overcategorization

Overcategorization is the tendency to assign items or concepts to an excessive number of categories, or to apply categories inappropriately, resulting in diluted meaning or fragmented knowledge. It contrasts with undercategorization, where too few categories are used, and with overly restrictive taxonomies. In humans, overcategorization can occur during learning or reasoning when features are interpreted as evidence for multiple overlapping categories, or when categories are subdivided beyond practical usefulness. This can lead to inconsistent judgments, label noise, and increased cognitive load.

In information systems, overcategorization refers to category proliferation in knowledge bases, taxonomies, or search indexes. Examples include labeling a single item with many narrowly defined tags or creating separate categories for nearly identical concepts. The result can hinder retrieval, complicate maintenance, and reduce user trust.

Causes include ambiguous or overlapping criteria, data spanning multiple contexts, and design choices that encourage fine-grained labeling. In machine learning, overcategorization can arise from training data with overlapping labels or from multi-label schemes without practical constraints.

Mitigation strategies involve establishing clear inclusion criteria for each category, employing hierarchical taxonomies to balance granularity, pruning rarely used categories, and applying validation or inter-annotator agreement processes. Evaluations of category stability and retrieval performance can help detect overcategorization. Related concepts include category learning, taxonomy design, label noise, overgeneralization, and category proliferation.