Präferenzlernen

Präferenzlernen, also known as preference learning, is a subfield of machine learning concerned with learning models from preference data. Instead of receiving explicit labels for individual data points, the learning system is presented with comparisons or rankings between items. For example, a user might indicate that item A is preferred over item B, or that a list of items is ranked from best to worst. The goal of Präferenzlernen is to build a model that can predict these preferences for new, unseen items or to understand the underlying factors that drive preferences.

This type of learning is particularly useful in scenarios where obtaining absolute scores or labels is difficult, subjective, or expensive. Applications include recommender systems, where users might prefer one movie over another, search result ranking, where users implicitly rank the relevance of different search snippets, and computational advertising, where users' click-through behavior indicates a preference.

Common approaches in Präferenzlernen involve learning a scoring function that assigns a value to each item, such that preferred items receive higher scores. Alternatively, models can learn a pairwise comparison function that directly predicts the probability of one item being preferred over another. Techniques often draw from areas like pairwise classification, ranking algorithms, and utility learning. The evaluation of Präferenzlernen models typically involves metrics that assess the accuracy of predicted rankings or the quality of learned preference models on held-out preference data.