CTR prediction models

Limitations

• Available only on Industry Algorithm Edition - Dedicated Cluster instances.

• Each application supports up to three CTR prediction models.

• CTR prediction models can only be called from Cava-based plugins.

Prerequisites

Before you begin, make sure you have:

• An Industry Algorithm Edition configuration

Create and train a model

1. Log on to the OpenSearch console and go to Search Algorithm Center > Sort Configuration in the left-side navigation pane.

2. On the Policy Management page, click CTR Prediction Models in the left-side pane, then click Create.

   

3. Enter a model name and map the training fields.

   • Commodity ID and Commodity Title are required. The more fields you map, the better the model performs.

   

   

4. Find the model on the CTR Prediction Models page and click Train in the Actions column.

   

5. Monitor training progress on the model details page.

   

6. After training completes, check the model status:

   • Available: The model is ready to use.

   • Unavailable: Adjust your data to meet the upgrade conditions in the Data integrity levels section, then retrain the following day. If issues persist, submit a ticket for technical support.

   

Enable scheduled training to retrain the model daily and keep it current.

Apply the model in a fine sort policy

1. Go to Search Algorithm Center > Sort Configuration > Policy Management and click Create.

   

2. Set the Policy Name, select Fine Sort from the Scope drop-down list, and select Cava Script from the Type drop-down list. Click Next.

   

3. Click Add Script File, paste the sample Cava script below into the editor, and replace "Name of your CTR prediction model" with your model name.

4. Click Compile. If compilation succeeds, click Save and then Publish.

   

   

Sample Cava script:

package users.scorer;
import com.aliyun.opensearch.cava.framework.OpsScoreParams;
import com.aliyun.opensearch.cava.framework.OpsScorerInitParams;
import com.aliyun.opensearch.cava.framework.OpsRequest;
import com.aliyun.opensearch.cava.framework.OpsDoc;
import com.aliyun.opensearch.cava.features.algo.AlgoModel;

class BasicSimilarityScorer {
    boolean init(OpsScorerInitParams params) {
        return true;
    }

    double score(OpsScoreParams params) {
        double score = 0;
        return score;
    }
};

class IntelligenceAlgorithmScorer {
    AlgoModel _algoModel;

    boolean init(OpsScorerInitParams params) {
        // tf_checkpoint is a fixed parameter name—do not change it.
        _algoModel = AlgoModel.create(params, "tf_checkpoint", "ctr", "Name of your CTR prediction model");
        return true;
    }

    double score(OpsScoreParams params) {
        OpsDoc doc = params.getDoc();
        double modelScore = _algoModel.evaluate(params);
        doc.trace("ctrModelScore: ", modelScore);

        double score = modelScore + 700;
        return score;
    }
};

The IntelligenceAlgorithmScorer loads your CTR model using AlgoModel.create and scores each document in the fine sort (second-stage ranking) phase.

Run a search test

After publishing the policy, send a search request with the following required parameters.

For example Java SDK code, see SDK for Java demo code for implementing the search feature.

Manage the model

Basic information

The model details page shows Created At, Status, Last Training Time, and Latest Version Status.

Training fields

Click Map Training Fields to modify or delete fields in the Map Training Fields panel. After changing training fields, retrain the model.

Scheduled training

Scheduled training is enabled by default and runs daily. Modify the schedule to customize the training cycle.

Data integrity levels

The integrity level reflects the volume and quality of your behavioral data. A higher level means richer data and more accurate model training.

Data Integrity has two valid values: Available Data and Abnormal Data.

Item page views (IPVs) measure CTR per search, where bhv_type=click. When a user clicks a result, upload two behavioral data entries: one with bhv_type=expose and one with bhv_type=click.

The table below describes each level and the conditions required to advance to the next.

What's next

• Algorithms API reference

• SDK for Java demo code for implementing the search feature