Yahoo Enhances Search Retargeting with Amazon Bedrock AI Models

Yahoo has upgraded its Search Retargeting (SRT) system within the Yahoo DSP ad tech suite by leveraging Amazon Bedrock's serverless access to advanced foundation models. SRT is a core tool that helps advertisers reach users based on their search behavior, both on Yahoo and across partner networks. Previously, Yahoo's keyword expansion relied on Word2Vec embeddings combined with locality-sensitive hashing, which had limitations such as outdated vocabulary and limited semantic understanding. To address these issues, Yahoo adopted generative AI models through Amazon Bedrock, experimenting with models like Amazon Nova, Meta Llama, and Anthropic’s Claude. The selected model, Claude 3.5 Sonnet v2, demonstrated superior capabilities in generating relevant, high-quality keyword expansions, increasing the reach and precision of targeted advertising. This new approach involves a verification step where expanded keywords are filtered based on semantic similarity and sensitivity checks, ensuring relevance and compliance. The integration of large language models has resulted in a fivefold increase in keyword expansion ratio and up to five times larger audience reach, translating into improved campaign performance and conversions. Amazon Bedrock's serverless architecture allowed Yahoo to experiment rapidly with multiple models without managing infrastructure, simplifying upgrades and scaling. This enhancement positions Yahoo DSP to deliver more effective advertising solutions, leveraging AI to better connect search intent with display, video, and native ads. The move underscores the growing importance of generative AI in digital advertising, offering advertisers more precise targeting tools and improving overall system efficiency.