In a monumental leap forward for e-commerce personalization, Tao Tian Group unveiled RecGPT, a cutting-edge recommendation model with a hundred billion parameters, at the "Hardcore Youth Technology Festival 4.0" on July 1st. This self-developed technology signals a revolutionary upgrade to the "You Might Like" feature on Taobao's homepage, transforming the way users interact with the platform.
The RecGPT model, rooted in the Taobao Xingchen Large Language Model (LLM), utilizes reinforcement learning to deeply train on user historical behavior data, significantly enhancing its reasoning and analytical capabilities in e-commerce scenarios. This model delves into over a decade of consumer trajectories on the Taobao platform, integrating billions of product text and image information through multimodal cognitive technology, and generating personalized recommendations by combining with external knowledge bases.
RecGPT's prowess in anticipating user needs is evident in its ability to predict changes in family situations based on purchases of baby products. It can then forecast the types of goods needed at different stages, proactively recommending items like walkers and age-appropriate formula as the baby approaches its first birthday. During shopping festivals, the system tailors precise promotional product combinations based on user brand preferences.
A standout feature of this upgrade is the automatic generation of personalized recommendation reasons. Each product recommendation is accompanied by a customized copy, such as "Check out the new trending top toy?" for a popular toy, or "A must-have moisture-proof device for the Hangzhou plum rain season" for dehumidifiers, greatly enhancing user interaction with the recommended content.
This upgrade is a significant application achievement of Tao Tian Group's AIGX technology system, which has been deployed on a large scale across various business lines of Taobao and Tmall. It covers the entire e-commerce business scenario, including indexing, recommendation, bidding, auction, creativity, and data. By reconstructing the recommendation algorithm with large model technology, the platform aims to enhance commercial conversion while creating a shopping experience that better meets user needs, paving a new direction for the technological evolution of the entire e-commerce industry.