BUNDLE CREATION FROM DESCRIPTION CONTENT USING MACHINE LEARNING

Abstract

A bundle generation engine provides recommendations for an item and at least one compatible item. Upon receiving an item listing for an item at a listing platform, the bundle generation engine determines that unstructured text of an item description for the item listing includes one or more compatible items. For example, the bundle generation engine may use a classification model to determine the unstructured text includes the one or more compatible items. Based on determining that the unstructured text includes the one or more compatible items, the bundle generation engine identifies one or more identifiers within the unstructured text of the item description. For example, the identifiers may be model identifiers or brand identifiers. In aspects, the identifiers are identified using one or more natural language processing models. A bundle recommendation is provided to a user device based on the one or more identifiers.

Description

Claims

1. A computer-implemented method comprising: receiving an item listing for an item at a listing platform;determining, using a classification model, that unstructured text of an item description for the item listing includes compatible items;based on determining the unstructured text of the item description includes the compatible items, identifying, using a natural language processing model, identifiers for the compatible items included within the unstructured text of the item description of the item listing for the item;selecting, from an inventory of the listing platform, an item listing for a first compatible item identified based on a first identifier of the identifiers for the first compatible item determined from the unstructured text of the item description of the item listing for the item; andproviding, to a user device, a bundle recommendation comprising the item listing for the item and the item listing for the first compatible item.

2. The computer-implemented method of claim 1, further comprising: prior to selecting the item listing for the first compatible item, extracting, from the inventory of the listing platform, a plurality of item listings for a plurality of compatible items based on the identifiers included within item descriptions of each of the plurality of compatible items, the plurality of compatible items extracted using one or more entity recognition natural language processing models; andselecting the item listing for the first compatible item of the plurality of compatible items.

3. The computer-implemented method of claim 2, further comprising: determining the first compatible item based on user interactions, by other users, with bundle recommendations that each comprise at least one of the plurality of item listings.

4. The computer-implemented method of claim 1, further comprising: identifying a first category for a first subset of the compatible items;identifying, from the inventory of the listing platform, a plurality of item listings for items of the first category based on one or more identifiers determined for the first subset of the compatible items;determining a ranking for each item listing from the plurality of item listings for the items of the first category; andwherein the item listing for the first compatible item is selected from the plurality of item listings for the items of the first category based on the rankings.

5. The computer-implemented method of claim 4, further comprising: identifying a second category for a second subset of the compatible items;identifying, from the inventory of the listing platform, a second plurality of item listings for items of the second category based on one or more identifiers determined for the second subset of the compatible items;determining a ranking for each item listing from the second plurality of item listings for the items of the second category;selecting an item listing for a second compatible item based on the ranking; andwherein the bundle recommendation further comprises the item listing for the second compatible item.

6. The computer-implemented method of claim 4, wherein the ranking for each item listing from the plurality of item listings for the items of the first category are based on one or more selected from the following: user interactions with each item listing, shipment data for each item listing, and price associated with each item listing.

7. The computer-implemented method of claim 1, further comprising: mapping entities to the identifiers for the compatible items;identifying an entity, from the mapping of the entities to the identifiers, for the first compatible item; andproviding, to the user device, the bundle recommendation including the entity for the first compatible item.

8. One or more non-transitory computer storage media storing computer-readable instructions that when executed by a processor, cause the processor to perform operations, the operations comprising: receiving an item listing for an item at a listing platform, the item listing including an item description for the item;determining that unstructured text of the item description includes one or more compatible items;based on determining the unstructured text of the item description includes the one or more compatible items, identifying identifiers for the one or more compatible items included within the unstructured text of the item description;selecting, from an inventory of the listing platform, a first compatible item from the one or more compatible items based on a first identifier of the identifiers, the first identifier corresponding to the first compatible item; andproviding, to a user device, a bundle recommendation comprising the first compatible item and the item.

9. The one or more non-transitory computer storage media of claim 8, wherein the identifiers are identified using a generative pre-trained transformer.

10. The one or more non-transitory computer storage media of claim 8, wherein the first compatible item is selected based on extracting the first identifier from a first item description within a first item listing of the first compatible item using entity recognition natural language processing.

11. The one or more non-transitory computer storage media of claim 8, further comprising: determining that the unstructured text of the item description includes a plurality of compatible items;identifying an identifier for each of the plurality of compatible items included within the unstructured text of the item description;identifying a compatible item listing for each of the plurality of compatible items;determining a ranking for each of the compatible item listings; andselecting the first compatible item for the bundle recommendation based on the ranking for each of the compatible item listings.

12. The one or more non-transitory computer storage media of claim 11, wherein each of the compatible item listings are ranked based on user feedback for each of the plurality of compatible items or shipment data for each of the plurality of compatible items.

13. The one or more non-transitory computer storage media of claim 8, wherein the first identifier is a model identifier.

14. The one or more non-transitory computer storage media 8, wherein the first compatible item is selected based on extracting the first identifier from unstructured text within a first item listing of the first compatible item using entity recognition natural language processing.

15. A system comprising: at least one processor; andone or more computer storage media storing computer-readable instructions that when executed by the at least one processor, cause the at least one processor to perform operations comprising: receiving an item listing for an item at a listing platform, the item listing comprising an item description for the item;determining, using a classification model, that unstructured text of the item description includes one or more compatible items;based on determining the unstructured text includes the one or more compatible items, identifying, using one or more natural language processing models, at least one identifier for each of the one or more compatible items included within the unstructured text of the item description;selecting, from an inventory of the listing platform, a first compatible item identified based on a first identifier of the at least one identifier; andproviding, to a user device, a bundle recommendation comprising the item and the first compatible item.

16. The system of claim 15, further comprising selecting the first compatible item identified based on identifying the first identifier from a first item description of a first item listing of the first compatible item, wherein the first identifier is identified from the first item description using the one or more natural language processing models.

17. The system of claim 16, wherein the bundle recommendation comprises the first item listing of the first compatible item.

18. The system of claim 15, wherein the at least one identifier comprises a model and brand identifier.

19. The system of claim 15, further comprising: determining that the unstructured text of the item description includes a plurality of compatible items;identifying a model identifier for each of the plurality of compatible items included within the unstructured text of the item description;identifying a compatible item listing for each of the plurality of compatible items;determining a ranking for each compatible item listing, the ranking based on prior bundle purchases comprising the corresponding compatible item of the plurality of compatible items; andselecting the first compatible item for the bundle recommendation based on the ranking for each of the compatible item listings.

20. The system of claim 19, wherein each ranking is further based on shipment data for each of the compatible item listings.