At the "Hardcore Youth Tech Festival 4.0" event held on July 1st, Taobao Group officially launched its independently developed recommendation large model with hundreds of billions of parameters, RecGPT. This innovation marks a revolutionary upgrade to the "Guess What You Like" feature on the Taobao homepage. This change is mainly attributed to the application of generative recommendation (AIGR) technology, significantly improving the accuracy of personalized recommendations.
According to official data, the recommendation system equipped with RecGPT has achieved double-digit growth in user clicks, while user add-to-cart behavior and page dwell time have also increased by more than 5%. This technological upgrade not only marks an important step for e-commerce platforms in the field of personalized recommendations, but also provides users with a more tailored shopping experience.
The core of RecGPT is based on the Taobao Xingchen LLM large model, which conducts in-depth training on users' historical behavior data within the platform through reinforcement learning, thus significantly enhancing its reasoning and analytical capabilities in e-commerce scenarios. The model can deeply mine users' consumption trajectories on Taobao over the past decade, and integrate the image and text information of hundreds of millions of products using multimodal cognitive technology, combining external knowledge bases to generate personalized recommendation content.
In practical applications, RecGPT demonstrates its advanced ability to anticipate user needs. For example, when the system identifies that a user has purchased baby products, it can infer changes in family situations and then predict the required products at different stages. When a baby is about to turn one year old, the system will proactively recommend items such as walking frames and age-appropriate milk powder; during shopping festivals, it will provide precise promotional product combinations based on the user's brand preferences.
Another highlight of this upgrade is the automatic generation of personalized recommendation reasons. Each item's recommendation feed is equipped with customized recommendation copy, such as labeling popular toys with "The new top trend, don't you want to check it out?" or adding regionalized tips like "Anti-moisture magic tool for Hangzhou's plum rain season" for dehumidifiers, greatly enhancing the user's interaction experience with the recommended content.
This upgrade of the recommendation system is an important application result of Taobao Group's AIGX technology system. This system has already been scaled and deployed across multiple business lines on Taobao and Tmall, covering the entire e-commerce business scenarios including indexing, recommendation, bidding, auction, creativity, and data. By reconstructing the recommendation algorithm with large model technology, the platform is expected to improve commercial conversion while creating a more user-centric shopping experience, opening up new directions for the technological evolution of the entire e-commerce industry.