Accelerating SEO-focused image tagging with AI-driven automation

Teezily turned to Addepto to develop an automatic model responsible for creating descriptions for images. Teezily’s massive image database needed to be properly described for SEO purposes. Once again, Teezily’s managers considered AI the right development direction and Addepto as the right partner to take care of the technical details.

Read about our first project delivered for Teezily

Tagging user-generated images with proper keywords

Teezily’s image catalog was vast, as the user-generated content is based on the company’s business model. And yet, to be searchable by Google, every single image needs to be described with proper tags to let Google know what is in it. Users who uploaded images could not be burdened with the obligation to tag their artworks, as – first – it had to be done according to very specific rules, and second – it would be friction in the UX, the step preventing them from focusing on creation.

SEO, however, has its own requirements that have to be met. Keywords describing the picture content are one of the crucial ranking factors for Google algorithms. Without proper tags, Google bots can’t understand after what search queries given images are supposed to be displayed for users, which is a killer for optimizing in search.

Enhance visibility in Google Search

Teezily wanted to implement AI-driven modules that would be able to “read” through the entire base of images, understand their content, choose the most proper keywords to describe it, and put it in a given catalog structure.

Customized NLP solution to automate SEO image tagging

Addepto decided to take a unique approach to tackle this challenge. The evident direction seemed to be computer vision models combined with NLP, but it would require a huge amount of training sets and lead to high costs. Not to mention that – given the existing Google Images Search – reaching for computer vision would look like taking a sledgehammer to crack a nut.

We wanted to tackle the challenge most cost-effectively. Thus, we decided to use the state-of-art solutions already available on the market and enhance them with custom-made NLP algorithms. This approach allowed us to accelerate the development process significantly.

– Wojciech Drężek, Data Scientist at Addepto

Addepto Team chose not to do that and looked for a more efficient way of automation tagging images instead. The company chose to reach out to the technology underlying Google Image Search because – when it comes to tagging images, it mastered it, though enhanced it with custom-made NLP algorithms trained with – among others – Wikipedia articles.

All of it was aimed at building the AI-driven engine able to pinpoint the main keyword that described the main and most visible image content but also enriched the description with the following keywords.