Systems, method, and non-transitory computer-readable storage media for multi-modal product classification

1. A multi-modal computer classification network system comprising:

a memory device including an item records database and a labeling database, the item records database including a plurality of item records, each item record of the plurality of item records including an item ID, a first classification metadata set, and a second classification metadata set, the labeling database including a plurality of labeling records, each labeling record of the plurality of labeling records including a class label and labeling data;

a first classification computer server including a first classifier program configured to:

select an item record from the item records database; and

generate a first classification record including a first ranked list of class labels, the first ranked list of class labels generated using first input data including the plurality of labeling records and the first classification metadata set included in the item record as selected;

a second classification computer server including a second classifier program configured to:

generate a second classification record including a second ranked list of class labels, the second ranked list of class labels generated using second input data including the plurality of labeling records and the second classification metadata set included in the item record as selected;

a policy computer server including a policy network programmed to:

determine a predicted class label based on the first and second ranked lists of class labels;

generate a labeled item record including the item ID included in the item record, as selected, and the predicted class label; and

generate a plurality of labeled item records and store the plurality of labeled item records in a labeled item database; and

a web server including a processor programmed to:

generate and display a product search website on a user computing device;

receive, from a user via the user computing device, a product search request;

determine a product category based on the product search request as received;

access the labeled item database and select labeled item records having predicted class labels matching the product category; and

generate and display a product search result webpage including product information associated with the selected labeled item records.

2. The system of claim 1 , wherein the policy network includes at least one of a Convolutional Neural Network (CNN), a logistic regression, a Support Vector Machine (SVM), a Generic Neural Network, or a Random Forest classifier.

3. The system of claim 1 , wherein the first classification metadata set includes text data.

4. The system of claim 3 , wherein the second classification metadata set includes image data.

5. The system of claim 3 , wherein the second classification metadata set includes video data.

6. The system of claim 1 , wherein the first classifier program includes a text CNN classifier and the second classifier program includes an image CNN classifier.

7. The system of claim 6 , wherein the image CNN classifier is a VGG network.

8. A multi-modal computer classification network system comprising:

a memory device including an item records database and a labeling database, the item records database including a plurality of item records, each item record of the plurality of item records including an item ID, a first classification metadata set, and a second classification metadata set, the labeling database including a plurality of labeling records, each labeling record of the plurality of labeling records including a class label and labeling data;

a first classification computer server including a first classifier program configured to:

select an item record from the item records database; and

generate a first classification record including a first ranked list of class labels, the first ranked list of class labels generated using first input data including the plurality of labeling records and the first classification metadata set included in the item record as selected;

a second classification computer server including a second classifier program configured to:

generate a second classification record including a second ranked list of class labels, the second ranked list of class labels generated using second input data including the plurality of labeling records and the second classification metadata set included in the item record as selected; and

a policy computer server including a policy network programmed to:

determine a predicted class label based on the first and second ranked lists of class labels; and

generate a labeled item record including the item ID included in the item record, as selected, and the predicted class label, wherein:

the first classifier program is further configured to generate a first classification probability associated with each class label included in the first ranked list of class labels;

the second classifier program is configured to generate a second classification probability associated with each class label included in the second ranked list of class labels;

the policy network is programmed to determine the predicted class label based on the first and second ranked lists of class labels and the corresponding first and second classification probabilities.

9. One or more non-transitory computer-readable storage media, having computer-executable instructions embodied thereon, that, when executed by at least one processor, the computer-executable instructions cause the at least one processor to function as:

a first classification computer server including a first classifier program configured to:

access a memory device including an item records database and a labeling database, the item records database including a plurality of item records, each item record of the plurality of item records including an item ID, a first classification metadata set, and a second classification metadata set, the labeling database including a plurality of labeling records, each labeling record of the plurality of labeling records including a class label and labeling data;

select an item record from the item records database; and

generate a first classification record including a first ranked list of class labels, the first ranked list of class labels generated using first input data including the plurality of labeling records and the first classification metadata set included in the item record as selected;

a second classification computer server including a second classifier program configured to:

generate a second classification record including a second ranked list of class labels, the second ranked list of class labels generated using second input data including the plurality of labeling records and the second classification metadata set included in the item record as selected;

a policy computer server including a policy network programmed to:

determine a predicted class label based on the first and second ranked lists of class labels;

generate a labeled item record including the item ID included in the item record, as selected, and the predicted class label; and

generate a plurality of labeled item records and store the plurality of labeled item records in a labeled item database; and

a web server including a processor programmed to:

generate and display a product search website on a user computing device;

receive, from a user via the user computing device, a product search request;

determine a product category based on the product search request as received;

access the labeled item database and select labeled item records having predicted class labels matching the product category; and

generate and display a product search result webpage including product information associated with the selected labeled item records.

10. The one or more non-transitory computer-readable storage media of claim 9 , wherein the policy network includes at least one of a Convolutional Neural Network (CNN), a logistic regression, a Support Vector Machine (SVM), a Generic Neural Network, or a Random Forest classifier.

11. The one or more non-transitory computer-readable storage media of claim 9 , wherein the first classification metadata set includes text data.

12. The one or more non-transitory computer-readable storage media of claim 11 , wherein the second classification metadata set includes image data.

13. The one or more non-transitory computer-readable storage media of claim 11 , wherein the second classification metadata set includes video data.

14. The one or more non-transitory computer-readable storage media of claim 9 , wherein the first classifier includes a text CNN classifier and the second classifier includes an image CNN classifier.

15. The one or more non-transitory computer-readable storage media of claim 14 , wherein the image CNN classifier is a VGG network.

16. One or more non-transitory computer-readable storage media, having computer-executable instructions embodied thereon, that, when executed by at least one processor, the computer-executable instructions cause the at least one processor to function as:

a first classification computer server including a first classifier program configured to:

access a memory device including an item records database and a labeling database, the item records database including a plurality of item records, each item record of the plurality of item records including an item ID, a first classification metadata set, and a second classification metadata set, the labeling database including a plurality of labeling records, each labeling record of the plurality of labeling records including a class label and labeling data;

select an item record from the item records database; and

generate a first classification record including a first ranked list of class labels, the first ranked list of class labels generated using input data including the plurality of labeling records and the first classification metadata set included in the item record as selected;

a second classification computer server including a second classifier program configured to:

generate a second classification record including a second ranked list of class labels, the second ranked list of class labels generated using input data including the plurality of labeling records and the second classification metadata set included in the item record as selected; and

a policy computer server including a policy network programmed to:

determine a predicted class label based on the first and second ranked lists of class labels; and

generate a labeled item record including the item ID included in the item record as selected and the predicted class label, wherein:

the first classifier program is further configured to generate a first classification probability associated with each class label included in the first ranked list of class labels;

the second classifier program is configured to generate a second classification probability associated with each class label included in the second ranked list of class labels; and

the policy network is programmed to determine the predicted class label based on the first and second ranked lists of class labels and the corresponding first and second classification probabilities.

17. A computer-implemented method for operating a multi-modal computer classification network system, the computer implemented method including steps of:

generating and storing, by a memory device, an item records database and a labeling database, the item records database including a plurality of item records, each item record of the plurality of item records including an item ID, a first classification metadata set, and a second classification metadata set, the labeling database including a plurality of labeling records, each labeling record of the plurality of labeling records including a class label and labeling data;

execute, on a first classification computer server, a first classifier program configured to:

select an item record from the item records database; and

generate a first classification record including a first ranked list of class labels, the first ranked list of class labels generated using first input data including the plurality of labeling records and the first classification metadata set included in the item record as selected;

execute, on a second classification computer server, a second classifier program configured to:

generate a second classification record including a second ranked list of class labels, the second ranked list of class labels generated using second input data including the plurality of labeling records and the second classification metadata set included in the item record as selected;

execute, on a policy computer server including a policy network, a program configured to:

determine a predicted class label based on the first and second ranked lists of class labels;

generate a labeled item record including the item ID included in the item record as selected and the predicted class label; and

generate a plurality of labeled item records and store the plurality of labeled item records in a labeled item database; and

execute, on a web server including a processor, a program configured to:

generate and display a product search website on a user computing device;

receive, from a user via the user computing device, a product search request;

determine a product category based on the product search request as received;

access the labeled item database and select labeled item records having predicted class labels matching the product category; and

generate and display a product search result webpage including product information associated with the selected labeled item records.

18. The computer-implemented method of claim 17 , wherein the policy network includes at least one of a Convolutional Neural Network (CNN), a logistic regression, a Support Vector Machine (SVM), a Generic Neural Network, or a Random Forest classifier.

19. The computer-implemented method of claim 17 , wherein:

the first classifier program is further configured to generate a first classification probability associated with each class label included in the first ranked list of class labels;

the second classifier program is configured to generate a second classification probability associated with each class label included in the second ranked list of class labels;

the policy network is programmed to determine the predicted class label based on the first and second ranked lists of class labels and the corresponding first and second classification probabilities.

20. The computer-implemented method of claim 17 , wherein the first classification metadata set includes text data.