Pinecone Systems Inc., a machine learning (ML) search infrastructure company, announced the release of a keyword-aware semantic search solution that enables accessible and advanced combination of semantic and keyword search results. “Vector search” allows companies to provide relevant results based on semantic, or similar meanings, as opposed to simple keyword-based searches. At the same time, keywords still matter in searches involving uncommon words like names or industry-specific terms. With few exceptions, companies have to choose between semantic search and keyword search, or running both systems in parallel.
Neither of these options is ideal. When companies choose one or the other, the results are not as complete as they could be, and when they run both systems in parallel and try to combine the results, cost and complexity goes up significantly. This technology can search across two data types — “dense vectors” generated by ML models to represent meaning, and “sparse vectors” generated by traditional keyword-ranking models such as BM25 — before automatically fusing everything into one ranked list of the most relevant results. The Pinecone hybrid search feature is available in beta.