Large-scale language models are excellent at reading the context of text and understanding intent. Applying this to search technology can help users understand what products they want and provide appropriate results. If this is done well, consumers will experience the joy of discovery. Using Gemini, Plateer is able to create a new product search experience by integrating the data analysis environment of Vertex AI.

"In order to provide generative AI search to online shopping services in the form of a SaaS, all data analysis and AI services need to be in the cloud. Vertex AI has the strength of being able to apply and manage many models systematically. In generative AI search, the Imagen model, which connects images to text, plays a big role. This is because advanced search begins when the system accurately understands the product," added Dr. Seo.

The Imagen model on Vertex AI can read the images of products listed on shopping portals. It not only turns text into images but also provides detailed explanations of the contents of the image as text. Groobee uses this to interpret various information such as the shape, color, material, and more about the product, along with existing product information, to provide a clear understanding of the product.

Unfortunately, many images used are often staged for promotional purposes in addition to providing product information. Staged images can cause distortion when processed by artificial intelligence. To improve accuracy, Groobee generated three possible results for each image and compared them to determine the best possible selection. Dr. Seo explained that Imagen’s diverse image analysis became the foundation for broader searches.

"Imagen, among many generative AIs, has a high level of understanding of images and provides rich interpretations. There are almost no errors in basic interpretation, but by comprehensively verifying the three results generated by Imagen, we were able to increase accuracy while looking at the images from a more diverse perspective, which made it possible for even better search results."

Both standardized search terms, such as ‘black one-piece dress’ and more general queries, like ‘Find things I need for my five-year-old daughter’s birthday party,’ will see relevant results come up. Gemini is able to interpret the context of the phrases entered by the user in the search bar and search for product metadata as Imagen understands. Since it performs text and vector searches simultaneously, it is possible to achieve advanced searches while increasing accuracy.

In general, the selection of products appearing in this area is manually selected or extracted based on data analysis and deep learning based on the direction pursued by each trading site. In most cases, product selections are obtained through data queries during low-traffic times daily.

"The new recommendations section is designed to collect all data through BigQuery and bring it to Vertex AI for analysis at any time. All data collected is in Google Cloud, and since separate data transfer is not required, it safely and quickly analyzes and infers to create appropriate results. Above all, the regular AI-based recommendation area configuration is generated immediately upon request, so it can respond appropriately in a shopping environment that requires a quick response."

The learning process was also much smoother than handling previous general deep learning models, as BigQuery data and AI embedding move quickly and intuitively within Vertex AI. Since there is no need to transfer or separately process data, performance has improved, and data processing costs have also been reduced.

The results were rapidly apparent. In particular, the fashion industry benefited significantly as it became easier to find appropriate products by quickly identifying consumers' intentions, needs, and wants. The growing expectation is that a chat-type search service will be possible in the future.

The importance of generative AI search is greatly magnified in the shopping industry. Global distribution companies have already improved their search experience by combining massive data and large-scale language models. The process of finding exactly what you want through intuitive conversation and helping you choose by categorizing a wide range of questions into detailed categories is becoming increasingly prevalent.

Dr. Seo says that AI can tremendously evolve the online commerce experience through Vertex AI and Gemini. This is as the long-standing idea of ​​creating a single flow from the process of registering a product to marketing through appropriate category classification and data analysis is taking a more service-oriented approach.

"As we introduced the Google Cloud AI environment to the Groobee service, we began to see changes in marketing. This is because we confirmed that although the shopping market is still largely handled by people, existing values ​can be further enhanced with AI. Groobee plans to develop a service that helps not only consumers who access the service and purchase products, but also product planners through broader application of AI."