Since simulation is a tool for generating data, a major task in executing simulations is to extract data from simulation runs. However, traditional methods of data extraction such as instrumenting the model code by hand or over-instrumenting the model and filtering data offline suffer from inflexibility and poor efficiency. To overcome these shortcomings, this paper promotes configurable targeted online data extraction, which also has special relevance in the field of real-time simulation. Nevertheless, there is no common terminology for the range of functions for targeted data extraction and there is no common concept for the implementation of flexible and efficient solutions. By decomposing the data extraction problem and by formalizing generalizable parts, this paper provides a conceptional framework for the assessment and implementation of language-based data extraction solutions. It turns out that data extraction can be decomposed into a sequential and a structural dimension, both of which having operations for selection, extraction, and windowed aggregation. As a proof of concept, the functionality of existing data extraction languages is analyzed using the proposed terminology.