How to improve search to get the best match as the first record?

Plugins Machine Learning
1

Versions (relevant - OpenSearch/Dashboard/Server OS/Browser):

OpenSearch : 2.16.0.0

Describe the issue:
I am trying to create a Neural Search Pipeline so that I can perform Hybrid Search. I want to use an Ollama llm which is deployed on my machine. I performed the following steps:

Step 1. Register model group:

localhost:9200/_plugins/_ml/model_groups/_register

{
  "name": "NLP_model_group",
  "description": "A model group for NLP models",
  "access_mode": "public"
}

Step 2. Register URL:

{
    "persistent": {
        "plugins.ml_commons.trusted_connector_endpoints_regex": [
          "^<Ollama url>/.*$"
        ]
    }
}

Step 3. Access Ctrl:

{
    "persistent": {
        "plugins.ml_commons.connector_access_control_enabled": true
    }
}

Step 4. Remote Model Connector:

{
    "name": "OpenAI Chat Connector",
    "description": "The connector to a remote Ollama model",
    "version": 1,
    "protocol": "http",
    "parameters": {
        "endpoint": "<Ollama URL>",
        "model": "mistral"
    },
    "actions": [
        {
            "action_type": "predict",
            "method": "POST",
            "url": "http://${parameters.endpoint}/api/embeddings",
            "request_body": "{ \"model\": \"${parameters.model}\", \"prompt\": \"${parameters.text}\"}",
            "pre_process_function": "\n    StringBuilder builder = new StringBuilder();\n    builder.append(\"\\\"\");\n    String first = params.text_docs[0];\n    builder.append(first);\n    builder.append(\"\\\"\");\n    def parameters = \"{\" +\"\\\"text\\\":\" + builder + \"}\";\n    return  \"{\" +\"\\\"parameters\\\":\" + parameters + \"}\";",
            "post_process_function": "\n   def name = \"embedding\";\n      def dataType = \"FLOAT32\";\n      if (params.embedding == null || params.embedding.length == 0) {\n          return null;\n      }\n      def shape = [params.embedding.length];\n      def json = \"{\" +\n                 \"\\\"name\\\":\\\"\" + name + \"\\\",\" +\n                 \"\\\"data_type\\\":\\\"\" + dataType + \"\\\",\" +\n                 \"\\\"shape\\\":\" + shape + \",\" +\n                 \"\\\"data\\\":\" + params.embedding +\n                 \"}\";\n      return json;\n    "
        }
    ]
}

Step 5. Register model to the model group:

{
    "name": "Ollama model",
    "function_name": "remote",
    "model_group_id": "<group id>",
    "description": "test model",
    "connector_id": "<connector id>"
}

Step 6. Deploy the model:

localhost:9200/_plugins/_ml/models/model id/_deploy

Step 7. Create NLP Pipeline:

localhost:9200/_ingest/pipeline/nlp-ingest-pipeline

{
  "description": "An new NLP ingest pipeline",
  "processors": [
    {
      "text_embedding": {
        "model_id": "<model id>",
        "field_map": {
          "text": "embedding"
        }
      }
    }
  ]
}

Step 8. Create NLP Index:

localhost:9200/<index-name>
{
  "settings": {
    "index.knn": true,
    "default_pipeline": "nlp-ingest-pipeline"
  },
  "mappings": {
    "properties": {
      "id": {
        "type": "text"
      },
      "embedding": {
        "type": "knn_vector",
        "dimension": 4096,
        "method": {
          "engine": "lucene",
          "space_type": "l2",
          "name": "hnsw",
          "parameters": {}
        }
      },
      "text": {
        "type": "text"
      }
    }
  }
}

Step 9. Create Post processor pipeline:

localhost:9200/_search/pipeline/nlp-ingest-pipeline

{
  "description": "Post processor for hybrid search",
  "phase_results_processors": [
    {
      "normalization-processor": {
        "normalization": {
          "technique": "min_max"
        },
        "combination": {
          "technique": "arithmetic_mean",
          "parameters": {
            "weights": [
              0.3,
              0.7
            ]
          }
        }
      }
    }
  ]
}

Step 10. Ingest Data:

localhost:9200/<index-name>/_doc/1
{
  "text": "some text ...",
  "id": "1"
}

localhost:9200/<index-name>/_doc/2
{
  "text": "some text ...",
  "id": "2"
}

Step 11: Hybrid Search query:

localhost:9200/<index-name>/_search?search_pipeline=nlp-search-pipeline

{
  "_source": {
    "exclude": [
      "embedding"
    ]
  },
  "query": {
    "hybrid": {
      "queries": [
        {
          "match": {
            "text": {
              "query": "some text"
            }
          }
        },
        {
          "neural": {
            "embedding": {
              "query_text": "some text",
              "model_id": "<model_id>",
              "k": 5
            }
          }
        }
      ]
    }
  }
}

I was able to resolve all the issue and save text as well as embeddings as document in the index. However when I search I get 2 results with best match as second one instead of first. How can I improve on the results to get the best match as the first record?

2

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.