Scale the boost values to a [min,max] range in the boost with signals stage

Goal

Explain how the “Scale the boost values to a [min,max] range” setting works in the Boost with Signals stage of a query pipeline, and how it interacts with signal aggregation, weighting, and other relevancy factors.

Environment

Fusion 5.9 and above
Applicable to self-hosted and cloud deployments on Kubernetes (EKS, GKE, AKS, or other supported platforms).

Guide

Understanding signal boost values

Fusion uses aggregated signals such as clicks, add-to-cart, and purchases to influence search result ranking. Each signal contributes to a calculated weight (weight_d) during the aggregation job.

Key points:

• weight_d is computed using a time decay function so that recent signals carry more weight than older ones. For example, with a half-life of 30 days, a signal 30 days old contributes half as much as one from today, and one 60 days old contributes one quarter as much.

• Different signal types are assigned different weights through the signalTypeWeights configuration (for example, purchases > add-to-cart > clicks).

What the scale setting does

The Scale the boost values to a [min,max] range option normalizes signal boost values before they are applied to Solr scoring.

• Without scaling: boost values may vary widely depending on raw counts and decay. A document with thousands of clicks might dominate relevance, overshadowing other signals or Solr’s natural scoring.

• With scaling: Fusion maps all signal boost values for a given query into the defined [min,max] range. This ensures consistent influence and prevents any one signal from overwhelming other relevancy components such as TF/IDF, field boosts, or business rules.

Example:
If the raw boost values range from 1 to 50, and you configure scaling to [0.1, 2.0], Fusion rescales the lowest boost to 0.1, the highest boost to 2.0, and interpolates all other values proportionally in between.

When to use scaling

Scaling is most useful when:

• Signal counts differ significantly across documents, causing large discrepancies in boost values.

• Business needs require balancing signals with other relevancy factors such as newness, inventory, or margin.

• You want to control the relative impact of signals without changing the aggregation job logic.

Query pipeline considerations

• Scaling is applied after the Final Boost Weight Expression in the Boost with Signals stage.

• Use scaling to ensure signal boosts stay within a reasonable range relative to Solr’s native scoring.

• Other factors to consider in query pipeline tuning:

  • Boost method: query-param vs query-parser — affects how boosts are passed to Solr.

  • Rollup weight strategy: max vs sum — determines how multiple signal types are combined.

  • Minimum should match (mm): controls how many optional clauses must match in the query parser.

• Scaling does not directly relate to the number of recommendations returned; it strictly controls the relative magnitude of boosts.

Verifying the effect

To inspect how scaling is applied:

1. Run a query with debug=true enabled.

2. In the response JSON, check the responseHeader.params section.

3. Look at the Solr boost parameter values to confirm they have been normalized to the defined [min,max] range.

This helps verify whether scaling is keeping boosts within the expected influence range compared to Solr’s relevance scoring.