Vector Embeddings for eCommerce Ad Optimization

Vector Embeddings for eCommerce Ad Optimization
While ChatGPT garners most of the buzz, vector embeddings might make search algorithms become highly irrelevant - especially for eCommerce companies.

While ChatGPT garners most of the buzz, especially in the areas of content generation, another generative AI innovation boasts the potential to revolutionize internet search. In fact, something known as vector embeddings might make search algorithms become highly irrelevant - especially for eCommerce companies.

Generative AI’s ability to use large language models (LLM) to generate different types of text that seem written by humans almost obscures its ability to understand complex phrases. It combines with other underpublicized capabilities of this AI variant to potentially make the current keyword matching logic used for internet search obsolete. 

That previous sentence reveals why Google fears for its revenue model that largely depends on internet advertising. Its parent company, Alphabet, hides the arcane algorithms used to generate Google search results in a Silicon Valley equivalent of Fort Knox. 

Vector embeddings provide a mathematical representation of a word and its context in a multidimensional space. Comparing the locations of two embeddings results in another important data point, known as semantic distance. Two words or phrases closely related to each other have a small semantic distance. If these AI concepts seem arcane, read further to learn how they stand poised to revolutionize internet advertising. 

A Simplified Definition of Vector Embeddings and Semantic Distance

As highlighted above, a vector embedding identifies a word and its context as a location in a 3-dimensional space. The semantic distance between two embeddings effectively describes the closeness of the two concepts. So cat and lion have a small semantic distance, while ice cream and synthesizer boast a large one. 

Why do these concepts matter in the world of generative AI and internet search? Semantic distance plays a key role in a gen AI-powered chatbot being able to generate and understand complex phrases. That improved understanding also enables it to find webpages and documents with content that’s actually relevant to the search query. As an example, the results when searching for glasses used for reading won’t include a variety of fancy beer steins. 

How Vector Embeddings Boast the Potential to Transform Internet Advertising

These simplified definitions of complex concepts like vector embeddings and semantic distance bely their potential power when applied to internet advertising. Notably, any form of customer communication – text chats, verbal conversations, emails, and more is able to collected in a large language model. 

In fact, each customer and their purchase history, topic preferences, demographic information, and the LLM data noted above ultimately generates its own vector embedding, taking the concept of a social graph to another level. In a similar manner as Google, Facebook and its advertising model is also vulnerable because of this AI innovation. 

Current keyword based eCommerce search results and advertisements largely rely on a user’s purchase history. For example, Facebook regularly shows you advertising in its news feed for products recently purchased online that you no longer need. On the other hand, vector embeddings offer the potential to generate advertising based on your actual intent. 

To accomplish this higher level of insight, it takes into account the previously noted large language model containing virtually anything generated online by a user. This might be reviews left on a website, comments made in an email or to a customer service representative (human or chatbot), and even that venerable search history. 

So if a user cancels an order with a company because of a vacation to Arizona, that information potentially gets captured in their model. So if they subsequently search for vacation wear for that trip, the search results reflect their intent for an upcoming visit to the desert landscape of the Southwest United States. Expect improved sales and brand loyalty as a result.  

Determining Intent is the Secret Sauce in Modern Advertising 

Notably, Google currently uses some form of vector embeddings in its search engine. However, remember that the advent of generative AI and large language models provides the ability to represent complex language as a vector embedding in addition to keywords. In essence, it provides the secret sauce to determine user intent. 

One relevant goal of this innovation involves Facebook not showing you an ad for an expensive product you already purchased 5 months ago! Not that the team at Meta cares that much about the repetitive advertising as long as their revenue keeps flowing, but rest assured that their advertisers do. 

Online retailers leveraging this technology gain the ability to profile customers beyond basic demographic information and their purchase history. This potentially includes anything that they posted publicly on the internet, efficiently analyzed in great detail. Blog posts, comments, social media activity, and so much more will make up each person’s own LLM. Expect a higher level of accuracy in targeted advertising – as well as additional data privacy concerns – in the near future!

If your company wants to leverage the new world of advertising powered by AI and vector embeddings, connect with the experts at NineTwoThree. We boast relevant technical expertise in machine learning development and business building to help your venture succeed. Schedule a meeting with us to discuss the potential of partnership.

Tim Ludy
Tim Ludy
Articles From Tim
Subscribe To Our Newsletter