AI Search: Improving Exact Match Scoring in Document Search

Question

AI Search: Improving Exact Match Scoring in Document Search

Shruti 40

Building a search functionality on a dataset containing approximately 65,000,000 documents and 30 GB of data, with fields such as name, address, city, telephone number, and postcode, along with filter fields like country and type.

The challenge lies in the scoring system where exact matches receive lower scores compared to documents that contain all search terms alongside additional terms.

For example, when searching for Microsoft Inc using the following query:

  {
	 "query": "Microsoft Inc",
    "options": {
        "includeTotalCount": true,
        "queryType": "FULL",
        "searchMode": "ALL",
        "filter": "(countryCode eq USA)",
        "searchFields": ["organisationName"],
        "scoringProfile": "default"
    }
}

The expected results are:

"Microsoft Inc."
"Microsoft Inc. US"
"Microsoft Corporation"

However, "Microsoft Inc. US" is getting a higher score than "Microsoft Inc." despite being an exact match.

A scoring profile has been added, but it hasn't resolved the issue:

{
  "scoringProfiles": [
    {
      "name": "default",
      "functionAggregation": "firstMatching",
      "text": {
        "weights": {
          "organisationName": 5
        }
      },
      "functions": []
    }
  ]
}

The field definition is as follows:

{
  "name": "organisationName",
  "type": "Edm.String",
  "searchable": true,
  "filterable": false,
  "retrievable": true,
  "stored": true,
  "sortable": false,
  "facetable": false,
  "key": false,
  "analyzer": "diacritics_analyzer",
  "synonymMaps": []
}

The scoring configuration in use is:

"@odata.type": "#Microsoft.Azure.Search.BM25Similarity"

How can the scoring for exact matches be improved in this scenario?

Shree Hima Bindu Maganti 4,155 Reputation points Microsoft External Staff Moderator

2025-04-24T12:19:40.2633333+00:00

Hi @Shruti
To improve the scoring for exact matches in your Azure AI Search implementation,

Modify your scoring profile to give higher weight to exact matches by adding a function that boosts documents matching the query exactly.

Utilize term boosting in your query to prioritize exact matches. For example, use the syntax Microsoft Inc^2 to give a higher score to documents containing this exact phrase.

If default scoring is insufficient, implement custom scoring functions that evaluate documents based on criteria like exact matches in the organisationName field.

Ensure the analyzer for the organisationName field is suitable. Test different analyzers to see if they improve exact match results.

Experiment with different query types (e.g., simple, full) to see if they favor exact matches in scoring.

By implementing these strategies, you should enhance the scoring system to better prioritize exact matches in your search results.
Full text search in Azure AI Search

Add scoring profiles to boost search scores
let mi know if you have any further queries.

1 answer

Your answer

Shree Hima Bindu Maganti 4,155 Reputation points Microsoft External Staff Moderator

2025-04-24T12:19:40.2633333+00:00

Hi @Shruti
To improve the scoring for exact matches in your Azure AI Search implementation,

Modify your scoring profile to give higher weight to exact matches by adding a function that boosts documents matching the query exactly.

Utilize term boosting in your query to prioritize exact matches. For example, use the syntax Microsoft Inc^2 to give a higher score to documents containing this exact phrase.

If default scoring is insufficient, implement custom scoring functions that evaluate documents based on criteria like exact matches in the organisationName field.

Ensure the analyzer for the organisationName field is suitable. Test different analyzers to see if they improve exact match results.

Experiment with different query types (e.g., simple, full) to see if they favor exact matches in scoring.

By implementing these strategies, you should enhance the scoring system to better prioritize exact matches in your search results.
Full text search in Azure AI Search

Add scoring profiles to boost search scores
let mi know if you have any further queries.

Answer 1

Hi Shruti,

To address this, introduce a new field in the index specifically designed for exact matches. Define this field using a custom analyzer with a keyword tokenizer. That way, the entire phrase “Microsoft Inc” is treated as a single token and not split into "Microsoft" and "Inc".

"analyzers": [
  {
    "name": "keyword_analyzer",
    "tokenizer": "keyword",
    "filters": [
      "lowercase"
    ]
  }
]

And the field definition:

{
  "name": "organisationName_exact",
  "type": "Edm.String",
  "searchable": true,
  "analyzer": "keyword_analyzer",
  "filterable": false,
  "retrievable": true,
  "sortable": false,
  "facetable": false,
  "key": false
}

When indexing documents, check the organisationName_exact field contains a lowercased version of the original name. For example, if the value is “Microsoft Inc”, the organisationName_exact field should store “microsoft inc”.

Now, in the search query, continue using the regular organisationName field for general term matching, but also introduce a scoring profile that gives a significant boost when the exact phrase appears in the organisationName_exact field.

{
  "search": "Microsoft Inc",
  "queryType": "full",
  "searchMode": "all",
  "searchFields": "organisationName",
  "filter": "countryCode eq 'USA'",
  "scoringProfile": "exactMatchBoost",
  "scoringParameters": [
    "searchText='Microsoft Inc'"
  ]
}

And the scoring profile:

"scoringProfiles": [
  {
    "name": "exactMatchBoost",
    "functions": [
      {
        "type": "magnitude",
        "fieldName": "organisationName_exact",
        "boost": 10,
        "interpolation": "constant",
        "parameters": {
          "values": ["Microsoft Inc"]
        }
      }
    ],
    "functionAggregation": "sum"
  }
]

This way, BM25 will continue to work for overall relevance, but the exact match will receive a strong score boost, helping “Microsoft Inc” appear above similar but less precise matches like “Microsoft Inc. US.”

Hope it helps!

Please do not forget to click "Accept the answer” and Yes wherever the information provided helps you, this can be beneficial to other community members.

User's image

If you have any other questions or still running into more issues, let me know in the "comments" and I would be happy to help you.

Suresh Chikkam 1,560 Reputation points Microsoft External Staff Moderator

2025-04-30T00:59:33.59+00:00

Hi Shruti,

Following up to see if the above answer was helpful. If this answers your query, do click Accept Answer and Yes, if you have any further query do let us know.

Share via

AI Search: Improving Exact Match Scoring in Document Search

1 answer

Your answer