Collecting training data using session-level randomization in an on-line social network
A news feed system of an on-line social network system news utilizes a relevance model to determine which updates from an inventory of updates are to be presented to a member on their news feed page. The relevance model is trained using historical data that reflects interactions of members of the on-line social network system with items in their respective news feed pages. In order to reduce potential biases in the historical data that is used to train the relevance model, the news feed system designates a certain portion of all member sessions to be random sessions. The news feed generated for a member during a random session includes updates that are selected and/or ordered for presentation using one or more randomization techniques.
1. A computer-implemented method comprising:
detecting, at a server system, a request to commence a login session for a member represented by a member profile in an on-line social network system;
generating, randomly with a predetermined probability, a session mode value;
invoking a relevance model to generate respective scores for items in an inventory of updates identified as potentially of interest to the member, the respective relevance scores utilized to determine ordering of updates from the inventory of updates on a news feed web page generated for the member;
using at least one processor, randomizing the order of presentation of items to be included in the news feed web page based on the generated session mode value indicating that the requested login session is to be a random session, the news feed page to be displayed on a display device of the member; and
training the relevance model using data representing interactions of the member with items presented in the news feed during the random session.
2. The method of claim 1 , comprising including a predetermined number of top-scored items from the inventory of updates into a presentation set of items, wherein the randomizing of the presentation of items to be included in the news feed web page comprises including items from the presentation set of items in the news feed web page in a random order.
3. The method of claim 1 , wherein the randomizing of the presentation of items to be included in the news feed web page comprises perturbing the generated scores by a random factor and using the resulting perturbed scores to select the items to be included in the news feed web page.
4. The method of claim 1 , wherein the predetermined probability used by the session mode selector to generate a session mode value indicating that the requested login session is to be a random session is a default probability.
5. The method of claim 4 , comprising:
detecting a further request to commence a further login session for a further member represented by a further member profile in the on-line social network system; and
based on a session requests history feature associated with the further member profile; utilizing probability other than the default probability to generate a session mode value for the further login session.
6. The method of claim 5 , comprising utilizing probability that is less than the default probability.
7. The method of claim 5 , wherein the session requests history feature reflects how recently a login session initiated for the further member was selected to be a random session.
8. The method of claim 5 , wherein the session requests history feature reflects how frequently login sessions are initiated for the further member.
9. The method of claim 5 , comprising training the relevance model using weighted data representing interactions of the member with items presented in the news teed during the random session.
10. The method of claim 1 , comprising:
constructing a news feed web page that includes the subset of items from the inventory; and
causing presentation of the news feed web page on a display device of the member.
11. A computer-implemented system comprising:
one or more processors; and
a non-transitory computer readable storage medium comprising instructions that when executed by the one or processors cause the one or more processors to perform operations comprising:
detecting, at a server system, a request to commence a login session for a member represented by a member profile in an on-line social network system;
generating, randomly with a predetermined probability, a session mode value;
invoking a relevance model to generate respective scores for items in an inventory of updates identified as potentially of interest to the member, the respective relevance scores utilized to determine ordering of updates from the inventory of updates on a news feed web page generated for the member;
randomizing the order of presentation of items to be included in the news feed web page based on the generated session mode value indicating that the requested login session is to be a random session, the news feed page to be displayed on a display device of the member; and
training the relevance model using data representing interactions of the member with items presented in the news feed during the random session.
12. The system of claim 11 , comprising including a predetermined number of top-scored items from the inventory of updates into a presentation set of items, wherein the randomizing of the presentation of items to be included in the news feed web page comprises including items from the presentation set of items in the news feed web page in a random order.
13. The system of claim 11 , comprising perturbing the generated scores by a random factor and using the resulting perturbed scores to select the items to be included in the news feed web page.
14. The system of claim 11 , wherein the predetermined probability is a default probability.
15. The system of claim 14 , comprising:
the request detector is to detect a further request to commence commencing a further login session for a further member represented by a further member profile in the on-line social network system; and
utilizing probability other than the default probability to generate a session mode value for the further login session, based on a session requests history feature associated with the further member profile.
16. The system of claim 15 , comprising utilizing probability that is less than the default probability.
17. The system of claim 15 , wherein the session requests history feature reflects how recently a login session initiated for the further member was selected to be a random session.
18. The system of claim 15 , wherein the session requests history feature reflects how frequently login sessions are initiated for the further member.
19. The system of claim 15 , wherein the training of the relevance model comprises using weighted data representing interactions of the member with items presented in the news feed during the random session.
20. A machine-readable non-transitory storage medium having instruction data executable by a machine to cause the machine to perform operations comprising:
detecting, at a server system, a request to commence a login session for a member represented by a member profile in an on-line social network system;
generating randomly with a predetermined probability, a session mode value;
invoking a relevance model to generate respective scores for items in an inventory of updates identified as potentially of interest to the member, the respective relevance scores utilized to determine ordering of updates from the inventory of updates on a news feed web page generated for the member;
randomizing the order of presentation of items to be included in the news feed web page based on the generated session mode value indicating that the requested login session is to be a random session, the news feed page to be displayed on a display device of the member; and
training the relevance model using data representing interactions of the member with items presented in the news feed during the random session.