On the Role of Language Representations in Auto-Bidding: Findings and Implications

2026-05-07

AI Summary

Existing automated bidding systems struggle to express high-level strategic intent and lack semantic controllability in their numerical state representations. To address this limitation, this work proposes SemBid, a novel framework that systematically investigates the role and boundaries of language-based representations in automated bidding. SemBid encodes task specifications, historical context, and strategic guidance from large language models into discrete tokens, which are injected into offline bidding trajectories. Crucially, it introduces a deep semantic-numerical fusion mechanism that leverages self-attention—rather than simple concatenation—to effectively steer bidding decisions with high-level semantics. Experimental results demonstrate that SemBid consistently outperforms existing methods across diverse scenarios and budget constraints, achieving significant improvements in overall performance, constraint satisfaction rates, and robustness.

Abstract

Auto-bidding is a crucial task in real-time advertising markets, where policies must optimize long-horizon value under delivery constraints (e.g., budget and CPA). Existing methods for auto-bidding rely on compact numerical state representations: while they can implicitly capture delivery dynamics, they offer limited support for explicitly representing and controlling high-level intent, evolving feedback, and operator-style strategic guidance in real campaigns. Meanwhile, Large Language Models (LLMs) offer a powerful method for encoding semantic information, it remains unclear when LLMs help and how to integrate them without sacrificing numerical precision. Through systematic preliminary studies, we find that (1) LLM embeddings contain bidding-relevant cues yet cannot replace numerical features, and (2) gains emerge only with careful semantic--numeric integration rather than naive concatenation. Motivated by these findings, we propose \textit{SemBid}, a novel auto-bidding framework that injects LLM-encoded semantics into offline bidding trajectories at the token level. SemBid introduces three semantic inputs: \textit{Task}, \textit{History}, and \textit{Strategy}. It injects these semantics as tokens alongside numerical trajectory tokens and uses self-attention to integrate them, improving controllability and generalization across objectives. Across diverse scenarios and budget regimes, SemBid outperforms competitive baselines from offline RL and generative sequence modeling, with more consistent gains in overall performance, constraint satisfaction, and robustness. Our code is available at: \href{https://github.com/AlanYu04/SemBid-KDD2026}{\textcolor{blue}{here}}.

Problem

• auto-bidding
• language representations
• semantic integration
• real-time advertising
• Large Language Models

Innovation

• SemBid
• LLM-enhanced auto-bidding
• semantic-numeric integration
• offline reinforcement learning
• token-level injection