Generating estimated trait-intersection counts utilizing semantic-trait embeddings and machine learning

1. In a digital medium environment for managing and analyzing digital data repositories, a computer-implemented method for efficiently generating trait-intersection counts from the digital data repositories, comprising;

maintaining one or more databases comprising a plurality of digital user profiles and a plurality of traits;

performing a step for utilizing semantic-trait-training embeddings and initial trait-intersection-training counts of users corresponding to training traits to train a trait-intersection-count-machine-learning model;

identifying a request for a trait-intersection count of users corresponding to a first trait from the plurality of traits and a second trait from the plurality of traits for a target time period; and

performing a step for generating an estimated trait-intersection count of users corresponding to the first trait and the second trait for the target time period utilizing the trained trait-intersection-count-machine-learning model.

2. The method of claim 1 , wherein the trait-intersection-count-machine-learning model comprises a decision-tree regressor or an artificial neural network.

3. The method of claim 1 , further comprising:

receiving the request for the trait-intersection count for the target time period from a client device; and

providing the estimated trait-intersection count for the target time period to the client device for display within a graphical user interface.

4. The method of claim 1 , further comprising:

determining an observed trait-intersection count of users corresponding to the first trait and the second trait for the target time period;

determining that a count difference between the estimated trait-intersection count for the target time period and the observed trait-intersection count for the target time period satisfies a difference threshold; and

identifying a count anomaly for the first trait or the second trait based on the count difference.

5. A non-transitory computer readable medium storing instructions thereon that, when executed by at least one processor, cause a computer system to:

identify a request for a trait-intersection count of digital entities corresponding to a first trait and a second trait for a target time period;

identify a semantic-trait embedding of the first trait and the second trait;

determine an initial trait-intersection count of digital entities corresponding to the first trait and the second trait for an initial time period; and

based on the request, generate an estimated trait-intersection count of digital entities corresponding to the first trait and the second trait for the target time period by analyzing the semantic-trait embedding and the initial trait-intersection count for the initial time period utilizing a machine-learning model trained to generate estimated trait-intersection counts.

6. The non-transitory computer readable medium of claim 5 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:

generate the semantic-trait embedding by generating a first semantic-embedding vector for the first trait and a second semantic-embedding vector for the second trait utilizing a semantic-word-vector model; and

generate a trait-intersection-count vector for the initial trait-intersection count by applying network layers of the machine-learning model to the initial trait-intersection count.

7. The non-transitory computer readable medium of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the computer system to generate the estimated trait-intersection count for the target time period utilizing the machine-learning model by:

generating a first modified semantic-embedding vector for the first trait and a second modified semantic-embedding vector for the second trait by applying a trait-embedding matrix of the machine-learning model to the first semantic-embedding vector and the second semantic-embedding vector; and

combining the first modified semantic-embedding vector, the second modified semantic-embedding vector, and the trait-intersection-count vector to create a combined-trait-intersection vector.

8. The non-transitory computer readable medium of claim 7 , further comprising instructions that, when executed by the at least one processor, cause the computer system to generate the estimated trait-intersection count for the target time period utilizing the machine-learning model by applying additional network layers of the machine-learning model to the combined-trait-intersection vector to determine the estimated trait-intersection count for the target time period.

9. The non-transitory computer readable medium of claim 5 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:

identify a time-segment indicator corresponding to the initial time period for the initial trait-intersection count; and

generate the estimated trait-intersection count for the target time period in part by analyzing the time-segment indicator utilizing the machine-learning model.

10. The non-transitory computer readable medium of claim 5 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:

determine, for a plurality of time periods, a rolling average trait-intersection count of digital entities corresponding to the first trait and the second trait or a rolling extremum trait-intersection count of digital entities corresponding to the first trait and the second trait; and

generate the estimated trait-intersection count for the target time period in part by analyzing the rolling average trait-intersection count or the rolling extremum trait-intersection count utilizing the machine-learning model.

11. The non-transitory computer readable medium of claim 5 , wherein the trait-intersection count of digital entities comprises a trait-intersection count of events, products, users, or transactions.

12. The non-transitory computer readable medium of claim 5 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:

identify the request for the trait-intersection count by identifying a segment-trait request for a trait-intersection count of users corresponding to the first trait and a segment for the second trait for the target time period; and

generate the estimated trait-intersection count by generating an estimated trait-intersection count of users corresponding to the first trait and a segment for the second trait for the target time period.

13. The non-transitory computer readable medium of claim 5 , wherein the machine-learning model is trained based on:

semantic-trait-training embeddings for a first training trait and a second training trait;

initial trait-intersection-training counts of digital entities corresponding to the first training trait and the second training trait for initial-training-time periods; and

ground-truth-trait-intersection counts of digital entities corresponding to the first training trait and the second training trait for a target-training-time period.

14. A system comprising:

at least one processor;

at least one non-transitory computer readable medium comprising:

digital training data comprising a plurality of training users and a plurality of training traits of the plurality of training users;

a semantic-word-vector model; and

a machine-learning model; and

instructions that, when executed by at least one processor, cause the system to train the machine-learning model by:

applying the semantic-word-vector model to a first training trait from the plurality of training traits and a second training trait from the plurality of training traits to generate a semantic-trait-training embedding;

identifying an initial trait-intersection-training count of users corresponding to the first training trait and the second training trait for an initial-training-time period;

applying the machine-learning model to the semantic-trait-training embedding and the initial trait-intersection-training count to generate an estimated trait-intersection-training count of users corresponding to the first training trait and the second training trait for a target-training-time period; and

modifying parameters of the machine-learning model based on a comparison of the estimated trait-intersection-training count for the target-training-time period with a ground-truth-trait-intersection count of users corresponding to the first training trait and the second training trait for the target-training-time period.

15. The system of claim 14 , further comprising instructions that, when executed by at least one processor, cause the system to:

generate the semantic-trait-training embedding by generating a first semantic-embedding-training vector for the first training trait and a second semantic-embedding-training vector for the second training trait; and

generate a trait-intersection-count-training vector for the initial trait-intersection-training count by applying network layers of the machine-learning model to the initial trait-intersection-training count.

16. The system of claim 15 , further comprising instructions that, when executed by the at least one processor, cause the system to generate the estimated trait-intersection-training count for the target-training-time period utilizing the machine-learning model by:

generating a first modified semantic-embedding-training vector for the first training trait and a second modified semantic-embedding-training vector for the second training trait by applying a trait-embedding matrix of the machine-learning model to the first semantic-embedding-training vector and the second semantic-embedding-training vector; and

combining the first modified semantic-embedding-training vector, the second modified semantic-embedding-training vector, and the trait-intersection-count-training vector to create a combined-trait-intersection-training vector.

17. The system of claim 16 , further comprising instructions that, when executed by the at least one processor, cause the system to generate the estimated trait-intersection-training count for the target-training-time period utilizing the machine-learning model by applying additional network layers of the machine-learning model to the combined-trait-intersection-training vector to determine the estimated trait-intersection-training count for the target-training-time period.

18. The system of claim 14 , further comprising instructions that, when executed by the at least one processor, cause the system to:

identify a time-segment-training indicator corresponding to the initial-training-time period for the initial trait-intersection-training count; and

generate the estimated trait-intersection-training count for the target-training-time period by analyzing the time-segment-training indicator utilizing the machine-learning model.

19. The system of claim 14 , further comprising instructions that, when executed by the at least one processor, cause the system to:

determine, for a plurality of training-time periods, a rolling average trait-intersection-training count of users corresponding to the first training trait and the second training trait or a rolling extremum trait-intersection-training count of users corresponding to the first training trait and the second training trait; and

generate the estimated trait-intersection-training count for the target-training-time period in part by analyzing the rolling average trait-intersection-training count or the rolling extremum trait-intersection-training count utilizing the machine-learning model.

20. The system of claim 14 , wherein the machine-learning model comprises a decision-tree regressor or an artificial neural network.