By default, OpenSearch calculates document scores using the [Okapi BM25](https://en.wikipedia.org/wiki/Okapi_BM25) algorithm. BM25 is a keyword-based algorithm that performs well on queries containing keywords but fails to capture the semantic meaning of the query terms. Semantic search, unlike keyword-based search, takes into account the meaning of the query in the search context. Thus, semantic search performs well when a query requires natural language understanding.
It's helpful to understand the following terms before starting this tutorial:
- _Neural search_: Facilitates vector search at ingestion time and at search time:
- At ingestion time, neural search uses language models to generate vector embeddings from the text fields in the document. The documents containing both the original text field and the vector embedding of the field are then indexed in a k-NN index, as shown in the following diagram.
- At search time, when you then use a _neural query_, the query text is passed through a language model, and the resulting vector embeddings are compared with the document text vector embeddings to find the most relevant results, as shown in the following diagram.
- _Semantic search_: Employs neural search in order to determine the intention of the user's query in the search context, thereby improving search relevance.
You'll find descriptions of all these components as you follow the tutorial, so don't worry if you're not familiar with some of them. Each link in the preceding list will take you to the documentation section for the corresponding component.
## Prerequisites
For this simple setup, you'll use an OpenSearch-provided machine learning (ML) model and a cluster with no dedicated ML nodes. To ensure that this basic local setup works, send the following request to update ML-related cluster settings:
- In production, it's best practice to separate the workloads by having dedicated ML nodes. On clusters with dedicated ML nodes, specify `"only_run_on_ml_node": "true"` for improved performance.
For more information about ML-related cluster settings, see [ML Commons cluster settings]({{site.url}}{{site.baseurl}}/ml-commons-plugin/cluster-settings/).
## Tutorial overview
This tutorial consists of the following steps:
1. [**Set up an ML language model**](#step-1-set-up-an-ml-language-model).
1. [Choose a language model](#step-1a-choose-a-language-model).
1. [Register a model group](#step-1b-register-a-model-group).
1. [Register the model to the model group](#step-1c-register-the-model-to-the-model-group).
1. [Deploy the model](#step-1d-deploy-the-model).
1. [**Ingest data with neural search**](#step-2-ingest-data-with-neural-search).
1. [Create an ingest pipeline for neural search](#step-2a-create-an-ingest-pipeline-for-neural-search).
1. [Create a k-NN index](#step-2b-create-a-k-nn-index).
1. [Ingest documents into the index](#step-2c-ingest-documents-into-the-index).
1. [**Search the data**](#step-3-search-the-data).
- [Search using a keyword search](#search-using-a-keyword-search).
- [Search using a neural search](#search-using-a-neural-search).
Some steps in the tutorial contain optional `Test it` sections. You can ensure that the step was successful by running requests in these sections.
After you're done, follow the steps in the [Clean up](#clean-up) section to delete all created components.
## Tutorial
You can follow this tutorial using your command line or the OpenSearch Dashboards [Dev Tools console]({{site.url}}{{site.baseurl}}/dashboards/dev-tools/run-queries/).
## Step 1: Set up an ML language model
Neural search requires a language model in order to generate vector embeddings from text fields, both at ingestion time and query time.
### Step 1(a): Choose a language model
For this tutorial, you'll use the [DistilBERT](https://huggingface.co/docs/transformers/model_doc/distilbert) model from Hugging Face. It is one of the pretrained sentence transformer models available in OpenSearch that has shown some of the best results in benchmarking tests (for details, see [this blog post](https://opensearch.org/blog/semantic-science-benchmarks/)). You'll need the name, version, and dimension of the model to register it. You can find this information in the [pretrained model table]({{site.url}}{{site.baseurl}}/ml-commons-plugin/pretrained-models/#sentence-transformers) by selecting the `config_url` link corresponding to the model's TorchScript artifact:
- The model name is `huggingface/sentence-transformers/msmarco-distilbert-base-tas-b`.
- The model version is `1.0.1`.
- The number of dimensions for this model is `768`.
- Use any other pretrained model provided by OpenSearch. For more information, see [OpenSearch-provided pretrained models]({{site.url}}{{site.baseurl}}/ml-commons-plugin/pretrained-models/).
- Upload your own model to OpenSearch. For more information, see [Custom local models]({{site.url}}{{site.baseurl}}/ml-commons-plugin/custom-local-models/).
- Connect to a foundation model hosted on an external platform. For more information, see [Connecting to remote models]({{site.url}}{{site.baseurl}}/ml-commons-plugin/remote-models/index/).
For information about choosing a model, see [Further reading](#further-reading).
For access control, models are organized into model groups (collections of versions of a particular model). Each model group name in the cluster must be globally unique. Registering a model group ensures the uniqueness of the model group name.
If you are registering the first version of a model without first registering the model group, a new model group is created automatically. For more information, see [Model access control]({{site.url}}{{site.baseurl}}/ml-commons-plugin/model-access-control/).
{: .tip}
To register a model group with the access mode set to `public`, send the following request:
```json
POST /_plugins/_ml/model_groups/_register
{
"name": "NLP_model_group",
"description": "A model group for NLP models",
"access_mode": "public"
}
```
{% include copy-curl.html %}
OpenSearch sends back the model group ID:
```json
{
"model_group_id": "Z1eQf4oB5Vm0Tdw8EIP2",
"status": "CREATED"
}
```
You'll use this ID to register the chosen model to the model group.
<detailsclosedmarkdown="block">
<summary>
Test it
</summary>
{: .text-delta}
Search for the newly created model group by providing its model group ID in the request:
```json
POST /_plugins/_ml/model_groups/_search
{
"query": {
"match": {
"_id": "Z1eQf4oB5Vm0Tdw8EIP2"
}
}
}
```
{% include copy-curl.html %}
The response contains the model group:
```json
{
"took": 0,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"skipped": 0,
"failed": 0
},
"hits": {
"total": {
"value": 1,
"relation": "eq"
},
"max_score": 1,
"hits": [
{
"_index": ".plugins-ml-model-group",
"_id": "Z1eQf4oB5Vm0Tdw8EIP2",
"_version": 1,
"_seq_no": 14,
"_primary_term": 2,
"_score": 1,
"_source": {
"created_time": 1694357262582,
"access": "public",
"latest_version": 0,
"last_updated_time": 1694357262582,
"name": "NLP_model_group",
"description": "A model group for NLP models"
}
}
]
}
}
```
</details>
### Step 1(c): Register the model to the model group
To register the model to the model group, provide the model group ID in the register request:
Registering a model is an asynchronous task. OpenSearch sends back a task ID for this task:
```json
{
"task_id": "aFeif4oB5Vm0Tdw8yoN7",
"status": "CREATED"
}
```
OpenSearch downloads the config file for the model and the model contents from the URL. Because the model is larger than 10 MB in size, OpenSearch splits it into chunks of up to 10 MB and saves those chunks in the model index. You can check the status of the task by using the Tasks API:
```json
GET /_plugins/_ml/tasks/aFeif4oB5Vm0Tdw8yoN7
```
{% include copy-curl.html %}
Once the task is complete, the task state will be `COMPLETED` and the Tasks API response will contain a model ID for the registered model:
```json
{
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"task_type": "REGISTER_MODEL",
"function_name": "TEXT_EMBEDDING",
"state": "COMPLETED",
"worker_node": [
"4p6FVOmJRtu3wehDD74hzQ"
],
"create_time": 1694358489722,
"last_update_time": 1694358499139,
"is_async": true
}
```
You'll need the model ID in order to use this model for several of the following steps.
<detailsclosedmarkdown="block">
<summary>
Test it
</summary>
{: .text-delta}
Search for the newly created model by providing its ID in the request:
The response contains the model information. You can see that the `model_state` is `REGISTERED`. Additionally, the model was split into 27 chunks, as shown in the `total_chunks` field.
</details>
#### Advanced: Registering a custom model
To register a custom model, you must provide a model configuration in the register request. For example, the following is a register request containing the full format for the model used in this tutorial:
"description": "This is a port of the DistilBert TAS-B Model to sentence-transformers model: It maps sentences & paragraphs to a 768 dimensional dense vector space and is optimized for the task of semantic search.",
Once the model is registered, it is saved in the model index. Next, you'll need to deploy the model. Deploying a model creates a model instance and caches the model in memory. To deploy the model, provide its model ID to the `_deploy` endpoint:
```json
POST /_plugins/_ml/models/aVeif4oB5Vm0Tdw8zYO2/_deploy
```
{% include copy-curl.html %}
Like the register operation, the deploy operation is asynchronous, so you'll get a task ID in the response:
```json
{
"task_id": "ale6f4oB5Vm0Tdw8NINO",
"status": "CREATED"
}
```
You can check the status of the task by using the Tasks API:
```json
GET /_plugins/_ml/tasks/ale6f4oB5Vm0Tdw8NINO
```
{% include copy-curl.html %}
Once the task is complete, the task state will be `COMPLETED`:
```json
{
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"task_type": "DEPLOY_MODEL",
"function_name": "TEXT_EMBEDDING",
"state": "COMPLETED",
"worker_node": [
"4p6FVOmJRtu3wehDD74hzQ"
],
"create_time": 1694360024141,
"last_update_time": 1694360027940,
"is_async": true
}
```
<detailsclosedmarkdown="block">
<summary>
Test it
</summary>
{: .text-delta}
Search for the deployed model by providing its ID in the request:
You can also receive statistics for all deployed models in your cluster by sending a [Models Profile API]({{site.url}}{{site.baseurl}}/ml-commons-plugin/api/profile/) request:
Neural search uses a language model to transform text into vector embeddings. During ingestion, neural search creates vector embeddings for the text fields in the request. During search, you can generate vector embeddings for the query text by applying the same model, allowing you to perform vector similarity search on the documents.
### Step 2(a): Create an ingest pipeline for neural search
Now that you have deployed a model, you can use this model to configure [neural search]({{site.url}}{{site.baseurl}}/search-plugins/neural-search/). First, you need to create an [ingest pipeline]({{site.url}}{{site.baseurl}}/api-reference/ingest-apis/index/) that contains one processor: a task that transforms document fields before documents are ingested into an index. For neural search, you'll set up a `text_embedding` processor that creates vector embeddings from text. You'll need the `model_id` of the model you set up in the previous section and a `field_map`, which specifies the name of the field from which to take the text (`text`) and the name of the field in which to record embeddings (`passage_embedding`):
```json
PUT /_ingest/pipeline/nlp-ingest-pipeline
{
"description": "An NLP ingest pipeline",
"processors": [
{
"text_embedding": {
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"field_map": {
"text": "passage_embedding"
}
}
}
]
}
```
{% include copy-curl.html %}
<detailsclosedmarkdown="block">
<summary>
Test it
</summary>
{: .text-delta}
Search for the created ingest pipeline by using the Ingest API:
```json
GET /_ingest/pipeline
```
{% include copy-curl.html %}
The response contains the ingest pipeline:
```json
{
"nlp-ingest-pipeline": {
"description": "An NLP ingest pipeline",
"processors": [
{
"text_embedding": {
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"field_map": {
"text": "passage_embedding"
}
}
}
]
}
}
```
</details>
### Step 2(b): Create a k-NN index
Now you'll create a k-NN index with a field named `text`, which contains an image description, and a [`knn_vector`]({{site.url}}{{site.baseurl}}/field-types/supported-field-types/knn-vector/) field named `passage_embedding`, which contains the vector embedding of the text. Additionally, set the default ingest pipeline to the `nlp-ingest-pipeline` you created in the previous step:
```json
PUT /my-nlp-index
{
"settings": {
"index.knn": true,
"default_pipeline": "nlp-ingest-pipeline"
},
"mappings": {
"properties": {
"id": {
"type": "text"
},
"passage_embedding": {
"type": "knn_vector",
"dimension": 768,
"method": {
"engine": "lucene",
"space_type": "l2",
"name": "hnsw",
"parameters": {}
}
},
"text": {
"type": "text"
}
}
}
}
```
{% include copy-curl.html %}
Setting up a k-NN index allows you to later perform a vector search on the `passage_embedding` field.
<detailsclosedmarkdown="block">
<summary>
Test it
</summary>
{: .text-delta}
Use the following requests to get the settings and the mappings of the created index:
```json
GET /my-nlp-index/_settings
```
{% include copy-curl.html %}
```json
GET /my-nlp-index/_mappings
```
{% include copy-curl.html %}
</details>
### Step 2(c): Ingest documents into the index
In this step, you'll ingest several sample documents into the index. The sample data is taken from the [Flickr image dataset](https://www.kaggle.com/datasets/hsankesara/flickr-image-dataset). Each document contains a `text` field corresponding to the image description and an `id` field corresponding to the image ID:
```json
PUT /my-nlp-index/_doc/1
{
"text": "A West Virginia university women 's basketball team , officials , and a small gathering of fans are in a West Virginia arena .",
"id": "4319130149.jpg"
}
```
{% include copy-curl.html %}
```json
PUT /my-nlp-index/_doc/2
{
"text": "A wild animal races across an uncut field with a minimal amount of trees .",
"id": "1775029934.jpg"
}
```
{% include copy-curl.html %}
```json
PUT /my-nlp-index/_doc/3
{
"text": "People line the stands which advertise Freemont 's orthopedics , a cowboy rides a light brown bucking bronco .",
"id": "2664027527.jpg"
}
```
{% include copy-curl.html %}
```json
PUT /my-nlp-index/_doc/4
{
"text": "A man who is riding a wild horse in the rodeo is very near to falling off .",
"id": "4427058951.jpg"
}
```
{% include copy-curl.html %}
```json
PUT /my-nlp-index/_doc/5
{
"text": "A rodeo cowboy , wearing a cowboy hat , is being thrown off of a wild white horse .",
"id": "2691147709.jpg"
}
```
{% include copy-curl.html %}
When the documents are ingested into the index, the `text_embedding` processor creates an additional field that contains vector embeddings and adds that field to the document. To see an example document that is indexed, search for document 1:
"text": "A West Virginia university women 's basketball team , officials , and a small gathering of fans are in a West Virginia arena .",
"id": "4319130149.jpg"
}
}
```
## Step 3: Search the data
Now you'll search the index using keyword search, neural search, and a combination of the two.
### Search using a keyword search
To search using a keyword search, use a `match` query. You'll exclude embeddings from the results:
```json
GET /my-nlp-index/_search
{
"_source": {
"excludes": [
"passage_embedding"
]
},
"query": {
"match": {
"text": {
"query": "wild west"
}
}
}
}
```
{% include copy-curl.html %}
Document 3 is not returned because it does not contain the specified keywords. Documents containing the words `rodeo` and `cowboy` are scored lower because semantic meaning is not considered:
<detailsclosedmarkdown="block">
<summary>
Results
</summary>
{: .text-delta}
```json
{
"took": 647,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"skipped": 0,
"failed": 0
},
"hits": {
"total": {
"value": 4,
"relation": "eq"
},
"max_score": 1.7878418,
"hits": [
{
"_index": "my-nlp-index",
"_id": "1",
"_score": 1.7878418,
"_source": {
"text": "A West Virginia university women 's basketball team , officials , and a small gathering of fans are in a West Virginia arena .",
"id": "4319130149.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "2",
"_score": 0.58093566,
"_source": {
"text": "A wild animal races across an uncut field with a minimal amount of trees .",
"id": "1775029934.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "5",
"_score": 0.55228686,
"_source": {
"text": "A rodeo cowboy , wearing a cowboy hat , is being thrown off of a wild white horse .",
"id": "2691147709.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "4",
"_score": 0.53899646,
"_source": {
"text": "A man who is riding a wild horse in the rodeo is very near to falling off .",
"id": "4427058951.jpg"
}
}
]
}
}
```
</details>
### Search using a neural search
To search using a neural search, use a `neural` query and provide the model ID of the model you set up earlier so that vector embeddings for the query text are generated with the model used at ingestion time:
```json
GET /my-nlp-index/_search
{
"_source": {
"excludes": [
"passage_embedding"
]
},
"query": {
"neural": {
"passage_embedding": {
"query_text": "wild west",
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"k": 5
}
}
}
}
```
{% include copy-curl.html %}
This time, the response not only contains all five documents, but the document order is also improved because neural search considers semantic meaning:
<detailsclosedmarkdown="block">
<summary>
Results
</summary>
{: .text-delta}
```json
{
"took": 25,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"skipped": 0,
"failed": 0
},
"hits": {
"total": {
"value": 5,
"relation": "eq"
},
"max_score": 0.01585195,
"hits": [
{
"_index": "my-nlp-index",
"_id": "4",
"_score": 0.01585195,
"_source": {
"text": "A man who is riding a wild horse in the rodeo is very near to falling off .",
"id": "4427058951.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "2",
"_score": 0.015748845,
"_source": {
"text": "A wild animal races across an uncut field with a minimal amount of trees.",
"id": "1775029934.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "5",
"_score": 0.015177963,
"_source": {
"text": "A rodeo cowboy , wearing a cowboy hat , is being thrown off of a wild white horse .",
"id": "2691147709.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "1",
"_score": 0.013272902,
"_source": {
"text": "A West Virginia university women 's basketball team , officials , and a small gathering of fans are in a West Virginia arena .",
"id": "4319130149.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "3",
"_score": 0.011347735,
"_source": {
"text": "People line the stands which advertise Freemont 's orthopedics , a cowboy rides a light brown bucking bronco .",
Hybrid search combines keyword and neural search to improve search relevance. To implement hybrid search, you need to set up a [search pipeline]({{site.url}}{{site.baseurl}}/search-plugins/search-pipelines/index/) that runs at search time. The search pipeline you'll configure intercepts search results at an intermediate stage and applies the [`normalization-processor`]({{site.url}}{{site.baseurl}}/search-plugins/search-pipelines/normalization-processor/) to them. The `normalization-processor` normalizes and combines the document scores from multiple query clauses, rescoring the documents according to the chosen normalization and combination techniques.
To configure a search pipeline with a `normalization-processor`, use the following request. The normalization technique in the processor is set to `min_max`, and the combination technique is set to `arithmetic_mean`. The `weights` array specifies the weights assigned to each query clause as decimal percentages:
"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
]
}
}
}
}
]
}
```
{% include copy-curl.html %}
#### Step 2: Search with the hybrid query
You'll use the [`hybrid` query]({{site.url}}{{site.baseurl}}/query-dsl/compound/hybrid/) to combine the `match` and `neural` query clauses. Make sure to apply the previously created `nlp-search-pipeline` to the request in the query parameter:
```json
GET /my-nlp-index/_search?search_pipeline=nlp-search-pipeline
{
"_source": {
"exclude": [
"passage_embedding"
]
},
"query": {
"hybrid": {
"queries": [
{
"match": {
"text": {
"query": "cowboy rodeo bronco"
}
}
},
{
"neural": {
"passage_embedding": {
"query_text": "wild west",
"model_id": "aVeif4oB5Vm0Tdw8zYO2",
"k": 5
}
}
}
]
}
}
}
```
{% include copy-curl.html %}
Not only does OpenSearch return documents that match the semantic meaning of `wild west`, but now the documents containing words related to the wild west theme are also scored higher relative to the others:
<detailsclosedmarkdown="block">
<summary>
Results
</summary>
{: .text-delta}
```json
{
"took": 27,
"timed_out": false,
"_shards": {
"total": 1,
"successful": 1,
"skipped": 0,
"failed": 0
},
"hits": {
"total": {
"value": 5,
"relation": "eq"
},
"max_score": 0.86481035,
"hits": [
{
"_index": "my-nlp-index",
"_id": "5",
"_score": 0.86481035,
"_source": {
"text": "A rodeo cowboy , wearing a cowboy hat , is being thrown off of a wild white horse .",
"id": "2691147709.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "4",
"_score": 0.7003,
"_source": {
"text": "A man who is riding a wild horse in the rodeo is very near to falling off .",
"id": "4427058951.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "2",
"_score": 0.6839765,
"_source": {
"text": "A wild animal races across an uncut field with a minimal amount of trees.",
"id": "1775029934.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "3",
"_score": 0.3007,
"_source": {
"text": "People line the stands which advertise Freemont 's orthopedics , a cowboy rides a light brown bucking bronco .",
"id": "2664027527.jpg"
}
},
{
"_index": "my-nlp-index",
"_id": "1",
"_score": 0.29919013,
"_source": {
"text": "A West Virginia university women 's basketball team , officials , and a small gathering of fans are in a West Virginia arena .",
"id": "4319130149.jpg"
}
}
]
}
}
```
</details>
Instead of specifying the search pipeline in every request, you can set it as a default search pipeline for the index as follows:
You can now experiment with different weights, normalization techniques, and combination techniques. For more information, see the [`normalization-processor`]({{site.url}}{{site.baseurl}}/search-plugins/search-pipelines/normalization-processor/) and [`hybrid` query]({{site.url}}{{site.baseurl}}/query-dsl/compound/hybrid/) documentation.
You can parameterize the search by using search templates. Search templates hide implementation details, reducing the number of nested levels and thus the query complexity. For more information, see [search templates]({{site.url}}{{site.baseurl}}/search-plugins/search-template/).
- Read about the basics of OpenSearch semantic search in [Building a semantic search engine in OpenSearch](https://opensearch.org/blog/semantic-search-solutions/).
- Read about the benefits of combining keyword and neural search, the normalization and combination technique options, and benchmarking tests in [The ABCs of semantic search in OpenSearch: Architectures, benchmarks, and combination strategies](https://opensearch.org/blog/semantic-science-benchmarks/).
