Learning to Hybrid Search: Combining BM25, Neural Embeddings, and Customer Behavior

Explore the intricacies of hybrid search techniques in this insightful conference talk from Haystack US 2023. Delve into the world of combining BM25, neural embeddings, and customer behavior to create an ultimate ranking ensemble. Discover how traditional term search, neural search, and Learning to Rank (LTR) approaches can be integrated to overcome their individual limitations. Examine the strengths and weaknesses of each method, including the precision of term search, the semantic understanding of neural search, and the adaptability of LTR to customer behavior. Through an e-commerce search example using Amazon's ESCI dataset, compare the effectiveness of various search approaches on real data. Learn how combining multiple techniques in a hybrid system can yield superior results compared to individual methods. Gain valuable insights from Roman Grebennikov, a Principal Engineer at DeliveryHero SE, as he shares his expertise in search personalization, recommendations, and performance engineering.