Recommendation

As explained in Causal Inference for Data Science

A recommendation is a suggested item, content choice, or action produced for a user or decision-maker.

A recommendation can predict what a customer will buy without changing what they buy. Predictive accuracy therefore does not establish that recommendations increase sales. section 3.3.1

When accuracy has no effect

Suppose a mobile app predicts every purchase correctly, including a customer buying bread. If the customer would have bought the bread anyway, the recommendation has no causal effect. Comparing users who receive recommendations with users who do not can reveal whether sales actually increase. section 3.3.1

Evaluate a recommendation as an intervention:

• Treat showing a specific item as a treatment and compare the user's response with the response when it is not shown. section 3.3.1
• Use an A/B test to estimate the average treatment effect of recommendations rather than relying only on cross-validation accuracy. section 3.3.1

Figure 3.12 Graph of a pure recommender system that does not influence the customer. This system may predict customer behavior accurately, but it does not affect their actions.

Use the causal graph to distinguish a recommender that predicts customer behavior from one that changes it; the corresponding standard graph is also available. section 3.3.1

How Causal Inference for Data Science covers it

How deep it goes

Contrasts predictive accuracy with causal impact and works through graphs, A/B testing, and average treatment effects.

Examples

Uses a bread purchase and a mobile-app experiment contrasting recommendation and no-recommendation groups.

What you can do after

Evaluate recommender impact with an A/B test instead of inferring effectiveness from predictive accuracy.

How it connects

Connects recommender systems with causal inference, A/B testing, average treatment effect, predictive modeling, and cross-validation.

liveBookmanning.com

Read Causal Inference for Data Science on liveBook Get the book at manning.com

This explainer was generated by Manning's knowledge graph from the text of Causal Inference for Data Science. The section links point to the parts of the book it draws on.