Training Data Generation

Construct training examples to match the ranking objective. A pointwise model learns from individual document labels; a pairwise model learns relative preferences between documents for a query. This lesson derives both kinds of examples from human ratings and user interactions.

The terms training data row and training example are used interchangeably here. When estimating data volume, distinguish a labeled document from a pair made from two labeled documents.

Training data generation for pointwise approach

Pointwise approach: In this approach of model training, the training data consists of relevance scores for each document. The loss function looks at the score of one document at a time as an absolute ranking.