Systems and methods for providing customer insights

1 . A system comprising:

a processor; and

a non-transitory memory storing instructions that, when executed, cause the processor to:

obtain user data associated with a user from a database, wherein the user data includes historical interaction data;

generate a historical interaction set of tensors representative of the historical interaction data;

transmit the historical interaction set of tensors to a plurality of processing units;

determine whether the user data includes a session entry within a predetermined time period;

responsive to determining the user data does not include the session entry within the predetermined time period, in real-time select a plurality of product types relevant to the user based on a relevance score related to each of the plurality of product types for the user;

for each product type of the plurality of product types:

execute an attribute model based on the historical interaction set of tensors, including a machine learning model configured to capture a non-linear relationship, by distributing a plurality of processing tasks to a first processing unit of the plurality of processing units to determine in real-time whether a first attribute of a first product type and a second attribute of a second product type are semantically similar attributes, and, if so, to represent the semantically similar attributes in a manner that reflects their similarity to generate a set of attributes associated with the product type, wherein each attribute in the set of attributes is associated with an affinity score for the user; and

for each attribute of the set of attributes, determine an overall score for an attribute and product type pair, based on the relevance score for the product type and the affinity score for the attribute, wherein the overall score is determined by:

executing a semantic similarity model by distributing a second plurality of processing tasks to the plurality of processing units to generate a first set of tensors representative of a semantic similarity of the plurality of product types;

generating, by a scoring engine, the relevance score for the product type based on the first set of tensors and the user data by executing a relevance model by distributing a third plurality of processing tasks to a second processing unit of the plurality of processing units;

generating a second set of tensors representative of a semantic similarity of the set of attributes based on a universal sentence encoding using the semantic similarity model, wherein the semantic similarity model encodes and embeds the set of attributes;

generating, by the scoring engine, the affinity score for the attribute based on the second set of tensors and the user data; and

combining the relevance score and the affinity score to generate the overall score; and

select for presentation, via a user interface, at least one of the attribute and product type pairs based on the overall score of each of the attribute and product type pairs, and wherein a format of the presentation is based on the at least one of the attribute and product type pairs selected for presentation.

2 . The system of claim 1 , wherein the user data includes historical user purchase data and historical user engagement data.

3 . The system of claim 1 , wherein the instructions, when executed, further cause the processor to:

determine a number of transactions within the user data; and

wherein the plurality of product types are obtained based on a determination that the number of transactions is greater than a predetermined threshold number within a predetermined time period.

4 . The system of claim 1 , wherein the set of attributes includes one more of a brand, a flavor, and a dietary preference.

5 . The system of claim 1 , wherein the affinity score for each attribute indicates a likelihood of the user buying a product based on the attribute.

6 . The system of claim 1 , wherein the instructions, when executed, further cause the processor to:

determine, based on the user data, a number of user transactions including a product with the at least one of the attribute and product type pairs selected for presentation; and

wherein the number of user transaction is selected for presentation with the at least one of the attribute and product type pairs.

7 . A method comprising:

obtaining user data associated with a user from a database, wherein the user data includes historical interaction data;

generating a historical interaction set of tensors representative of the historical interaction data;

transmitting the historical interaction set of tensors to a plurality of processing units;

determining whether the user data includes a session entry within a predetermined time period;

responsive to determining the user data does not include the session entry within the predetermined time period, in real-time selecting a plurality of product types relevant to the user based on a relevance score related to each of the plurality of product types for the user;

for each product type of the plurality of product types:

executing an attribute model based on the historical interaction set of tensors, including a machine learning model configured to capture a non-linear relationship, by distributing a plurality of processing tasks to a first processing unit of the plurality of processing units to determine in real-time whether a first attribute of a first product type and a second attribute of a second product type are semantically similar attributes, and, if so, to represent the semantically similar attributes in a manner that reflects their similarity to generate a set of attributes associated with the product type, wherein each attribute in the set of attributes is associated with an affinity score for the user; and

for each attribute of the set of attributes, determining an overall score for an attribute and product type pair based on the relevance score for the product type and the affinity score for the attribute, wherein the overall score is determined by:

executing a semantic similarity model by distributing a second plurality of processing tasks to the plurality of processing units to generate a first set of tensors representative of a semantic similarity of the plurality of product types;

generating, by a scoring engine, the relevance score for the product type based on the first set of tensors and the user data by executing a relevance model by distributing a third plurality of processing tasks to a second processing unit of the plurality of processing units;

generating a second set of tensors representative of a semantic similarity of the set of attributes based on a universal sentence encoding using the semantic similarity model, wherein the semantic similarity model encodes and embeds the set of attributes;

generating, by the scoring engine, the affinity score for the attribute based on the second set of tensors and the user data; and

combining the relevance score and the affinity score to generate the overall score; and

selecting for presentation at least one of the attribute and product type pairs based on the overall score of each of the attribute and product type pairs, and wherein a format of the presentation is based on the at least one of the attribute and product type pairs selected for presentation.

8 . The method of claim 7 , wherein the user data includes historical user purchase data and historical user engagement data.

9 . The method of claim 7 , the method further comprising:

determining a number of transactions within the user data; and

obtaining the plurality of product types based at least in part on a determination that the number of transactions is greater than a predetermined threshold within a predetermined time period.

10 . The method of claim 7 , wherein the set of attributes includes one more of a brand, a flavor, and a dietary preference.

11 . The method of claim 7 , wherein the affinity score for each attribute indicates a likelihood of the user buying a product based on the attribute.

12 . The method of claim 7 , the method further comprising:

determining, based on the user data, a number of user transactions including a product with the at least one of the attribute and product type pairs selected for presentation; and

selecting for presentation the number of user transaction with the at least one of the attribute and product type pairs.

13 . A non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed by at least one processor, cause a device to perform operations comprising:

obtaining user data associated with a user from a database, wherein the user data includes historical interaction data;

generating a historical interaction set of tensors representative of the historical interaction data;

transmitting the historical interaction set of tensors to a plurality of processing units;

determining whether the user data includes a session entry within a predetermined time period;

responsive to determining the user data does not include the session entry within the predetermined time period, selecting in real-time a plurality of product types relevant to the user based on a relevance score related to each of the plurality of product types for the user;

for each product type of the plurality of product types:

executing an attribute model based on the historical interaction set of tensors, including a machine learning model configured to capture a non-linear relationship, by distributing a plurality of processing tasks to a first processing unit of the plurality of processing units to determine in real-time whether a first attribute of a first product type and a second attribute of a second product type are semantically similar attributes, and, if so, to represent the semantically similar attributes in a manner that reflects their similarity to generate a set of attributes associated with the product type, wherein each attribute in the set of attributes is associated with an affinity score for the user; and

for each attribute of the set of attributes, determining an overall score for an attribute and product type pair based on the relevance score for the product type and the affinity score for the attribute, wherein the overall score is determined by:

executing a semantic similarity model by distributing a second plurality of processing tasks to the plurality of processing units to generate a first set of tensors representative of a semantic similarity of the plurality of product types;

generating, by a scoring engine, the relevance score for the product type based on the first set of tensors and the user data by executing a relevance model by distributing a third plurality of processing tasks to a second processing unit of the plurality of processing units;

generating a second set of tensors representative of a semantic similarity of the set of attributes based on a universal sentence encoding using the semantic similarity model, wherein the semantic similarity model encodes and embeds the set of attributes;

generating, by the scoring engine, the affinity score for the attribute based on the second set of tensors and the user data; and

combining the relevance score and the affinity score to generate the overall score; and

selecting for presentation at least one of the attribute and product type pairs based on the overall score of each of the attribute and product type pairs, and wherein a format of the presentation is based on the at least one of the attribute and product type pairs selected for presentation.

14 . The non-transitory computer readable medium of claim 13 , the operations further comprising:

determining a number of transactions within the user data; and

obtaining the plurality of product types based at least in part on a determination that the number of transactions is greater than a predetermined threshold within a predetermined time period.

15 . The non-transitory computer readable medium of claim 14 , the operations further comprising:

determining, based on the user data, that the user has been inactive for a predetermined minimum threshold of time; and

obtaining the plurality of product types based at least in part on the determination that the user has been inactive for the predetermined minimum threshold of time.

16 . The non-transitory computer readable medium of claim 13 , the operations further comprising:

determining, based on the user data, a number of user transactions including a product with the at least one of the attribute and product type pairs selected for presentation; and

selecting for presentation the number of user transaction with the at least one of the attribute and product type pairs.

17 . The system of claim 1 , wherein the instructions, when executed, further cause the processor to:

determine the overall score for the attribute and product pair type by executing a favorites model by distributing a fourth plurality of processing tasks to a third processing unit of the plurality of processing units.

18 . The system of claim 1 , wherein the machine learning model of the attribute model includes a neural network.

19 . The method of claim 7 , further comprising:

determining the overall score for the attribute and product pair type by executing a favorites model by distributing a fourth plurality of processing tasks to a third processing unit of the plurality of processing units.

20 . The method of claim 7 , wherein the machine learning model of the attribute model includes a neural network.