Skip to content
    Flow-alerta.cab

    Contextual Bandit

    AI fundamentals & modelscontext-dependent bandit methodContextual Multi-Armed Bandit
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 17 words

    In short

    Learning method that selects an action based on a context and receives only its result as feedback.

    Flow-alerta.cab (2026): Contextual Bandit. https://flow-alerta.cab/en/glossar/contextual-bandit. Editorially reviewed on: 09 October 2026.

    How much detail would you like?

    Practical

    Relevant for adaptive recommendations and the selection of content or offer variants.

    Example scenario: A portal selects suitable auxiliary articles and learns from helpful feedback.

    Technically define rewards, document decision history and evaluate them before productive use and set limits.

    At a glance

    Core
    The system observes features of the situation, selects an available action and receives their reward or cost. It must use successful options while exploring alternatives. Unlike comprehensive reinforcement learning, the immediate decision is typically the priority.
    Use
    Relevant for adaptive recommendations and the selection of content or offer variants.
    Example
    Example scenario: A portal selects suitable auxiliary articles and learns from helpful feedback.
    Limits
    Only optimize clicks; suppress exploration; do not log selection probabilities; falsify offline comparisons by selection bias.

    Questions about this term

    Is it A/B testing?

    Another decision-making process.

    Does it see all the results?

    No.

    Do you need exploration?

    For new learning usually yes.

    All details and background

    The system observes features of the situation, selects an available action and receives their reward or cost. It must use successful options while exploring alternatives. Unlike comprehensive reinforcement learning, the immediate decision is typically the priority.

    ELI5 – Explained simply

    You choose a possibility for the situation and learn from the result without being able to try out all the alternatives.

    Why it matters in practice

    Relevant for adaptive recommendations and the selection of content or offer variants.

    Example

    Example scenario: A portal selects suitable auxiliary articles and learns from helpful feedback.

    Pitfalls and limits

    • Only optimize clicks; suppress exploration; do not log selection probabilities; falsify offline comparisons by selection bias.

    Origin & history

    Developed from Bandit research and online learning; procedures were used for personalised recommendations, among other things.

    Use in mid-sized companies

    • Technically define rewards, document decision history and evaluate them before productive use and set limits.

    Related terms

    • Multi-Armed Bandit
    • Exploration
    • Exploitation
    • Reinforcement learning
    • Policy evaluation

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting – designing adaptive decision-making systems and measuring their success.

    Matching tool / use case

    View use casesTool: Vowpal Wabbit; Use Case: Selection of helpful knowledge articles.

    Sources

    Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases