Automated product attribute selection

A non-transitory computer-readable storage medium storing executable computer program instructions, the instructions executable to perform operations comprising:

• receiving product data for a product, the product data comprising product image data for the product and product text data for the product, wherein the product is associated with a category, and wherein the category comprises a plurality of attributes;
• generating a plurality of individual probability distributions for the attributes based on the product data, wherein each individual probability distribution is generated using a different technique, wherein each individual probability distribution includes a respective individual probability for each of the attributes, and wherein the respective probability for each of the individual attributes represents a likelihood that the attribute is applicable to the product, wherein generating the plurality of individual probability distributions comprises:
• generating a first individual probability distribution using a first generator that determines probabilities based on the product image data, and
• generating a second individual probability distribution using a second generator that determines probabilities based on the product text data;
• generating a combined probability distribution for the attributes based on a weighted sum of the plurality of individual probability distributions, wherein the combined probability distribution comprises a respective combined probability for each of the attributes;
• selecting an attribute from the plurality of attributes having a highest combined probability;
• and
• associating an attribute label for the selected attribute with the product.

A system comprising:

• one or more computers;
• and
• one or more storage devices storing executable computer program instructions that when executed by the one or more computers cause the one or more computers to perform operations comprising:
• receiving product data for a product, the product data comprising product image data for the product and product text data for the product, wherein the product is associated with a category, and wherein the category comprises a plurality of attributes;
• generating a plurality of individual probability distributions for the attributes based on the product data, wherein each individual probability distribution is generated using a different technique, wherein each individual probability distribution includes a respective individual probability for each of the attributes, and wherein the respective probability for each of the individual attributes represents a likelihood that the attribute is applicable to the product, wherein generating the plurality of individual probability distributions comprises:
• generating a first individual probability distribution using a first generator that determines probabilities based on the product image data, and
• generating a second individual probability distribution using a second generator that determines probabilities based on the product text data;
• generating a combined probability distribution for the attributes based on a weighted sum of the plurality of individual probability distributions, wherein the combined probability distribution comprises a respective combined probability for each of the attributes;
• selecting an attribute from the plurality of attributes having a highest combined probability;
• and
• associating an attribute label for the selected attribute with the product.