KuaFu: Compressing Long User Behavior into Understanding at Billion Scale

2026-09-25

AI Summary

This study addresses the prohibitive computational costs and compression-induced profile distortion and hallucination inherent in modeling long behavior sequences for billion-scale users. We propose a unified behavior compression framework that treats individual behavior items as minimal units, employing a dual-axis projector to achieve token-level compression requiring only two to four tokens per behavior. Integrated with a four-stage training paradigm and a hierarchical evaluation mechanism, this approach effectively mitigates hallucination generation. Experimental results demonstrate that the proposed method surpasses uncompressed baselines in performance while increasing GPU throughput by 37%–350% and saving 190 GPUs. Furthermore, online deployment yields a 1.37% increase in Gross Merchandise Volume (GMV), confirming that the framework achieves both highly efficient and faithful user understanding at industrial scale.

Abstract

Conversational agents, generative recommenders, and personalized advertising all rest on one capability: understanding each user from raw behavior. Prevailing industrial practice is task-specific: for each task, a relevant subsequence is extracted from the full history and a dedicated model trained on it. In production it hits two bottlenecks. First, even after filtering, a single-task sequence stays extremely long: content-interest summarization reads several hundred items per user, tens of thousands of tokens once serialized as prompt text. Second, profiles are refreshed routinely: a billion users weekly, roughly 100K QPM in aggregate, which under a fixed GPU budget sets a hard throughput floor. Compression is therefore mandatory, yet truncation or coarse compression can silently distort the profile, introducing four hallucination types (fabrication, omission, date misattribution, broken logic) that, with no way to evaluate the compressed representation itself, surface only as diffuse degradation in downstream metrics. We present KuaFu, a unified behavior-compression layer whose minimal unit is one behavior item. A two-axis projector compresses each item into 2-4 tokens of width 128-256 (about 10x along the token axis, 20x along width; per-item cache 10 KB to 0.5 KB), with fidelity-oriented four-stage training and layered intermediate evaluation. Across four production profiling tasks it matches or exceeds uncompressed single-task production models on all five headline metrics, raises per-GPU throughput by 37%-350%, and saves 190 GPUs. On public benchmarks it nearly always beats prior compressors at the same compression ratio (up to +17.7 EM on out-of-domain MRQA); on RecBench, a 4B model surpasses its 8B counterpart by 1.90 points. KuaFu has run on the Tencent advertising and recommendation platform for ten months, lifting overall GMV by 1.37%.

Problem

• user behavior compression
• long sequence modeling
• hallucination mitigation
• large-scale profiling
• throughput optimization

Innovation

• Behavior Compression
• Two-axis Projector
• Hallucination Mitigation
• Unified Profiling
• Four-stage Training