Experiments and Trends Dataflow
Dataflow Overview
Experiments and Trends in Bloomreach Content use the concept of a visit to track user interactions. The system groups requests from a single visitor into a visit. When there is no activity from the visitor for a defined period (30 minutes by default), the visit is considered complete. Any new request after this period starts a new visit.
For Experiments, completed visits are evaluated to determine if the visitor achieved the goal of any active experiment. These outcomes update the statistical models for the experiments, which in turn adjust how often each variant is shown to visitors. For Trends, all visit data is stored in Elasticsearch. The Trends panel allows users to analyze this data to understand visitor behavior.
Configuration
The Relevance Module manages two scheduled jobs: Visits Aggregator and Model Trainer.
- Visits Aggregator: Collects new request log entries from the Request Log Store and creates or updates visit records in the Visit Store. The Visit Store always uses Elasticsearch. Aggregated visits can be queried using the Trends feature.
- Model Trainer: Processes visits that have not been updated recently and uses them to train the statistical "bandit" models for any active experiments.
If you only use Trends and do not plan to run experiments, you do not need to run the Model Trainer job. However, the performance impact of running the Model Trainer without experiments is minimal, so it is recommended to keep it enabled.
Configuration Parameters
Set the following parameter on /targeting:targeting:
| Parameter | Default | Description |
|---|---|---|
targeting:newVisitIdleTimeMinutes | 30.0 (Double) | Maximum allowed inactivity (in minutes) before a visit is considered complete. |
Configure the Visits Aggregator and Model Trainer jobs at /targeting:targeting/targeting:dataflow/visitsAggregator and /targeting:targeting/targeting:dataflow/modelTrainer, respectively. Available parameters:
| Parameter | Default | Description |
|---|---|---|
running | false (Boolean) | Controls whether the job is active. See Disable Relevance for instructions on disabling it for a single cluster node. |
processedUntil | - | Automatically updated by the job to track progress. Manual changes are only needed if request aggregation must be repeated. |
Important:
Do not set the Model Trainer or Visits Aggregator jobs to running until the data stores are correctly configured. The processedUntil parameter is managed automatically and should only be changed manually if you need to re-aggregate requests.