A Survey of Retrieval Algorithms in Ad and Content Recommendation Systems

TL;DR

This survey addresses how retrieval algorithms underpin both ad targeting and organic content recommendation, contrasting monetized targeting with user-centric retrieval. It centers on the two-tower network as a core architectural approach, detailing its training, inference, and retrieval workflows, as well as its advantages and practical challenges such as cold start and data quality. The paper highlights potential cross-pollination with retrieval-augmented generation and LLM-assisted data augmentation, underscoring the importance of scalable, flexible, and privacy-conscious systems. Overall, it maps the methodological landscape, benchmarks, and design considerations shaping the next generation of retrieval algorithms for both ads and content. The findings have practical impact for engineers building scalable recommender pipelines, as well as policymakers concerned with privacy and ethical data use in personalized systems.

Abstract

This survey examines the most effective retrieval algorithms utilized in ad recommendation and content recommendation systems. Ad targeting algorithms rely on detailed user profiles and behavioral data to deliver personalized advertisements, thereby driving revenue through targeted placements. Conversely, organic retrieval systems aim to improve user experience by recommending content that matches user preferences. This paper compares these two applications and explains the most effective methods employed in each.

Figures (5)

• Figure 1: Inverted Index
• Figure 2: Re-targeting
• Figure 3: Two-Tower Neural Network
• Figure 4: Three Tower Neural Network
• Figure 5: AB Experimentation