Hybrid vigor: a winning way to use hybrid search

trec_tok_vec

PUT trec
{
  "mappings": {
    "properties": {
      "title" : {"type": "text"},
      "episode": {"type": "text"},
      "text": {"type": "text"},
      "embed": {"type": "dense_vector", "dims": 3}
    }
  }
}


POST trec/_doc
{
  "title": "Great Steaks",
  "episode": "T-bones FTW",
  "text": "When it comes down to it, in my option, the best cut of steak in the cow is the T-Bone! You can't beat the juicy tenderness",
  "embed": [3.0, 2.1, 0.2]
}


POST trec/_doc
{
  "title": "Beef Eater Anonymos",
  "episode": "Why red meat is actually good for you",
  "text": "Sure a steak tasts great, but is it also great for you? I think it is and my guess Teddy Steaks here on the show today say YES!",
  "embed": [2.0, 2.9, 0.1]
}


GET trec/_search // first we gotta get the maxScore
{
  "size": 1,
  "query": {
    "multi_match": {
      "fields": [
        "title",
        "episode",
        "text^3"
      ],
      "query": "tasty steaks"
    }
  }
}


GET trec/_search
{
  "query": {
    "script_score": {
      "query": {
        "multi_match": {
          "fields": [
            "title",
            "episode",
            "text^3"
          ],
          "query": "tasty steaks"
        }
      },
      "script": {
        "params": {
          "maxScore": 2.064089,
          "queryVector": [3.1, 2.5, 0.1]
        },
        "source": """
        double myScore = _score / params.maxScore;
        double sbertScore = cosineSimilarity(params.queryVector, 'embed');
        return myScore + sbertScore;
        """
      }
    }
  }
}

DELETE trec