features and one or more user-features;

rank, using a user-engagement model, the third set of candidate-users based on a feature superset, wherein the user-engagement model ranks the candidate-users based on a calculated probability of the first user connecting with the respective candidate-user; and

send, to a client system of the first user, instructions for generating a suggested-friends interface for display to the first user, wherein the suggested-friends interface comprises one or more candidate-users from the third set of candidate-users having a ranking greater than a threshold rank.

17. A system comprising: one or more processors; and a non-transitory memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:

access a social graph comprising a plurality of nodes and a plurality of edges connecting the nodes, each of the edges between two of the nodes representing a single degree of separation between them, the nodes comprising:

a first node corresponding to a first user associated with an online social network; and

a plurality of second nodes corresponding to a plurality of second users of the online social network, respectively;

identify, for the first user, a first set of candidate-users from the plurality of second users, wherein each candidate-user corresponds to a second user within a threshold degree of separation of the first user within the social graph;

filter, using a first-stage filtering model, the first set of candidate-users to generate a second set of candidate-users, wherein the first-stage filtering model filters the candidate-users based on one or more edge-features;

filter, using a second-stage filtering model, the second set of candidate-users to generate a third-set of candidate-users, wherein the second-stage filtering model filters the candidate-users based on the one or more edge-features and one or more user-features;

rank, using a user-engagement model, the third set of candidate-users based on a feature superset, wherein the user-engagement model ranks the candidate-users based on a calculated probability of the first user connecting with the respective candidate-user; and

send, to a client system of the first user, instructions for generating a suggested-friends interface for display to the first user, wherein the suggested-friends interface comprises one or more candidate-users from the third set of candidate-users having a ranking greater than a threshold rank.