Preprint reports AI-evolved recommendation pipeline lifted Tencent click-through rate

Researchers report in a preprint that a recommendation pipeline evolved by AI agents increased online click-through rate by a statistically significant 0.40% in deployment on Tencent's large-scale commercial media platform. The deployed system also improved offline Hit@1, a measure of first-choice recommendation accuracy, by 12.16% relative to its comparison. The framework, TGR-EvoDAG, maintains multiple branches of code changes and merges complementary improvements. It plans around different retraining costs across pipeline stages and adapts downstream components so promising upstream changes are not discarded after temporary regressions. Separate public-dataset tests reported offline gains of up to 37.4% over the strongest baseline; those results are distinct from the commercial deployment.