Inferring brand similarities using graph neural networks and selection prediction

1. A non-transitory computer-readable medium embodying a program executable in at least one computing device, wherein when executed the program causes the at least one computing device to at least:

generate a brand-to-brand graph indicating similarities between a set of brands according to click-through data, wherein edges of the brand-to-brand graph represent a respective probability of a click-through to an item detail page associated with an item of a given brand associated with a first node following search queries associated with another brand associated with a second node;

receive a search query;

identify a first brand based at least in part on the search query;

analyze, using a first graph convolutional network (GCN) tower, the brand-to-brand graph to determine brand similarities among the first brand and a first set of other brands;

analyze, using a second GCN tower, the brand-to-brand graph to determine brand similarities among a second brand and a second set of other brands;

determine a level of similarity between the first brand and the second brand based at least in part on an output of the first GCN tower and an output of the second GCN tower; and

generate a user interface in response to the search query, wherein the user interface includes one or more items from the second brand that do not match the search query based at least in part on the level of similarity.

2. The non-transitory computer-readable medium of claim 1 , wherein determining the level of similarity between the first brand and the second brand is further based at least in part on determining that the first brand and the second brand include items associated with a particular item category or a particular price category.

3. The non-transitory computer-readable medium of claim 1 , wherein when executed the program further causes the at least one computing device to at least:

analyze, using a third GCN tower, the brand-to-brand graph to determine brand similarities among a third brand excluded from results following the search query and a third set of other brands; and

wherein the level of similarity is determined further based at least in part on an output of the third GCN tower.

4. A computer-implemented method, comprising:

generating a brand-to-brand graph indicating similarities between a set of brands according to at least one of: click-through data or conversion data, wherein edges of the brand-to-brand graph represent a respective probability of a selection of a given brand associated with a first node following search queries associated with another brand associated with a second node;

analyzing, using a first graph neural network (GNN) tower, the brand-to-brand graph to determine brand similarities among a first brand identified from a search query and a first set of other brands;

analyzing, using a second GNN tower, the brand-to-brand graph to determine brand similarities among a second brand and a second set of other brands; and

determining a level of similarity between the first brand and the second brand based at least in part on an output of the first GNN tower and an output of the second GNN tower.

5. The computer-implemented method of claim 4 , wherein the first GNN tower corresponds to a first graph convolutional network (GCN) tower, and the second GNN tower corresponds to a second GCN tower.

6. The computer-implemented method of claim 4 , further comprising:

analyzing, using a third GNN tower, the brand-to-brand graph to determine brand similarities among a third brand excluded from results following the search query and a third set of other brands; and

wherein the level of similarity is determined further based at least in part on an output of the third GNN tower.

7. The computer-implemented method of claim 4 , further comprising:

determining initial brand similarities for the first brand and initial brand similarities for the second brand using a node2vec algorithm; and

wherein the first GNN tower corresponds to a first deep neural network (DNN) tower, and the second GNN tower corresponds to a second DNN tower.

8. The computer-implemented method of claim 4 , wherein the selection corresponds to a click-through to an item detail page associated with an item of the second brand.

9. The computer-implemented method of claim 4 , wherein the selection corresponds to an order of an item of the second brand.

10. The computer-implemented method of claim 4 , wherein determining the level of similarity between the first brand and the second brand is further based at least in part on determining that the first brand and the second brand include items associated with a particular price category.

11. The computer-implemented method of claim 4 , wherein determining the level of similarity between the first brand and the second brand is further based at least in part on determining that the first brand and the second brand include items associated with a particular item category.

12. The computer-implemented method of claim 4 , further comprising generating a ranking of respective levels of similarity between the first brand and individual ones of a plurality of second brands.

13. The computer-implemented method of claim 4 , further comprising identifying the first brand as being one or more keywords in the search query.

14. The computer-implemented method of claim 4 , further comprising identifying the first brand as being associated with one or more keywords in the search query, the first brand being excluded from the search query.

15. The computer-implemented method of claim 4 , further comprising generating a user interface in response to receiving another search query for the first brand, wherein the user interface includes a recommendation for one or more items from the second brand based at least in part on the level of similarity determined between the first brand and the second brand.

16. A system, comprising:

at least one computing device; and

instructions executable in the at least one computing device, wherein when executed the instructions cause the at least one computing device to at least:

generate a brand-to-brand graph indicating similarities between a set of brands according to at least one of: click-through data or conversion data, wherein edges of the brand-to-brand graph represent a respective probability of a selection of a given brand associated with a first node following search queries associated with another brand associated with a second node;

analyze, using a first graph neural network (GNN) tower, the brand-to-brand graph to determine brand similarities among a first brand identified from a search query and a first set of other brands;

analyze, using a second GNN tower, the brand-to-brand graph to determine brand similarities among a second brand and a second set of other brands; and

determine whether to include one or more items from the second brand in a user interface generated in response to the search query based at least in part on an output of the first GNN tower and an output of the second GNN tower.

17. The system of claim 16 , wherein determining whether to include the one or more items from the second brand in the user interface is further based at least in part on determining that the first brand and the second brand include items associated with a particular price category and a particular item category.

18. The system of claim 16 , wherein the first brand is identified as being one or more keywords in the search query.

19. The system of claim 16 , wherein the first GNN tower corresponds to a first graph convolutional network (GCN) tower, and the second GNN tower corresponds to a second GCN tower.

20. The system of claim 16 , wherein the instructions further cause the at least one computing device to at least:

analyze, using a third GNN tower, the brand-to-brand graph to determine brand similarities among a third brand excluded from results following the search query and a third set of other brands; and

wherein determining whether to include the one or more items from the second brand in the user interface is further based at least in part on an output of the third GNN tower.