læringtilrangeringmodeller
læringtilrangeringmodeller, also known as learning‑to‑rank models, are supervised machine learning techniques designed to predict an order or ranking of items rather than a single label or numeric score. The goal of these models is to arrange items in a list so that the most relevant items appear first for a given query or context. Learning‑to‑rank encompasses three primary approaches: pointwise, pairwise, and listwise. In the pointwise method, each item is treated as a separate regression or classification task, predicting a relevance score that can be thresholded to form a ranking. Pairwise learning focuses on relative comparisons between pairs of items, training the model to predict which of two items should come first. Listwise methods take an entire set of items as input and optimize a global ranking objective, such as normalised discounted cumulative gain (nDCG).
Common algorithms include RankSVM and RankBoost for pairwise learning, and LambdaMART, Gradient Boosted Decision Trees, and