Source domain

A source domain is the area or set of data from which metaphorical concepts or computational knowledge is drawn to understand another concept, the target domain. In linguistics, it’s the concrete concept in a metaphor that explains a more abstract concept. In computer science, particularly in transfer learning and domain adaptation, it refers to the data or environment on which a model is trained, distinct from the target domain where it will be used.

In Linguistics (Conceptual Metaphor Theory) Conceptual metaphor

Definition: The conceptual domain that provides the basis for a metaphor.

Example: In the metaphor “Life is a journey,” the journey is the source domain, providing concepts like “path,” “destination,” and “obstacles” to understand “life”.

Purpose: To explain abstract concepts (like arguments, love, or ideas) using more concrete, familiar concepts.

In Computer Science (Domain Adaptation)

Definition: The original data distribution or dataset used to train a machine learning model.

Purpose: To leverage knowledge learned from a source domain with a specific data distribution to improve a related but different target domain with a potentially different data distribution.

Example: A spam filter trained on emails from one user (the source domain) is adapted for a different user with new email patterns (the target domain).