Trends Scores

Overview

The Trends tab in the Content Audiences application displays a Top N list of the highest-performing categories. Each category receives a score represented by up to 5 stars. This page describes the calculation method for these scores.

Trends tab showing audience filters, chart, and peak categories table

Score Calculation

Each item's star rating is based on its relative frequency in the selected audience compared to the full population.

  • Let f (foreground) represent the item's frequency in the selected segment.
    For example, in the screenshot, 'energy' has f = 0.38.
  • Let b (background) represent the item's frequency in the full population.
    For example, 'energy' has b = 0.09.

The score s for an item is calculated as:

s = (f - b) * (f / b)

There are two primary ways an item can achieve a high score:

  1. If an item's frequency increases from a low value (e.g., 4% to 8%), the f / b term has the greatest impact.
  2. If an item's frequency increases from a higher value (e.g., 50% to 60%), the (f - b) term has the greatest impact.

Scores can be unbounded. To display the score as a star rating, the following formula is used:

stars = floor(6 * s / (|s| + c))

where c = 0.1 (empirically determined).

Note on Terminology

Info:
The term "Score" on this page refers specifically to the calculation described above. This is distinct from the meaning of "score" in the context of scoring and normalization of segment scores.

Share Feedback
Page: /build/enterprise-plugins/targeting-relevance/trends
Section: Build
Category *
Trends Scores | Bloomreach Content Documentation