How We Enhanced the MIND Multi‑Interest Model for Short‑Video Recommendations

This article analyzes the original MIND recommendation model, identifies its limitations in capsule design and routing, proposes three practical improvements—including max‑min and Markov‑based capsule initialization and a sparse routing scheme—and validates the gains with offline experiments and real‑world business metrics.

AIMINDcapsule network

0 likes · 20 min read

How We Enhanced the MIND Multi‑Interest Model for Short‑Video Recommendations

vivo Internet Technology

Oct 25, 2023 · Backend Development

Optimizing Dubbo Routing and Load Balancing at Scale: Vivo's Practice

Vivo tackled high CPU overhead in large‑scale Dubbo deployments by disabling unused routers, caching routing results with BitMap intersections and epoch validation, optimizing weight calculations, and adding a grouping router, which together delivered over 100 % TPS gains for 20 k+ providers and cut CPU usage by roughly 27 %.

Backend ArchitectureDubboMicroservices

0 likes · 18 min read

Optimizing Dubbo Routing and Load Balancing at Scale: Vivo's Practice