/mta

Multi-Touch Attribution

Primary LanguagePython

mta

Multi-Touch Attribution. Find out which channels contribute most to user conversion.

Models

This package contains implementations the following Multi-Touch Attribution models:

  • Shapley
  • Markov
  • So-called Simple Probabilistic Model by Shao and Li
  • Bagged Logistic Regression by Shao and Li
  • Additive Hazard (Survival)

In addition, some popular heuristic “models” are included, specifically

  • First Touch
  • Linear
  • Last Touch
  • Time Decay
  • Position Based

Included Data

The package comes with the same test data set as an R package called ChannelAttribution - there are 10,000 rows containing customer journeys across 12 channels: alpha, beta, delta, epsilon, eta, gamma, iota, kappa, lambda, mi, theta and zeta.

!data_snippet

These are conversion aggregations by path. Suppose there’s a path (customer journey)

a > b > c

with total_conversions equal to 2 and total_null equal to 5. This means that we recorded 2 consumer journeys

a > b > c > (conversion)

and 5 customer journeys

a > b > c > (null)

There’s an option to generate timestamp data if you want to use the Additive Hazard model (the only model that explicitly incorporates exposure times).

References

  • Nisar and Yeung (2015) - Purchase Conversions and Attribution Modeling in Online Advertising: An Empirical Investigation pdf
  • Shao and Li (2011) - Data-driven Multi-touch Attribution Models pdf
  • Dalessandro et al (2012) - Causally Motivated Attribution for online Advertising pdf
  • Cano-Berlanga et al (2017) - Attribution models and the Cooperative Game Theory pdf
  • Ren et al (2018) - Learning Multi-touch Conversion Attribution with Dual-attention Mechanisms for Online Advertising pdf
  • Zhang et al (2014) - Multi-Touch Attribution in Online Advertising with Survival Theory pdf
  • Geyik et al (2014) - Multi-Touch Attribution Based Budget Allocation in Online Advertising pdf