An end-t-end approach to dertermining high-quality digital content recommendations
Topics: Microsoft, News and Discover, Ranking, Scoring, Search Query Processing
This Microsoft patent application describes an innovative, end-to-end method for dramatically improving the quality of digital content recommendations. The core concept involves leveraging a Large Language Model (LLM) to evaluate the relevance of content recommendations generated by a traditional machine learning model in response to a user’s search query. The LLM produces a relevance score, which is then used as a high-quality training signal (or “ground truth”) to refine and update the original recommendation engine, ensuring it learns to deliver better, more accurate results.
