PushDualGen: Enabling LLMs to Generate Semantic IDs with Interpretable Copy for Industrial Push Recommendation

2026-08-08

AI Summary

This work addresses the black-box nature of large language models in push recommendation systems, where the generation of semantic IDs lacks interpretability, hindering traceability of recommendation logic and limiting industrial deployment. To overcome this, the authors propose PushDualGen, a lightweight generative framework that introduces an interpretable mechanism into industrial-scale push systems for the first time. The approach employs a two-stage architecture: it first generates semantic IDs and then appends optional, skippable explanatory copies, thereby enhancing transparency with negligible additional inference cost. Online A/B experiments demonstrate that PushDualGen achieves a relative 8.50% improvement in effective play rate, reduces user dissatisfaction by 37.70%, and significantly boosts exposure of long-tail videos.

Abstract

Push recommendation in KuaiShou proactively delivers personalized content to nearly one billion users to facilitate their engagement. Recently, generative recommendation has achieved end-to-end user personalization through semantic ID. However, their black- box characteristics make recommendation logics difficult to trace, hindering their deployment. OneRec-Thinking addresses this by incorporating CoT before generating SIDs, but this significantly increases inference cost. To support large-scale industrial applications, we propose PushDualGen, a lightweight generator, which first generates the SID and then produces a copy as a skippable explanation. PushDualGen has been deployed in Kuaishou's push recommendation system. Online A/B tests demonstrate the effectiveness of PushDualGen, delivering significant improvements in both user attraction and satisfaction. The effective play rate for videos recommended to users has relatively increased by 8.50%, while the dissatisfaction rate has relatively fallen by 37.70%. In the long term, PushDualGen optimises the content ecosystem, providing more exposure for long-tail videos.

Problem

• generative recommendation
• semantic ID
• interpretability
• push recommendation
• black-box model

Innovation

• PushDualGen
• semantic ID
• interpretable copy
• generative recommendation
• industrial push recommendation