Automated social message stream population

1. A method performed by a messaging system for automatically populating a stream of messages for a user, the method comprising:

providing, by the messaging system and to a client device, for presentation to and selection by a user of the client device, multiple system-selected accounts, the multiple system-selected accounts being accounts selected by the messaging system, wherein the accounts of the messaging system include first accounts, and the system-selected accounts are selected from among the first accounts;

receiving, by the messaging system, a request for content from the user of the client device, the request specifying two or more user-selected accounts, a user-selected account being an account selected by the user from among the multiple system-selected accounts provided by the messaging system to the client device;

obtaining, for each of the two or more user-selected accounts, a respective feature vector, each respective feature vector representing those first accounts to which the user-selected account is interesting, wherein the respective feature vector is a vector of real numbers;

generating, by the messaging system, a seed vector, comprising summing the respective feature vectors for each of the two or more user-selected accounts;

determining, for each first account, a respective measure of relevance of the first account to the user based on a measure of similarity between a respective feature vector for the first account and the seed vector; and

providing, by the messaging system and to the client device, a response to the request, wherein the response includes messages authored by first accounts in a ranked order according to a ranking score of each message, wherein the ranking score of each message is computed as a function of at least the determined measure of relevance of the first account that authored the message.

2. The method of claim 1 , wherein each of the first accounts is a top account, wherein top accounts are accounts that are top when measured by numbers of followers or when measured by numbers of engagements on messages by the accounts.

3. The method of claim 2 , comprising obtaining, for each of the first accounts, a respective feature vector by:

generating a follow matrix F of dimension N×T in which element (i,j) is a binary value that indicates whether the i-th account is followed by the j-th account, wherein each i-th account is one of N accounts of the messaging system, the messaging system has T top accounts and N>T, and each j-th account is one of T top accounts;

factoring the follow matrix F into a matrix product UΣV, where U is an N×A matrix, Σ is a diagonal A×A matrix, and V is an A×T matrix, wherein A is a predetermined number less than or equal to a total number of features for each top account; and

generating the respective feature vector as a respective row vector of a matrix M=(Σ·V) T corresponding to the top account.

4. The method of claim 3 , wherein generating, for each first account, the respective feature vector defining one or more respective features of the first account further comprises normalizing the row vectors of the matrix M in accordance with a computed variance for values of the row vectors.

5. The method of claim 1 ,

wherein each of the respective feature vectors is a vector of real numbers having as many elements as the number of first accounts; and

wherein the measure of similarity between two feature vectors is computed from a dot product between the vectors.

6. The method of claim 1 , wherein the content is a stream of messages, and wherein providing the messages comprises providing the messages in the ranked order as the stream of messages for display on the client device, wherein higher ranked messages are displayed before lower ranked messages.

7. The method of claim 1 , wherein the ranking score for each message is computed as a function of the determined measure of relevance of the account that authored the message, a measure of engagement with the message, and an age of the message.

8. A system comprising:

one or more computers and one or more storage devices on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations performed by a messaging system for automatically populating a stream of messages for a user, the operations comprising:

providing, by the messaging system and to a client device, for presentation to and selection by a user of the client device, multiple system-selected accounts, the multiple system-selected accounts being accounts selected by the messaging system, wherein the accounts of the messaging system include first accounts, and the system-selected accounts are selected from among the first accounts;

receiving, by the messaging system, a request for content from the user of the client device, the request specifying two or more user-selected accounts, a user-selected account being an account selected by the user from among the multiple system-selected accounts provided by the messaging system to the client device;

obtaining, for each of the two or more user-selected accounts, a respective feature vector, each respective feature vector representing those first accounts to which the user-selected account is interesting, wherein the respective feature vector is a vector of real numbers;

generating, by the messaging system, a seed vector, comprising summing the respective feature vectors for each of the two or more user-selected accounts;

determining, for each first account, a respective measure of relevance of the first account to the user based on a measure of similarity between a respective feature vector for the first account and the seed vector; and

providing, by the messaging system and to the client device, a response to the request, wherein the response includes messages authored by first accounts in a ranked order according to a ranking score of each message, wherein the ranking score of each message is computed as a function of at least the determined measure of relevance of the first account that authored the message.

9. The system of claim 8 , wherein each of the first accounts is a top account, wherein top accounts are accounts that are top when measured by numbers of followers or when measured by numbers of engagements on messages by the accounts.

10. The system of claim 9 , comprising obtaining, for each of the first accounts, a respective feature vector by:

generating a follow matrix F of dimension N×T in which element (i,j) is a binary value that indicates whether the i-th account is followed by the j-th account, wherein each i-th account is one of N accounts of the messaging system, the messaging system has T top accounts and N>T, and each j-th account is one of T top accounts;

factoring the follow matrix F into a matrix product UΣV, where U is an N×A matrix, Σ is a diagonal A×A matrix, and V is an A×T matrix, wherein A is a predetermined number less than or equal to a total number of features for each top account; and

generating the respective feature vector as a respective row vector of a matrix M=(Σ·V) T corresponding to the top account.

11. The system of claim 10 , wherein obtaining, for each first account, the respective feature vector defining one or more respective features of the first account further comprises normalizing the row vectors of the matrix M in accordance with a computed variance for values of the row vectors.

12. The system of claim 8 ,

wherein each of the respective feature vectors is a vector of real numbers having as many elements as the number of first accounts; and

wherein the measure of similarity between two feature vectors is computed from a dot product between the vectors.

13. The system of claim 8 , wherein the content is a stream of messages, and wherein providing the messages comprises providing the messages in the ranked order as the stream of messages for display on the client device, wherein higher ranked messages are displayed before lower ranked messages.

14. The system of claim 8 , wherein the ranking score for each message is computed as a function of the determined measure of relevance of the account that authored the message, a measure of engagement with the message, and an age of the message.

15. One or more non-transitory computer-readable storage media encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising:

providing, by a messaging system and to a client device, for presentation to and selection by a user of the client device, multiple system-selected accounts, the multiple system-selected accounts being accounts selected by the messaging system, wherein the accounts of the messaging system include first accounts, and the system-selected accounts are selected from among the first accounts;

receiving, by the messaging system, a request for content from the user of the client device, the request specifying two or more user-selected accounts, a user-selected account being an account selected by the user from among the multiple system-selected accounts provided by the messaging system to the client device;

obtaining, for each of the two or more user-selected accounts, a respective feature vector, each respective feature vector representing those first accounts to which the user-selected account is interesting, wherein the respective feature vector is a vector of real numbers;

generating, by the messaging system, a seed vector, comprising summing the respective feature vectors for each of the two or more user-selected accounts;

determining, for each first account, a respective measure of relevance of the first account to the user based on a measure of similarity between a respective feature vector for the first account and the seed vector; and

providing, by the messaging system and to the client device, a response to the request, wherein the response includes messages authored by first accounts in a ranked order according to a ranking score of each message, wherein the ranking score of each message is computed as a function of at least the determined measure of relevance of the first account that authored the message.

16. The computer-readable media of claim 15 , wherein each of the first accounts is a top account, wherein top accounts are accounts that are top when measured by numbers of followers or when measured by numbers of engagements on messages by the accounts.

17. The computer-readable media of claim 16 , comprising obtaining, for each of the first accounts, a respective feature vector by:

generating a follow matrix F of dimension N×T in which element (i,j) is a binary value that indicates whether the i-th account is followed by the j-th account, wherein each i-th account is one of N accounts of the messaging system, the messaging system has T top accounts and N>T, and each j-th account is one of T top accounts;

factoring the follow matrix F into a matrix product UΣV, where U is an N×A matrix, Σ is a diagonal A×A matrix, and V is an A×T matrix, wherein A is a predetermined number less than or equal to a total number of features for each top account; and

generating the respective feature vector as a respective row vector of a matrix M=(Σ·V) T corresponding to the top account.

18. The computer-readable media of claim 17 , wherein obtaining, for each first account, the respective feature vector defining one or more respective features of the first account further comprises normalizing the row vectors of the matrix M in accordance with a computed variance for values of the row vectors.

19. The computer-readable media of claim 15 ,

wherein each of the respective feature vectors is a vector of real numbers having as many elements as the number of accounts in the plurality of accounts; and

wherein the measure of similarity between two feature vectors is computed from a dot product between the vectors.

20. The computer-readable media of claim 15 , wherein the content is a stream of messages, and wherein providing the messages comprises providing the messages in the ranked order as the stream of messages for display on the client device, wherein higher ranked messages are displayed before lower ranked messages.

21. The computer-readable media of claim 15 , wherein the ranking score for each message is computed as a function of the determined measure of relevance of the account that authored the message, a measure of engagement with the message, and an age of the message.