labeling

Labeling is the process of assigning descriptors, tags, or categories to objects, records, or phenomena to identify, classify, or provide metadata. Labels can be textual, symbolic, or coded, and labeling can be manual, automated, or a combination. The purpose is to enable organization, search, analysis, compliance, or communication, and labels often function as the basis for taxonomy, indexing, or decision-making.

In data science and machine learning, data labeling (annotation) creates ground truth targets for supervised learning. This involves human annotators or crowdsourcing to assign correct labels to data such as images, text, audio, or video. Quality control relies on clear guidelines, training, duplicate labeling, adjudication, and measures of agreement like Cohen’s or Fleiss’ kappa. Label quality can influence model performance, and techniques such as active learning aim to label only the most informative samples.

Product and regulatory labeling refers to labels that convey essential information about a product, including ingredients, safety warnings, usage instructions, and compliance marks. Regulatory requirements vary by jurisdiction and industry, highlighting the need for accuracy, legibility, and accessibility in design.

Labeling theory, in sociology, describes how being labeled as deviant can influence identity and behavior, potentially reinforcing the labeled role. Originating with scholars such as Howard Becker, it emphasizes social processes, stigma, and the consequences of category membership.

In scientific contexts, labeling attaches markers to molecules or cells—radioisotopes or fluorescent dyes—to track, quantify, or observe processes. In metadata systems, labeling uses controlled vocabularies and taxonomies to improve interoperability, searchability, and data integration across platforms.

Ethical and practical considerations include label bias, inconsistency, privacy, and the need for transparent guidelines and provenance to ensure reliable and reproducible labeling outcomes.