liveBook · Manning

Discount function

As explained in AI Recommender Systems

A discount function reduces the contribution of a relevance value according to an item's position in a ranked list.

A discount function reduces an item's contribution according to its position in a ranked list. Logarithmic discounting gives position 1 a weight of 1.0, position 2 about 0.63, and position 3 about 0.5; lower-ranked relevant items contribute less. section 4.6.3

Within discounted cumulative gain, the function supports three linked steps:

  • Multiply each item's relevance gain by the discount assigned to its position. section 4.6.3
  • Sum the discounted gains across the evaluated positions to obtain DCG. section 4.6.3
  • Normalize DCG by ideal DCG when comparing users, datasets, or experiments. section 4.6.3

Same relevance, different order

Two recommendation lists can contain the same single relevant item and therefore share cumulative gain. The list placing it first scores higher after discounting because its relevance receives the largest positional weight. section 4.6.3

How AI Recommender Systems covers it

How deep it goes
Defines logarithmic weighting, gives position values, and connects discounted gains to DCG and NDCG.
Examples
Compares two recommendation lists containing the same relevant item at different positions.
What it assumes
Assumes ranked lists and relevance gain, then builds from cumulative gain before introducing positional discounting.
What you can do after
Evaluate ranked lists with DCG or NDCG while accounting for item position.
How it connects
Connects relevance, gain, discounted cumulative gain, ideal DCG, and normalized discounted cumulative gain.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Recommender Systems. The section links point to the parts of the book it draws on.