Abstract: The disclosure is directed to determining a social signature of a user of an application, e.g., a social networking application. The social signature can be a representation of social properties of a person, and can be determined based on various factors, e.g., user profile of the user and friends of the user. The social signature can be used for various applications, e.g., determining socially similarity of a user pair and determining if a user is a spammer. The social signature is determined as a vector of k buckets in which each of the k buckets indicates a count of friends of the user assigned to the corresponding bucket. A social similarity of a user pair is determined as a function of the social signatures of the user pair. A user is determined as likely to be a spammer if the social signature of the user is of a specified pattern.

Abstract: Embodiments are disclosed for improving cache or memory efficiency of a social network system. A method according to some embodiments includes steps of: receiving an instruction to improve cache or memory efficiency of social graph data of a social graph; generating based on the social graph a partitioning tree including multiple bottom-level buckets, the partitioning tree dividing the vertices of the social graph into the bottom-level buckets and ordering the bottom-level buckets such that a social network metric regarding the vertices is optimized; assigning user IDs to the vertices of the social network in a numerical sequence based on the ordering of the bottom-level buckets; storing the social graph data of the users in storage locations in an order according to the numeral sequence of the assigned user IDs of the vertices.

Abstract: A technology for routing traffic from similar users to a same server cluster to improve data center efficiency is disclosed. When a traffic routing server receives a request from a user, the traffic routing server determines an identifier of a partition to which the user is assigned. The user and many other users with whom the user shares a social attribute are co-located in the same partition. The traffic routing server then computes a hash of the identifier using a hash function and locates a server cluster on a consistent hash ring using the computed hash. The traffic routing server then sends the request from the user to that server cluster. By consistently sending requests from users assigned to the same partition to the same server cluster, the technology improves cache hit rates and reduces data duplication across the server clusters, which in turn improves datacenter efficiency.

Abstract: Embodiments are disclosed for improving cache or memory efficiency of a social network system. A method according to some embodiments includes steps of: receiving an instruction to improve cache or memory efficiency of social graph data of a social graph; generating based on the social graph a partitioning tree including multiple bottom-level buckets, the partitioning tree dividing the vertices of the social graph into the bottom-level buckets and ordering the bottom-level buckets such that a social network metric regarding the vertices is optimized; assigning user IDs to the vertices of the social network in a numerical sequence based on the ordering of the bottom-level buckets; storing the social graph data of the users in storage locations in an order according to the numeral sequence of the assigned user IDs of the vertices.

Abstract: Online systems generate predictors for predicting actions of users of the online system. The online system receives requests to generate predictor models for predicting whether a user is likely to take an action of a particular action type. The request specifies the type of action and criteria for identifying a successful instance of the action type and a failure instance of the action type. The online system collects data including successful and failure instances of the action type. The online system generates one or more predictors of different types using the generated data. The online system evaluates and compares the performance of the different predictors generated and selects a predictor based on the performance. The online system returns a handle to access the generated predictor to the requester of the predictor.

Abstract: A technology for routing traffic from similar users to a same server cluster to improve data center efficiency is disclosed. When a traffic routing server receives a request from a user, the traffic routing server determines an identifier of a partition to which the user is assigned. The user and many other users with whom the user shares a social attribute are co-located in the same partition. The traffic routing server then computes a hash of the identifier using a hash function and locates a server cluster on a consistent hash ring using the computed hash. The traffic routing server then sends the request from the user to that server cluster. By consistently sending requests from users assigned to the same partition to the same server cluster, the technology improves cache hit rates and reduces data duplication across the server clusters, which in turn improves datacenter efficiency.

Abstract: Online systems generate predictors for predicting actions of users of the online system. The online system receives requests to generate predictor models for predicting whether a user is likely to take an action of a particular action type. The request specifies the type of action and criteria for identifying a successful instance of the action type and a failure instance of the action type. The online system collects data including successful and failure instances of the action type. The online system generates one or more predictors of different types using the generated data. The online system evaluates and compares the performance of the different predictors generated and selects a predictor based on the performance. The online system returns a handle to access the generated predictor to the requester of the predictor.

Abstract: Members of a social network user's social graph are automatically segregated into overlapping clusters according to patterns of their past communications. Each cluster within the social graph represents a group of members having a high degree of intra-cluster communication or other connection with one another. The clustering is performed according to a sorting or ranking in accordance with non-principal eigenvectors of connectivity matrices describing the intra-cluster communications/connections. The overlapping clusters exhibit maximum internal density and minimum external sparsity.

Abstract: Members of a social network user's social graph are automatically segregated into overlapping clusters according to patterns of their past communications. Each cluster within the social graph represents a group of members having a high degree of intra-cluster communication or other connection with one another. The clustering is performed according to a sorting or ranking in accordance with non-principal eigenvectors of connectivity matrices describing the intra-cluster communications/connections. The overlapping clusters exhibit maximum internal density and minimum external sparsity.

Abstract: Content items in a distributed system are defined by a respective key, and each such content item is copied to R1 computer systems of the distributed system which have unique identifiers closest to a value of the respective key, where R1 is less than R2 which is less than R3, at least R2 number of the computer systems have copies of any respective one of the content items for all of the content items, and none of the computer systems farther than R3 in an address space of the distributed system have a copy of the subject content item. Modifications of individual content items are synchronized across all instances thereof responsive to a put operation and/or at periodic intervals.

Abstract: Content items in a distributed system are defined by a respective key, and each such content item is copied to R1 computer systems of the distributed system which have unique identifiers closest to a value of the respective key, where R1 is less than R2 which is less than R3, at least R2 number of the computer systems have copies of any respective one of the content items for all of the content items, and none of the computer systems farther than R3 in an address space of the distributed system have a copy of the subject content item. Modifications of individual content items are synchronized across all instances thereof responsive to a put operation and/or at periodic intervals.