patternoften

Patternoften is a metric used in pattern mining and data analysis to quantify how frequently a particular pattern occurs relative to a baseline expectation. It combines observed recurrence with a null model to identify patterns that persist across samples, time, or domains. The concept applies to text, time-series, genomics, and user data.

Formally, patternoften(P) is the ratio of the observed frequency of pattern P to its expected frequency under a predefined null model. If observed_count is how many times P appears and expected_count is the count predicted by the model, then patternoften(P) = observed_count / expected_count. Values > 1 indicate above-baseline recurrence; near 1 indicates baseline likelihood; < 1 suggests suppression.

Patternoften is a coinage from blending 'pattern' and 'often,' introduced in technical literature as a concise shorthand for comparative pattern recurrence. While not as widely formalized as other similarity measures, it was proposed to complement metrics such as support, confidence, and lift in pattern mining.

It is used to identify robust patterns in noisy data, compare cross-domain patterns, and flag patterns that persist under varying conditions. In natural language processing, patternoften can highlight phrases that remain common across corpora; in genomics and time-series analysis, it helps reveal motifs or cycles with persistent significance.

Its interpretation depends on the chosen null model; an inappropriate baseline can mislead. Patternoften relates to enrichment concepts such as lift, but emphasizes relative recurrence rather than absolute frequency and may require normalization for sample size, pattern length, or domain-specific characteristics.

See also pattern mining, lift, support, and anomaly detection.