Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real-time Ad retrieval via LLM-generative Commercial Intention for Sponsored Search Advertising

Tongtong Liu, Zhaohui Wang, Meiyue Qin, Zenghui Lu, Xudong Chen, Yuekui Yang, Peng Shu

TL;DR

This paper tackles real-time sponsored-search ad retrieval by replacing heavy DocIDs with LLM-generated Commercial Intentions (CIs) as semantic tokens. It introduces RARE, an end-to-end architecture featuring knowledge-injected, format-finetuned LLMs, a CI-Ads inverted index, and constrained decoding plus caching to achieve millisecond-scale real-time retrieval. Online deployment across large-scale systems yields meaningful gains in consumption, GMV, CTR, and conversions, while offline experiments show superiority over multiple baselines in HR@500, MAP, and ACR. The work demonstrates the practical viability of using generated commercial intents to bridge queries and ads at scale, with robust ablations and real-world online validation supporting its effectiveness.

Abstract

The integration of Large Language Models (LLMs) with retrieval systems has shown promising potential in retrieving documents (docs) or advertisements (ads) for a given query. Existing LLM-based retrieval methods generate numeric or content-based DocIDs to retrieve docs/ads. However, the one-to-few mapping between numeric IDs and docs, along with the time-consuming content extraction, leads to semantic inefficiency and limits scalability in large-scale corpora. In this paper, we propose the Real-time Ad REtrieval (RARE) framework, which leverages LLM-generated text called Commercial Intentions (CIs) as an intermediate semantic representation to directly retrieve ads for queries in real-time. These CIs are generated by a customized LLM injected with commercial knowledge, enhancing its domain relevance. Each CI corresponds to multiple ads, yielding a lightweight and scalable set of CIs. RARE has been implemented in a real-world online system, handling daily search volumes in the hundreds of millions. The online implementation has yielded significant benefits: a 5.04% increase in consumption, a 6.37% rise in Gross Merchandise Volume (GMV), a 1.28% enhancement in click-through rate (CTR) and a 5.29% increase in shallow conversions. Extensive offline experiments show RARE's superiority over ten competitive baselines in four major categories.

Real-time Ad retrieval via LLM-generative Commercial Intention for Sponsored Search Advertising

TL;DR

This paper tackles real-time sponsored-search ad retrieval by replacing heavy DocIDs with LLM-generated Commercial Intentions (CIs) as semantic tokens. It introduces RARE, an end-to-end architecture featuring knowledge-injected, format-finetuned LLMs, a CI-Ads inverted index, and constrained decoding plus caching to achieve millisecond-scale real-time retrieval. Online deployment across large-scale systems yields meaningful gains in consumption, GMV, CTR, and conversions, while offline experiments show superiority over multiple baselines in HR@500, MAP, and ACR. The work demonstrates the practical viability of using generated commercial intents to bridge queries and ads at scale, with robust ablations and real-world online validation supporting its effectiveness.

Abstract

The integration of Large Language Models (LLMs) with retrieval systems has shown promising potential in retrieving documents (docs) or advertisements (ads) for a given query. Existing LLM-based retrieval methods generate numeric or content-based DocIDs to retrieve docs/ads. However, the one-to-few mapping between numeric IDs and docs, along with the time-consuming content extraction, leads to semantic inefficiency and limits scalability in large-scale corpora. In this paper, we propose the Real-time Ad REtrieval (RARE) framework, which leverages LLM-generated text called Commercial Intentions (CIs) as an intermediate semantic representation to directly retrieve ads for queries in real-time. These CIs are generated by a customized LLM injected with commercial knowledge, enhancing its domain relevance. Each CI corresponds to multiple ads, yielding a lightweight and scalable set of CIs. RARE has been implemented in a real-world online system, handling daily search volumes in the hundreds of millions. The online implementation has yielded significant benefits: a 5.04% increase in consumption, a 6.37% rise in Gross Merchandise Volume (GMV), a 1.28% enhancement in click-through rate (CTR) and a 5.29% increase in shallow conversions. Extensive offline experiments show RARE's superiority over ten competitive baselines in four major categories.

Paper Structure

This paper contains 33 sections, 8 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Ad Retrieval.
  4. Generative-based LLMs Retriever.
  5. Semantic DocIDs.
  6. Method
  7. An End-to-end Generative Architecture
  8. Indexing.
  9. Retrieval.
  10. Customized LLM
  11. Stage 1: Knowledge Injection.
  12. Stage 2: Format Fine-Tuning.
  13. Indexing
  14. Commercial Intentions (CIs).
  15. Ad Indexing Building.
  16. ...and 18 more sections

Figures (5)

  • Figure 1: The Real-time LLM-Generative Ad Retrieval framework (RARE) processes user queries by generating commercial intentions (CIs) through LLM/caching, which are subsequently used to retrieve ads from the dynamic index. The customized LLM are created by injecting knowledge and learning rules based on vanilla LLM.
  • Figure 2: Comparison of RARE and Traditional Retrieval Methods. The Direct Generation of Candidate Ads from User Queries Shortens Link Structure.
  • Figure 3: Constrained Beam Search Decoding Process.
  • Figure 4: Time Consumption for Different Lengths.
  • Figure 5: RARE Outperforms Online Benchmark Models Across Major Real-World Industries.