Predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks
This disclosure involves predictive modeling with entity representations computed from neural network models simultaneously trained on multiple tasks. For example, a method includes a processing device performing operations including accessing input data for an entity and transforming the input data into a dense vector entity representation representing the entity. Transforming the input data includes applying, to the input data, a neural network including simultaneously trained propensity models. Each propensity model predicts a different task based on the input data. Transforming the input data also includes extracting the dense vector entity representation from a common layer of the neural network to which the propensity models are connected. The operations performed by the processing device include computing a predicted behavior by applying a predictive model to the dense vector entity representation and transmitting the predicted behavior to a computing device that customizes a presentation of electronic content at a remote user device.
1. A method comprising:
accessing, by one or more processing devices of a model development system, from a non-transitory computer-readable medium, input data for an entity;
transforming, by the one or more processing devices, the input data into a dense vector entity representation representing the entity, wherein transforming the input data comprises:
applying, to the input data, a neural network comprising simultaneously trained subnets representing propensity models, wherein each propensity model is configured to predict a different task based on the input data, and
extracting the dense vector entity representation from a common layer of the neural network to which the simultaneously trained subnets are connected; and
computing, by the one or more processing devices, a predicted behavior by applying a predictive model to the dense vector entity representation; and
transmitting, by the one or more processing devices, the predicted behavior to a computing device that customizes, in accordance with the predicted behavior, an interaction experience of a user of a remote user device associated with the entity, wherein customizing the interaction experience includes altering a presentation of information in electronic content displayed on the remote user device.
2. The method of claim 1 , further comprising, by the one or more processing devices:
generating a lookalike model, wherein the predictive model comprises the lookalike model, wherein computing the predicted behavior comprises identifying a set of entities with similar characteristics to the entity by applying the lookalike model to the dense vector entity representation of the entity.
3. The method of claim 2 , wherein the lookalike model compares the dense vector entity representation with other dense vector entity representations of other entities to identify the set of entities with the similar characteristics to the entity.
4. The method of claim 1 , further comprising, by the one or more processing devices:
generating a plurality of output predictions of a plurality of tasks, each task of the plurality of tasks associated with one of a plurality of propensity models, wherein the plurality of output predictions are usable to customize the presentation of the electronic content at the remote user device associated with the entity.
5. The method of claim 4 , wherein the plurality of output predictions comprises a first output prediction and a second output prediction, and wherein the second output prediction relies at least in part on the first output prediction.
6. The method of claim 1 , further comprising:
generating, by the one or more processing devices, the neural network, wherein generating the neural network comprises:
selecting portions of the input data of a plurality of entities to generate training data;
identifying a plurality of propensity models of the neural network; and
simultaneously fitting the plurality of propensity models to the training data.
7. The method of claim 1 , further comprising, by the one or more processing devices:
customizing the presentation of electronic content at the remote user device associated with the entity based on the dense vector entity representation generated by the neural network.
8. The method of claim 1 , further comprising:
identifying, by the one or more processing devices, categorical variables of the input data;
autoencoding, by the one or more processing devices, the categorical variables to generate a numerical representation of the categorical variables; and
providing, by the one or more processing devices, the numerical representation of the categorical variables to the neural network.
9. A computing system comprising:
means for accessing, from a non-transitory computer-readable medium, input data for an entity;
means for transforming the input data into a dense vector entity representation representing the entity, wherein the means for transforming the input data comprises:
means for applying, to the input data, a neural network comprising simultaneously trained subnets representing propensity models, wherein each propensity model is configured to predict a different task based on the input data, and
means for extracting the dense vector entity representation from a common layer of the neural network to which the simultaneously trained subnets are connected; and
means for computing a predicted behavior by applying a predictive model to the dense vector entity representation; and
means for transmitting the predicted behavior to a computing device that customizes, in accordance with the predicted behavior, an interaction experience of a user of a remote user device associated with the entity, wherein customizing the interaction experience includes altering a presentation of information in electronic content displayed on the remote user device.
10. The computing system of claim 9 , further comprising:
means for generating a lookalike model, wherein the predictive model comprises the lookalike model, wherein computing the predicted behavior comprises identifying a set of entities with similar characteristics to the entity by applying the lookalike model to the dense vector entity representation of the entity.
11. The computing system of claim 10 , wherein the lookalike model compares the dense vector entity representation with other dense vector entity representations of other entities to identify the set of entities with the similar characteristics to the entity.
12. The computing system of claim 9 , further comprising:
means for generating a plurality of output predictions of a plurality of tasks, each task of the plurality of tasks associated with one of a plurality of propensity models, wherein the plurality of output predictions are usable to customize the presentation of the electronic content at the remote user device associated with the entity.
13. The computing system of claim 12 , wherein the plurality of output predictions comprises a first output prediction and a second output prediction, and wherein the second output prediction relies at least in part on the first output prediction.
14. The computing system of claim 9 , further comprising:
means for customizing the presentation of electronic content at the remote user device associated with the entity based on the dense vector entity representation generated by the neural network.
15. The computing system of claim 9 , further comprising:
means for identifying categorical variables of the input data;
means for autoencoding the categorical variables to generate a numerical representation of the categorical variables; and
means for providing the numerical representation of the categorical variables to the neural network.
16. A non-transitory computer-readable medium having instructions stored thereon, the instructions executable by a processing device to perform operations comprising:
accessing, from a non-transitory computer-readable medium, input data for an entity;
transforming the input data into a dense vector entity representation representing the entity, wherein transforming the input data comprises:
applying, to the input data, a neural network comprising simultaneously trained subnets representing propensity models, wherein each propensity model is configured to predict a different task based on the input data, and
extracting the dense vector entity representation from a common layer of the neural network to which the simultaneously trained subnets are connected; and
computing a predicted behavior by applying a predictive model to the dense vector entity representation; and
transmitting the predicted behavior to a computing device that customizes, in accordance with the predicted behavior, an interaction experience of a user of a remote user device associated with the entity, wherein customizing the interaction experience includes altering a presentation of information in electronic content displayed on the remote user device.
17. The non-transitory computer-readable medium of claim 16 , the instructions executable to perform operations further comprising:
generating a lookalike model, wherein the predictive model comprises the lookalike model, wherein computing the predicted behavior comprises identifying a set of entities with similar characteristics to the entity by applying the lookalike model to the dense vector entity representation of the entity.
18. The non-transitory computer-readable medium of claim 17 , wherein the lookalike model compares the dense vector entity representation with other dense vector entity representations of other entities to identify the set of entities with the similar characteristics to the entity.
19. The non-transitory computer-readable medium of claim 16 , the instructions executable to perform operations further comprising:
generating the neural network, wherein generating the neural network comprises:
selecting portions of the input data of a plurality of entities to generate training data;
identifying a plurality of propensity models of the neural network; and
simultaneously fitting the plurality of propensity models to the training data.
20. The non-transitory computer-readable medium of claim 16 , the instructions executable to perform operations further comprising:
identifying categorical variables of the input data;
autoencoding the categorical variables to generate a numerical representation of the categorical variables; and
providing the numerical representation of the categorical variables to the neural network.