Recommending edges via importance aware machine learned model
In an example embodiment, a single machine learned model that allows for ranking of entities across all of the different combinations of node types and edge types is provided. The solution calibrates the scores from Edge-FPR models to a single scale. Additionally, the solution may utilize a per-edge type multiplicative factor dictated by the true importance of an edge type, which is learned through a counterfactual experimentation process. The solution may additionally optimize on a single, common downstream metric, specifically downstream interactions that can be compared against each other across all combinations of node types and edge types.
1. A system comprising:
a non-transitory computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to perform operations comprising:
training a first cohort-specific machine learned pre diction model corresponding to a first cohort by specifying a first optimization parameter and then feeding first training data and the first optimization parameter into a first machine learning algorithm, the first training data including sample entities of a first type of entity in an online network and sample entries of a first type of connection between entities in the online network;
receiving an identification of a first user in an online network;
obtaining, from the first cohort-specific machine learned prediction model, first scores for a plurality of different entities of the first cohort based on the identification of the first user, the first scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the first cohort via a graphical user interface of the online network, wherein the first cohort is a first combination of the first type of entity in an online network and the first type of connection between entities in the online network, the first cohort-specific machine learned prediction model calculating first scores on a first scale;
calibrating the first scores onto a common scale utilized by multiple cohort-specific machine learned prediction models;
obtaining, from a second cohort-specific machine learned prediction model corresponding to a second cohort based on the identification of the first user, second scores for a plurality of different entities of the second cohort, the second scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the second cohort via a graphical user interface of the online network, wherein the second cohort is a second combination of the first type of entity in the online network and a second type of connection between entities in the online network, the second cohort-specific machine learned prediction model calculating the second scores on a second scale, wherein the second cohort-specific machine learned prediction model is trained separately from the first cohort-specific machine learned prediction model;
calibrating the second scores onto the common scale;
feeding the identification of the first user, the calibrated first scores, and the calibrated second scores into a single calibrated importance-aware machine learned model trained to predict a probability of downstream interaction by users with top-k entities within each of a plurality of cohorts to obtain a ranking of the plurality of cohorts for the first user; and
causing display, in the graphical user interface, to the first user, of one or more entities of the first cohort and one or more entities of the second cohort in a manner that entities of a higher ranking cohort are displayed above entities of a lower ranking cohort.
2. The system of claim 1 , wherein the first cohort-specific machine learned prediction model is a neural network.
3. The system of claim 1 , wherein the single calibrated importance-aware machine learned model converts entity-level scores to cohort-level scores using an aggregation function.
4. The system of claim 3 , wherein the aggregation function is a weighted average.
5. The system of claim 3 , wherein the aggregation function is a non-linear aggregation function.
6. The system of claim 3 , wherein each cohort-level score is multiplied by a different importance factor unique to the corresponding cohort, wherein the importance factor for each cohort is estimated by dropping edges corresponding to the corresponding cohort from member heterogonous graphs of entities in training data and measuring impact to the downstream interaction of members.
7. The system of claim 1 , wherein the calibrating is performed by machine learned function mapping from quantiles of scores from cohort-specific machine learned models to a quantized observed response.
8. The system of claim 7 , wherein the quantized observed response is a click, like, comment, or share action in a graphical user interface.
9. The system of claim 1 , wherein the operations further comprise:
training the second cohort-specific machine learned prediction model by specificizing a second optimization parameter and then feeding second training data and the second optimization parameter into a second machine learning algorithm different than the first machine learning algorithm, the second training data including sample entities of the first type of entity and sample entries of the second type of connection.
10. The system of claim 1 , wherein the operations further comprise:
dynamically adjusting how the one or more entities of the first cohort and the one or more entities of the second cohort are displayed in the graphical user interface in response to a second execution of the obtaining from a second cohort-specific machine learned prediction model corresponding to a second cohort, scores for a plurality of different entities of the second cohort.
11. A computerized method comprising:
training a first cohort-specific machine learned pre diction model corresponding to a first cohort by specifying a first optimization parameter and then feeding first training data and the first optimization parameter into a first machine learning algorithm, the first training data including sample entities of a first type of entity in an online network and sample entries of a first type of connection between entities in the online network;
receiving an identification of a first user in an online network;
obtaining, from the first cohort-specific machine learned prediction model, first scores for a plurality of different entities of the first cohort based on the identification of the first user, the first scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the first cohort via a graphical user interface of the online network, wherein the first cohort is a first combination of the first type of entity in an online network and the first type of connection between entities in the online network, the first cohort-specific machine learned prediction model calculating first scores on a first scale;
calibrating the first scores onto a common scale utilized by multiple cohort-specific machine learned prediction models;
obtaining, from a second cohort-specific machine learned prediction model corresponding to a second cohort based on the identification of the first user, second scores for a plurality of different entities of the second cohort, the second scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the second cohort via a graphical user interface of the online network, wherein the second cohort is a second combination of the first type of entity in the online network and a second type of connection between entities in the online network, the second cohort-specific machine learned prediction model calculating the second scores on a second scale, wherein the second cohort-specific machine learned prediction model is trained separately from the first cohort-specific machine learned prediction model;
calibrating the second scores onto the common scale;
feeding the identification of the first user, the calibrated first scores, and the calibrated second scores into a single calibrated importance-aware machine learned model trained to predict a probability of downstream interaction by users with top-k entities within each of a plurality of cohorts to obtain a ranking of the plurality of cohorts for the first user; and
causing display, in the graphical user interface, to the first user, of one or more entities of the first cohort and one or more entities of the second cohort in a manner that entities of a higher ranking cohort are displayed above entities of a lower ranking cohort.
12. The method of claim 11 , wherein the first cohort-specific machine learned prediction model is a neural network.
13. The method of claim 11 , wherein the single calibrated importance-aware machine learned model converts entity-level scores to cohort-level scores using an aggregation function.
14. The method of claim 13 , wherein the aggregation function is a weighted average.
15. The method of claim 13 , wherein the aggregation function is a non-linear aggregation function.
16. The method of claim 13 , wherein each cohort-level score is multiplied by a different importance factor unique to the corresponding cohort, wherein the importance factor for each cohort is estimated by dropping edges corresponding to the corresponding cohort from member heterogonous graphs of entities in training data and measuring impact to the downstream interaction of members.
17. The method of claim 11 , wherein the calibrating is performed by machine learned function mapping from quantiles of scores from cohort-specific machine learned models to a quantized observed response.
18. The method of claim 17 , wherein the quantized observed response is a click, like, comment, or share action in a graphical user interface.
19. A non-transitory machine-readable storage medium comprising instructions which, when implemented by one or more machines, cause the one or more machines to perform operations comprising:
training a first cohort-specific machine learned pre diction model corresponding to a first cohort by specifying a first optimization parameter and then feeding first training data and the first optimization parameter into a first machine learning algorithm, the first training data including sample entities of a first type of entity in an online network and sample entries of a first type of connection between entities in the online network;
receiving an identification of a first user in an online network;
obtaining, from the first cohort-specific machine learned prediction model, first scores for a plurality of different entities of the first cohort based on the identification of the first user, the first scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the first cohort via a graphical user interface of the online network, wherein the first cohort is a first combination of the first type of entity in an online network and the first type of connection between entities in the online network, the first cohort-specific machine learned prediction model calculating first scores on a first scale;
calibrating the first scores onto a common scale utilized by multiple cohort-specific machine learned prediction models;
obtaining, from a second cohort-specific machine learned prediction model corresponding to a second cohort based on the identification of the first user, second scores for a plurality of different entities of the second cohort, the second scores indicative of a likelihood of the first user interacting with each of the plurality of different entities of the second cohort via a graphical user interface of the online network, wherein the second cohort is a second combination of the first type of entity in the online network and a second type of connection between entities in the online network, the second cohort-specific machine learned prediction model calculating the second scores on a second scale, wherein the second cohort-specific machine learned prediction model is trained separately from the first cohort-specific machine learned prediction model;
calibrating the second scores onto the common scale;
feeding the identification of the first user, the calibrated first scores, and the calibrated second scores into a single calibrated importance-aware machine learned model trained to predict a probability of downstream interaction by users with top-k entities within each of a plurality of cohorts to obtain a ranking of the plurality of cohorts for the first user; and
causing display, in the graphical user interface, to the first user, of one or more entities of the first cohort and one or more entities of the second cohort in a manner that entities of a higher ranking cohort are displayed above entities of a lower ranking cohort.
20. The non-transitory machine-readable storage medium of claim 19 , wherein the first cohort-specific machine learned prediction model is a neural network.
21. The non-transitory machine-readable storage medium of claim 19 , wherein the single calibrated importance-aware machine learned model converts entity-level scores to cohort-level scores using an aggregation function.
22. The non-transitory machine-readable storage medium of claim 21 , wherein the aggregation function is a weighted average.