Recommendation vs Advertising vs Search: Differences in Goals, Optimization Targets, and Model Design

This article, part of Wang Zhe’s Machine Learning Notes, summarizes the author’s four‑year experience in advertising systems and five‑year experience in recommendation systems, and his understanding of search algorithms, to clarify the distinctions among these three core internet applications.

At the most fundamental level, advertising algorithms aim directly at increasing company revenue, recommendation algorithms target higher user engagement to indirectly boost revenue, and search algorithms focus on efficiently retrieving information relevant to the user’s query.

Consequently, their optimization goals differ: advertising optimizes precise CTR and CVR estimation; recommendation optimizes engagement‑related metrics such as watch time, CTR, AUC, gAUC, MAP, and emphasizes list‑wise ranking; search optimizes recall‑oriented metrics like recall rate, MAP, and NDCG, emphasizing correct answer retrieval.

These goal differences lead to distinct model design emphases: advertising relies on point‑wise models with strong calibration to predict exact CTR/CVR; recommendation employs point‑wise, pair‑wise, list‑wise models, incorporates diversity, freshness, and exploration‑exploitation strategies; search uses dual‑tower structures, heavy cross‑features, and NLP models to understand queries and content.

Auxiliary strategies also vary: advertising systems integrate pacing, bidding, budget control, and ad allocation; recommendation systems add exploration, long‑tail content handling, and diversity constraints; search systems prioritize query understanding and multimodal content embedding.

System‑level pain points differ as well: advertising faces complex module coordination for profit maximization; recommendation struggles with balancing short‑term and long‑term user interests; search contends with multimodal content understanding beyond textual matching.

The author invites readers to follow his WeChat public account for further discussions on computational advertising, recommendation systems, and related machine‑learning topics.