Comparing semantic representations (embeddings) to find similar content without reprocessing raw data.