RankFeatureQuery for Numeric Relevance Signals

RankFeatureQuery turns a numeric field into a scoring signal.

Typical examples:

• popularity
• download count
• click-through count
• editorial quality score

Basic example

import {
  DocumentIndex,
  NumericFieldIndex,
  RankFeatureQuery
} from "@tryformation/querylight-ts";

const index = new DocumentIndex({
  popularity: new NumericFieldIndex()
});

index.index({ id: "1", fields: { popularity: ["5"] } });
index.index({ id: "2", fields: { popularity: ["20"] } });
index.index({ id: "3", fields: { popularity: ["50"] } });

const hits = index.searchRequest({
  query: new RankFeatureQuery({ field: "popularity" })
});

Supported modes

Saturation

Default behavior.

new RankFeatureQuery({ field: "popularity", options: { pivot: 10 } });

Useful when you want diminishing returns.

Log

new RankFeatureQuery({ field: "popularity", options: { type: "log", scalingFactor: 1 } });

Useful when raw values grow very large.

Sigmoid

new RankFeatureQuery({ field: "popularity", options: {
  type: "sigmoid",
  pivot: 10,
  exponent: 2
} });

Useful when you want a stronger curve around a pivot.

Linear

new RankFeatureQuery({ field: "popularity", options: { type: "linear", factor: 0.5 } });

Useful when the feature should scale directly.

Notes

• Map the field with NumericFieldIndex when possible.
• Documents with missing or non-positive values do not contribute useful scores.
• Rank features work best as one signal among several, not as the only ranking rule.