Elasticsearch example for chunked documents to multi-vectors and two different retrieval strategies

chunking-multi-vector.json

DELETE my-long-text-index

PUT my-long-text-index
{
  "mappings": {
    "properties": {
      "my_long_text_field": {
        "type": "nested", //because there can be multiple vectors per doc
        "properties": {
          "vector": {
            "type": "dense_vector" //the vector used for ranking
          },
          "text_chunk": {
            "type": "text" //the text from which the vector was created
          }
        }
      }
    }
  }
}

// We'll search for the vector [5,5]
// Document with the closest chunk
PUT my-long-text-index/_doc/1
{
  "my_long_text_field" : [
    {
      "vector" : [5,4],
      "text_chunk" :  "doc 1 chunk 1"
    },
    {
      "vector" : [5,1],
      "text_chunk" :  "doc 1 chunk 2"
    },
    {
      "vector" : [5,0],
      "text_chunk" :  "doc 1 chunk 3"
    }
  ]
}
// Document with the second and third closest chunk
PUT my-long-text-index/_doc/2
{
  "my_long_text_field" : [
    {
      "vector" : [5,3],
      "text_chunk" :  "doc 2 chunk 1"
    },
    {
      "vector" : [5,2],
      "text_chunk" :  "doc 2 chunk 2"
    },
    {
      "vector" : [5,0],
      "text_chunk" :  "doc 2 chunk 3"
    }
  ]
}
// Document with the closest aggregated chunks
PUT my-long-text-index/_doc/3
{
  "my_long_text_field" : [
    {
      "vector" : [5,1.9],
      "text_chunk" :  "doc 3 chunk 1"
    },
    {
      "vector" : [5,1.8],
      "text_chunk" :  "doc 3 chunk 2"
    },
    {
      "vector" : [5,1.7],
      "text_chunk" :  "doc 3 chunk 3"
    }
  ]
}

GET my-long-text-index/_search
{
  "knn": {
    "field": "my_long_text_field.vector",
    "query_vector": [5,5],
    "inner_hits":{
      "_source": false,
      "fields": [ "my_long_text_field.text_chunk"
        ],
        "size": 1 // Best chunk
    }
  },
  "size": 2, // 2 closest documents
  "_source": false
}

GET my-long-text-index/_search
{
  "knn": {
    "field": "my_long_text_field.vector",
    "query_vector": [5,5],
    "inner_hits":{
      "_source": false,
      "fields": [ "my_long_text_field.text_chunk"
        ],
        "size": 2 // 2 best chunks
    }
  },
  "size": 1, // Best document
  "_source": false
}