4 ms·

You index an embedding for each sentence and match against sentence embeddings, then rank documents based on highest scoring sentence matches.

• petulla 7y ago
  Seems problematic if the term entities/topics are spread across sentences.
  • m_ke 7y ago
    It really depends on your application but you can come up with a lot of different aggregation schemes to rerank top results based on the whole document.