Automating the generation of product descriptions using catalog images
The concept was to create duplicates, monitor customer engagement with the product, and consistently enhance the writing to boost interaction and sales. This was to be achieved without training human writers in the company’s specific writing style for the descriptions.
Temporibus autem quibusdam et aut officiis debitis aut rerum necessitatibus saepe eveniet ut et voluptates repudiandae sint et molestiae non recusandae. Itaque earum rerum hic tenetur a sapiente delectus, ut aut reiciendis voluptatibus maiores alias consequatur aut perferendis doloribus asperiores repellat.
The AI description generator has been trained to create various types of product copy, including a product name, a short description, and a longer description that expands on the elements mentioned in the short description. This automated process ensures that all product-related content on the website can be generated quickly, enabling the client to efficiently update their online inventory.
The model was trained on the merchant's existing product descriptions to ensure that the automated descriptions aligned with the merchant's style guide and brand positioning. This AI-powered solution reduced the time and effort required for onboarding new copywriters. Existing staff were able to focus on reviewing and generating product descriptions for items that the model hadn't been trained on yet.
A custom deep learning model, pre-trained on ImageNet, was used to analyze the product descriptions connected to product images.
NLP was used to learn how to describe the product features and combine the feature descriptions into the product copy.
The model was deployed to the customer’s on-premises infrastructure. The alerts and dashboard graphs were then pushed to a React-frontend website, deployed on the cloud.
The automated approach reduces the manual effort required to generate, validate, and check product descriptions. Additionally, it ensures that the resulting copy is free from any grammatical or spelling errors. A trained NLP model generates the descriptions using only the product images as input, detecting and appropriately describing the key features of the product.
