Researchers from Amazon have introduced artificial intelligence (AI) algorithms that help users buy clothes. One of them allows you to fine-tune your search queries, the other – offers products that can come to what the client has already chosen. The third one analyzes the user’s photos and then shows how the clothes that he plans to buy will look on him.
Engineers explained that to communicate with the Amazon AI, you can use reviews in the form of text, with different levels of detail. For example, a client may say that “something more formal is needed” or that he needs clothes “with a different throat”. AI will retain the features of the image, but only part of its parameters will change.
The system operates on three inputs: the source image, the text reaction, and the target image corresponding to the recall. Data sequentially passes through three submodels; at different points in the process, the original image is edited according to the user’s review. Amazon notes that the lower levels of the model are responsible for the incoming characteristics of a lower level (for example, texture and colors) and a higher level (sleeve length or fit); hierarchical reconciliation helps to train the system to work with text modifications.
Each selection is performed by a separate two-component model. Researchers note that AI helped to better match clothes based on reviews 58% more efficiently.