> ## Documentation Index
> Fetch the complete documentation index at: https://algolia.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Upload events from a CSV file

> Learn how you can upload historical user events to Algolia Recommend to get started quicker.

export const Application = () => <Tooltip tip="An Algolia application is a self-contained environment with its own indices, configuration, and API keys. Applications don't share data or settings with each other.">
    application
  </Tooltip>;

export const AlgoliaRecommend = () => <svg xmlns="http://www.w3.org/2000/svg" className="inline" viewBox="0 0 80 80" width="20" height="20" fill="none" role="presentation" ariaLabel="Algolia Recommend">
    <path d="m50 65-4 12H34l-4-12h20Z" fill="#36395A"></path>
    <path d="M68 32c0 15.464-12.536 28-28 28S12 47.464 12 32 24.536 4 40 4s28 12.536 28 28Z" fill="#FF2A6A"></path>
    <path d="M43 34V17L26 34h17Z" fill="#fff"></path>
    <path d="M37 30v17l17-17H37Z" fill="#fff"></path>
  </svg>;

<Callout icon="flask-conical" color="#14b8a6">
  This is a **beta feature** according to [Algolia's Terms of Service ("Beta Services")](https://www.algolia.com/policies/terms/).
</Callout>

Capturing events [through the Insights API](/doc/guides/sending-events/getting-started) is important
to continuously train and improve the Recommend models.
Collecting enough events can take some time.
You can import past user events from a CSV file to benefit from Recommend earlier.

## Format of the CSV file with historical events

The CSV file with historical events must have the following format:

* The CSV file must be smaller than 100 MB.
* The first row must contain the strings `userToken`, `timestamp`, `objectID`, `eventType`, and `eventName`.
  Extra columns are ignored.
* Each row represents an event tied to a single `objectID`.
* The timestamps should cover a period of **at least 30 days**.
* Events older than 90 days are ignored.

<Note>
  Some programs like Microsoft Excel can add invisible characters when exporting a CSV file.
  They might not be visible in regular text editors. The easiest way to find and remove
  these characters is by using a code editor such as Visual Studio Code.
</Note>

Each event must have the following properties:

| Property    | Description                                                                                                                              |
| ----------- | ---------------------------------------------------------------------------------------------------------------------------------------- |
| `userToken` | A unique identifier for the user session                                                                                                 |
| `timestamp` | The date of the event in a standard format: [ISO8601 or RFC3339](https://en.wikipedia.org/wiki/ISO_8601#RFCs) (with or without the time) |
| `objectID`  | A unique identifier for the item the event is tied to                                                                                    |
| `eventType` | Either "click" or "conversion"                                                                                                           |
| `eventName` | A name for the event, which can be the same as `eventType`                                                                               |

For more information, see [Send events properties](/doc/rest-api/insights/push-events).

## Upload historical events for Recommend

To import historical events from a CSV file, follow these steps:

1. Go to the [Algolia dashboard](https://dashboard.algolia.com/explorer/browse) and select your Algolia <Application />.
2. On the left sidebar, select <AlgoliaRecommend /> **Recommend**.
3. Select a model you want to train and click **Start using**.
4. In the **Select data source** section, select an Algolia index as a source for your recommendations.
5. In the **Collect events** section, click **Upload .csv** to upload your CSV file with the historical events.

   <img src="https://mintcdn.com/algolia/QUuhkPGiow1bP-ae/images/guides/recommend/csv-upload.png?fit=max&auto=format&n=QUuhkPGiow1bP-ae&q=85&s=043facead98473cd62a1a7ce42c1b24f" alt="Import a CSV file with historical events for training your Recommend model" width="1039" height="337" data-path="images/guides/recommend/csv-upload.png" />

When you've collected enough [click and conversion events](/doc/guides/sending-events),
Algolia Recommend stops using the events from the CSV file.
Only the user-generated events are used to train the model.

You can also **re-upload a CSV file**.
The training only takes the newer file into account and discards the old events.

## Export events from Google Analytics 360 with BigQuery

If you track user events with Google Analytics,
you can export these events with BigQuery.
You can then save these events in a CSV file which you can import in the Algolia dashboard to start training your Recommend models.

### Before you begin

Before you can export your events from Google Analytics with BigQuery,
you must meet the following requirements:

* A [Google Analytics 360](https://marketingplatform.google.com/about/analytics-360/) account with a website tracking ID
* [Enhanced Ecommerce](https://support.google.com/analytics/answer/6014841) activated and set up for your website
* [BigQuery Export enabled in Google Analytics 360](https://support.google.com/analytics/answer/3416092) for setting up daily imports into BigQuery
* The `productSKU` property from Google Analytics 360 must match the `objectID` property of your Algolia records.

### Set up a BigQuery export

To export user events for training Algolia Recommend models,
adapt the following query.

Replace the following variables:

* `GCP_PROJECT_ID`: the name of the project that holds the Analytics 360 data in BigQuery
* `BQ_DATASET`: the name of the dataset that stores the exported events
* `DATE_FROM` and `DATE_TO` with the corresponding dates in `YYYY-MM-DD` format for a time window of at least 30 days.

You can run this query in the SQL workplace for BigQuery or use one of the [BigQuery API client libraries](https://cloud.google.com/bigquery/docs/reference/libraries).

```sql SQL icon="database" theme={"system"}
WITH ecommerce_data AS (
    SELECT
        fullVisitorId as user_token,
        TIMESTAMP_SECONDS(visitStartTime + CAST(hits.time/1000 AS INT64)) as timestamp,
        products.productSKU as object_ids,
        CASE WHEN hits.eCommerceAction.action_type = "2" THEN 'click'
            WHEN hits.eCommerceAction.action_type = "3" THEN 'click'
            WHEN hits.eCommerceAction.action_type = "5" THEN 'click'
            WHEN hits.eCommerceAction.action_type = "6" THEN 'conversion'
        END
        AS event_type,
        CASE WHEN hits.eCommerceAction.action_type = "2" THEN "product_view"
            WHEN hits.eCommerceAction.action_type = "3" THEN "add_to_cart"
            WHEN hits.eCommerceAction.action_type = "5" THEN "checkout"
            WHEN hits.eCommerceAction.action_type = "6" THEN "purchase"
        END
        AS event_name
    FROM
        `GCP_PROJECT_ID.BQ_DATASET.ga_sessions_*`,
        UNNEST(hits) as hits,
        UNNEST(hits.product) as products
    WHERE
        _TABLE_SUFFIX BETWEEN FORMAT_DATE('%Y%m%d',DATE('DATE_FROM')) AND FORMAT_DATE('%Y%m%d',DATE('DATE_TO'))
    AND
        fullVisitorId IS NOT NULL
    AND hits.eCommerceAction.action_type in UNNEST(["2", "3", "5", "6"])
),
dedup_ecommerce_data AS (
    SELECT user_token as userToken,
    timestamp, event_name as eventName,
    event_type as eventType,
    object_id as objectID
    FROM ecommerce_data
    GROUP BY userToken, timestamp, eventName, eventType, objectID
)

select * from dedup_ecommerce_data
```