Create recommendations based on implicit feedback with a matrix factorization model

This tutorial teaches you how to create a matrix factorization model and train it on the Google Analytics 360 user session data in the public GA360_test.ga_sessions_sample table. You then use the matrix factorization model to generate content recommendations for site users.

Using indirect customer preference information, like user session duration, to train the model is called training with implicit feedback. Matrix factorization models are trained using the Weighted-Alternating Least Squares algorithm when you use implicit feedback as training data.

Create a dataset

Create a BigQuery dataset to store your ML model.

Console

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to the BigQuery page

  2. In the Explorer pane, click your project name.

  3. Click View actions > Create dataset

  4. On the Create dataset page, do the following:

    • For Dataset ID, enter bqml_tutorial.

    • For Location type, select Multi-region, and then select US (multiple regions in United States).

    • Leave the remaining default settings as they are, and click Create dataset.

bq

To create a new dataset, use the bq mk command with the --location flag. For a full list of possible parameters, see the bq mk --dataset command reference.

  1. Create a dataset named bqml_tutorial with the data location set to US and a description of BigQuery ML tutorial dataset:

    bq --location=US mk -d \
     --description "BigQuery ML tutorial dataset." \
     bqml_tutorial

    Instead of using the --dataset flag, the command uses the -d shortcut. If you omit -d and --dataset, the command defaults to creating a dataset.

  2. Confirm that the dataset was created:

    bq ls

API

Call the datasets.insert method with a defined dataset resource.

{
  "datasetReference": {
     "datasetId": "bqml_tutorial"
  }
}

BigQuery DataFrames

Before trying this sample, follow the BigQuery DataFrames setup instructions in the BigQuery quickstart using BigQuery DataFrames. For more information, see the BigQuery DataFrames reference documentation.

To authenticate to BigQuery, set up Application Default Credentials. For more information, see Set up ADC for a local development environment.

import google.cloud.bigquery

bqclient = google.cloud.bigquery.Client()
bqclient.create_dataset("bqml_tutorial", exists_ok=True)

Prepare the sample data

Transform the data from the GA360_test.ga_sessions_sample table into a better structure for model training, and then write this data to a BigQuery table. The following query calculates the session duration for each user for each piece of content, which you can then use as implicit feedback to infer the user's preference for that content.

Follow these steps to create the training data table:

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. Create the training data table. In the query editor, paste in the following query and click Run:

    CREATE OR REPLACE TABLE `bqml_tutorial.analytics_session_data`
    AS
    WITH
      visitor_page_content AS (
        SELECT
          fullVisitorID,
          (
            SELECT
              MAX(
                IF(
                  index = 10,
                  value,
                  NULL))
            FROM
              UNNEST(hits.customDimensions)
          ) AS latestContentId,
          (LEAD(hits.time, 1) OVER (PARTITION BY fullVisitorId ORDER BY hits.time ASC) - hits.time)
            AS session_duration
        FROM
          `cloud-training-demos.GA360_test.ga_sessions_sample`,
          UNNEST(hits) AS hits
        WHERE
          # only include hits on pages
          hits.type = 'PAGE'
        GROUP BY
          fullVisitorId,
          latestContentId,
          hits.time
      )
    # aggregate web stats
    SELECT
      fullVisitorID AS visitorId,
      latestContentId AS contentId,
      SUM(session_duration) AS session_duration
    FROM
      visitor_page_content
    WHERE
      latestContentId IS NOT NULL
    GROUP BY
      fullVisitorID,
      latestContentId
    HAVING
      session_duration > 0
    ORDER BY
      latestContentId;
  3. View a subset of the training data. In the query editor, paste in the following query and click Run:

    SELECT * FROM `bqml_tutorial.analytics_session_data` LIMIT 5;

    The results should look similar to the following:

    +---------------------+-----------+------------------+
    | visitorId           | contentId | session_duration |
    +---------------------+-----------+------------------+
    | 7337153711992174438 | 100074831 | 44652            |
    +---------------------+-----------+------------------+
    | 5190801220865459604 | 100170790 | 121420           |
    +---------------------+-----------+------------------+
    | 2293633612703952721 | 100510126 | 47744            |
    +---------------------+-----------+------------------+
    | 5874973374932455844 | 100510126 | 32109            |
    +---------------------+-----------+------------------+
    | 1173698801255170595 | 100676857 | 10512            |
    +---------------------+-----------+------------------+
    

Create the model

Create a matrix factorization model and train it on the data in the analytics_session_data table. The model is trained to predict a confidence rating for every visitorId-contentId pair. The confidence rating is created with centering and scaling by the median session duration. Records where the session duration is more than 3.33 times the median are filtered out as outliers.

The following CREATE MODEL statement uses these columns to generate recommendations:

  • visitorId—The visitor ID.
  • contentId—The content ID.
  • rating—The implicit rating from 0 to 1 calculated for each visitor-content pair, centered and scaled.
  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. In the query editor, paste in the following query and click Run:

    CREATE OR REPLACE MODEL `bqml_tutorial.mf_implicit`
      OPTIONS (
        MODEL_TYPE = 'matrix_factorization',
        FEEDBACK_TYPE = 'implicit',
        USER_COL = 'visitorId',
        ITEM_COL = 'contentId',
        RATING_COL = 'rating',
        L2_REG = 30,
        NUM_FACTORS = 15)
    AS
    SELECT
      visitorId,
      contentId,
      0.3 * (1 + (session_duration - 57937) / 57937) AS rating
    FROM `bqml_tutorial.analytics_session_data`
    WHERE 0.3 * (1 + (session_duration - 57937) / 57937) < 1;

    The query takes about 10 minutes to complete, after which the mf_implicit model appears in the Explorer pane. Because the query uses a CREATE MODEL statement to create a model, you don't see query results.

Get training statistics

Optionally, you can view the model's training statistics in the Cloud de Confiance console.

A machine learning algorithm builds a model by creating many iterations of the model using different parameters, and then selecting the version of the model that minimizes loss. This process is called empirical risk minimization. The model's training statistics let you see the loss associated with each iteration of the model.

Follow these steps to view the model's training statistics:

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. In the left pane, click Explorer:

    Highlighted button for the Explorer pane.

    If you don't see the left pane, click Expand left pane to open the pane.

  3. In the Explorer pane, expand your project and click Datasets.

  4. Click the bqml_tutorial dataset. You can also use the search feature or filters to find the dataset.

  5. Click the Models tab.

  6. Click the mf_implicit model and then click the Training tab

  7. In the View as section, click Table. The results should look similar to the following:

    +-----------+--------------------+--------------------+
    | Iteration | Training Data Loss | Duration (seconds) |
    +-----------+--------------------+--------------------+
    |  5        | 0.0027             | 47.27              |
    +-----------+--------------------+--------------------+
    |  4        | 0.0028             | 39.60              |
    +-----------+--------------------+--------------------+
    |  3        | 0.0032             | 55.57              |
    +-----------+--------------------+--------------------+
    |  ...      | ...                | ...                |
    +-----------+--------------------+--------------------+
    

    The Training Data Loss column represents the loss metric calculated after the model is trained. Because this is a matrix factorization model, this column shows the mean squared error.

Evaluate the model

Evaluate the performance of the model by using the ML.EVALUATE function. The ML.EVALUATE function evaluates the predicted content ratings returned by the model against the evaluation metrics calculated during training.

Follow these steps to evaluate the model:

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. In the query editor, paste in the following query and click Run:

    SELECT
      *
    FROM
      ML.EVALUATE(MODEL `bqml_tutorial.mf_implicit`);

    The results should look similar to the following:

    +------------------------+-----------------------+---------------------------------------+---------------------+
    | mean_average_precision |  mean_squared_error   | normalized_discounted_cumulative_gain |    average_rank     |
    +------------------------+-----------------------+---------------------------------------+---------------------+
    |     0.4434341257478137 | 0.0013381759837648962 |                    0.9433280547112802 | 0.24031636088594222 |
    +------------------------+-----------------------+---------------------------------------+---------------------+
    

    For more information about the ML.EVALUATE function output, see Output.

Get the predicted ratings for a subset of visitor-content pairs

Use the ML.RECOMMEND to get the predicted rating for each piece of content for five site visitors.

Follow these steps to get predicted ratings:

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. In the query editor, paste in the following query and click Run:

    SELECT
      *
    FROM
      ML.RECOMMEND(
        MODEL `bqml_tutorial.mf_implicit`,
        (
          SELECT
            visitorId
          FROM
            `bqml_tutorial.analytics_session_data`
          LIMIT 5
        ));

    The results should look similar to the following:

    +-------------------------------+---------------------+-----------+
    | predicted_rating_confidence   | visitorId           | contentId |
    +-------------------------------+---------------------+-----------+
    | 0.0033608418060270262         | 7337153711992174438 | 277237933 |
    +-------------------------------+---------------------+-----------+
    | 0.003602395397293956          | 7337153711992174438 | 158246147 |
    +-------------------------------+---------------------+--  -------+
    | 0.0053197670652785356         | 7337153711992174438 | 299389988 |
    +-------------------------------+---------------------+-----------+
    | ...                           | ...                 | ...       |
    +-------------------------------+---------------------+-----------+
    

Generate recommendations

Use the predicted ratings to generate the top five recommended content IDs for each visitor ID.

Follow these steps to generate recommendations:

  1. In the Cloud de Confiance console, go to the BigQuery page.

    Go to BigQuery

  2. Write the predicted ratings to a table. In the query editor, paste in the following query and click Run:

    CREATE OR REPLACE TABLE `bqml_tutorial.recommend_content`
    AS
    SELECT
      *
    FROM
      ML.RECOMMEND(MODEL `bqml_tutorial.mf_implicit`);
  3. Select the top five results per visitor. In the query editor, paste in the following query and click Run:

    SELECT
      visitorId,
      ARRAY_AGG(
        STRUCT(contentId, predicted_rating_confidence)
        ORDER BY predicted_rating_confidence DESC
        LIMIT 5) AS rec
    FROM
      `bqml_tutorial.recommend_content`
    GROUP BY
      visitorId;

    The results should look similar to the following:

    +---------------------+-----------------+---------------------------------+
    | visitorId           | rec:contentId   | rec:predicted_rating_confidence |
    +---------------------+-----------------+-------------------------  ------+
    | 867526255058981688  | 299804319       | 0.88170525357178664             |
    |                     | 299935287       | 0.54699439944935124             |
    |                     | 299410466       | 0.53424780863188659             |
    |                     | 299826767       | 0.46949603950374219             |
    |                     | 299809748       | 0.3379991197434149              |
    +---------------------+-----------------+---------------------------------+
    | 2434264018925667659 | 299824032       | 1.3903516407308065              |
    |                     | 299410466       | 0.9921995618196483              |
    |                     | 299903877       | 0.92333625294129218             |
    |                     | 299816215       | 0.91856701667757279             |
    |                     | 299852437       | 0.86973661454890561             |
    +---------------------+-----------------+---------------------------------+
    | ...                 | ...             | ...                             |
    +---------------------+-----------------+---------------------------------+