liveBook · Manning

Continuous improvement

As explained in AI Recommender Systems

A feedback loop is a cycle in which system outputs prompt user interactions that are logged and fed back to improve subsequent outputs.

Continuous improvement is the feedback loop between recommendations and user behavior. A system shows items, users interact with them, and those interactions are logged as data for future recommendations. Repeating the cycle lets recommendations adapt over time; it does not guarantee that they improve for every user or item. section 1.3 section 1.6

Figure 1.6 The recommender system feedback loop. Users receive recommendations, interact with content (clicks, purchases, views), interactions are logged, this data trains models, and improved models generate better recommendations.

The diagram traces the production cycle from recommendations through user interactions and logging to model training and subsequent recommendations. section 1.3

The logged signal needs interpretation:

  • Implicit feedback, such as clicks, purchases, watch time, and skips, is plentiful but noisy: a click does not necessarily mean the user liked an item. section 1.3
  • Explicit feedback, such as ratings, reviews, and thumbs-up or thumbs-down, has a clearer meaning but is sparse. It can sharpen the behavioral signal when available. section 1.3
  • Behavioral data can reveal shared tastes that item metadata misses, making the loop useful for collaborative filtering. section 1.3.2

When more interaction data reinforces a bias

If a popular comic such as Asterix is recommended often, it can collect more interactions and appear still more popular in later training data. A niche title such as Little Frog may struggle for visibility. Repeated similar recommendations can also narrow what users discover. New users and items present the opposite problem: they have little interaction history to feed back. The system must balance showing items it expects users to like with trying less-certain items. section 1.3.2 section 1.6

How AI Recommender Systems covers it

How deep it goes
Traces recommendation, interaction, logging, and model training, then identifies signal quality and bias challenges without specifying a retraining procedure.
Examples
Illustrates reinforcement with Asterix gaining interactions while Little Frog struggles for visibility in a comic catalog.
What you can do after
Evaluate/understand how interaction data can adapt recommendations while amplifying exposure bias or leaving new users and items without history.
How it connects
Connects implicit feedback, explicit feedback, collaborative filtering, popularity bias, filter bubbles, cold start, and exploration versus exploitation.

Also explained in

Compare how 43 books explain continuous improvement →

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.