← Cases
ChinaTechnology and telecom2024 – ongoing

JOYY

Proprietary moderation and recommendation models on Bigo Live

JOYY said its own recommendation model on Bigo Live lifted average viewing time 2.1% in one quarter

In progressCompany's own claimOfficial ↗

Problem

Bigo Live is a livestreaming platform that must screen large volumes of live content and keep viewers watching. JOYY did not describe the baseline in its release.

Approach

In Q4 2024 JOYY introduced a multimodal content-moderation model, fine-tuned on scenario-specific data and built with third-party large models. It also ran a proprietary AI model that analyzes very long sequences of user behavior to refine recommendations.

Results

2.1%quarter-on-quarter rise in average viewing time per user in Q4 2024, attributed by JOYY to the model plus content and feature improvementsCompany's own claim[1]

Takeaway

JOYY credited the viewing-time gain to the recommendation model together with better content and viewing features, so the model's own share is unknown.

Categories

Evidence
Company's own claim

Sources

  1. JOYY Reports Steady Profit Growth for 2024, Full-Year Buybacks Exceed US$300 Million
    TechNode (PR Newswire Asia)20 Mar 2025Company publication

Source summaries

[1]TechNode (PR Newswire Asia)20 Mar 2025

JOYY's full-year 2024 earnings release says Bigo Live introduced a multimodal moderation model in Q4 and used a proprietary model on long user behavior sequences for recommendations. It reports a 2.1% quarterly rise in average viewing time per user. It gives no moderation figures.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026
Open case page ↗