# Analytics & Query Suggestions

Typesense can aggregate search queries for both analytics purposes and for query suggestions.

# Enabling the feature

# When Self-Hosting

The Search analytics feature needs to be explicitly enabled via the --enable-search-analytics and --analytics-dir flags for query suggestions and other analytics features to work.

./typesense-server --data-dir=/path/to/data --api-key=abcd \
  --enable-search-analytics=true \
  --analytics-dir=/path/to/analytics-data \ 
  --analytics-flush-interval=60

The --analytics-flush-interval flag determines how often the search query aggregations are persisted to the suggestion collection.

Set this to a smaller value (minimum value is 60 seconds) to get more frequent updates to the suggestion collection. Default value is: 3600 seconds (every hour).

# On Typesense Cloud

We automatically set --enable-search-analytics=true and --analytics-flush-interval=300 (every 5 minutes) in Typesense Cloud (more context in the section above).

# Disabling for specific queries

To disable analytics aggregation for specific search queries (for e.g. those originating from a test script), you can set enable_analytics: false as a search query parameter.

# Query suggestions

You can capture the search queries that are happening in the system and use that to either track the popular queries or power an autosuggest feature with them.

# Create a collection for queries

Let's first create a new collection to which search queries will be aggregated and logged to.

This collection is just like any other Typesense collection, except that it is automatically populated by Typesense with search queries that were sent to other collections.

The q and count fields are mandatory.

# Create an analytics rule

We can now create a popular_queries analytics rule that stores the most frequently occurring search queries in the collection we created above. We limit the popular queries to the top 1000 queries via the limit parameter.

That's it!

Search queries will now be aggregated and stored in the product_queries (destination) collection, whenever you make a search on the products (source) collection.

Queries will be aggregated and sent to the destination collection based on the analytics-flush-interval configuration on your Typesense cluster.

Please note that the popular_queries analytics type will only track queries that have a search result.

Automatically expanding prefix search queries:

While creating the analytics rule, you can set expand_query: true to make Typesense aggregate the expanded versions of the search queries made. For e.g. if a user stops typing at sho and the hits were fetched for the word shoe, setting this parameter to true will make Typesense aggregate the word shoe instead of sho. By default, this parameter is set to false, i.e. we will capture the actual user queries (including prefix queries) without any expansion.

NOTE: This aggregation will happen every 5 minutes on Typesense Cloud.

# Send events via API

If you wish to send the search queries as events via the API, you can create an analytics rule that defines a search event for consumption. By setting enable_auto_aggregation to false, only the search queries sent via the API will be aggregated.

{
  "name": "product_queries_aggregation",
  "type": "popular_queries",
  "params": {
    "source": {
      "collections": [
        "products"
      ],
      "enable_auto_aggregation": false,
      "events": [
        {
          "type": "search",
          "name": "products_search_event"
        }
      ]
    },
    "destination": {
      "collection": "product_queries"
    },
    "limit": 1000
  }
}

With this rule in place, you can now directly send the queries as search events, via the API.

curl "http://localhost:8108/analytics/events" -X POST \
     -H "X-TYPESENSE-API-KEY: ${TYPESENSE_API_KEY}" \
     -d '{                           
            "type": "search",                   
            "name": "products_search_event",
            "data": {                     
                  "q": "running shoes",
                  "user_id": "1234"
            }
        }'   

# Query suggestion UX

Once you've set up the analytics rules above, top search terms will start appearing in the destination collection.

You can now use the data in this collection just like any other Typesense collection and send search queries to it, to power a query suggestion UX.

You can also send queries to your main indices in parallel to show both query suggestions and actual results side-by-side, using multi_search.

# Aggregation key

When you send a search query to the source collection, you can optionally send a x-typesense-user-id parameter or a X-TYPESENSE-USER-ID header to indicate the user who made this particular search request. When not specified, Typesense uses the client IP address for aggregation by default.

Since Typesense could be used for type-ahead searches, a search query is counted for aggregation only when there is at least a 4-second pause after the query. For example, f -> fo -> foo -> 4 second pause will register the foo query.

TIP

When testing locally, please be mindful of this 4-second pause and also the analytics-flush-interval configuration.

If you send a lot of queries to the source collection in a short period of time, search terms might not appear in the destination collection right away.

# No hits queries

Like popular queries, you can track queries that produced no hits as well. You can use these queries to identify gaps in your content.

# Create a collection for queries

Let's first create a new collection to which search queries that produce no hits will be aggregated and logged to.

This collection is just like any other Typesense collection, except that it is automatically populated by Typesense with no-hits search queries.

The q and count fields are mandatory.

# Create an analytics rule

We will track the most 1000 frequently occurring queries that don't produce a hit when the products collection is searched. These queries are then aggregated in no_hits_queries collection.

# Counting events for popularity

Typesense allows you to track how often a particular document is clicked on or purchased. You can then use this counter value to rank search results based on popularity.

Let's say there's a collection with a popularity field that we will use as a counter:

{                                        
  "name": "products",           
  "fields": [       
    {"name": "title", "type": "string"},
    {"name": "popularity", "type": "int32", "optional": true}
  ]                                    
}

We mark the popularity field as optional since we want to skip this field during indexing and let Typesense increment the value of this field based on user interactions.

# Create an analytics rule

Let's define a counter analytics rule that will increment this field whenever a click event happens.

The counter rule indicates which collection must be tracked and where the counter value should be stored. The weight property of the event parameter determines the size of the increments. In this case, we want to increment the popularity field by 1 every time a click event is sent to Typesense that's associated with that document.

# Sending click events

Once this counter rule is in-place, you can send start sending click events using the event name that we configured in the counter rule earlier (products_click_counter):

curl "http://localhost:8108/analytics/events" -X POST \
     -H "X-TYPESENSE-API-KEY: ${TYPESENSE_API_KEY}" \ 
     -d '{
            "type": "click",
            "name": "products_click_event",
            "data": {
                  "doc_id": "1024",
                  "user_id": "111112"
            }
        }'

The click events are aggregated, and the popularity field of the document with id specified in the doc_id event data is incremented. Frequency of aggregation is controlled by the --analytics-flush-interval option.

TIP

Ensure that the doc_id sent in the event payload matches with the actual id of the document stored in the destination collection.

# Aggregating multiple events

In fact, you can set up a counter rule that gives different weights to different events.

Event Type Description
click Tracking clicks against documents returned in search response.
conversion Tracking conversion (e.g. purchase) of specific documents.
visit Tracking page visits to specific documents, useful for recommendations.

Let's setup a rule for aggregating both click and conversion events.

{
  "name": "products_popularity",
  "type": "counter",
  "params": {
    "source": {
      "collections": [
        "products"
      ],
      "events": [
        {
          "type": "click",
          "weight": 1,
          "name": "products_click_event"
        },
        {
          "type": "conversion",
          "weight": 2,
          "name": "products_purchase_event"
        }
      ]
    },
    "destination": {
      "collection": "products",
      "counter_field": "popularity"
    }
  }
}

You can now send a conversion event via the API.

curl "http://localhost:8108/analytics/events" -X POST \
     -H "X-TYPESENSE-API-KEY: ${TYPESENSE_API_KEY}" \ 
     -d '{
            "type": "conversion",
            "name": "products_purchase_event",
            "data": {
                  "doc_id": "1022",
                  "user_id": "111117"
            }
        }'

Since we've configured a conversion event to have a weight of 2, the popularity field in the products collection will be incremented by 2 for every conversion event.

# Listing all rules

The listing API allows you to list all the analytics rules stored in your Typesense cluster.

# Remove a rule

Removing an analytics rule will stop aggregation of new queries, but the already aggregated queries will still be present in the destination collection.

Last Updated: 9/23/2024, 11:30:44 AM