BM25 encoding is known to experience drift over time as the number of documents in the corpus increases, potentially impacting the quality of search results. Specifically, as more documents are added, the relevance scoring can become less accurate unless the encodings are updated to reflect the expanded dataset.
Does Pinecone have a mechanism to handle this drift by automatically updating or recalibrating the sparse encodings for BM25? If so, how does it manage this to maintain optimal search performance?