Noise Cleaning

In the semantic audit pipeline, noise cleaning occurs at multiple stages: after keyword collection (removing branded queries, nonsense terms, foreign-language phrases), after embedding generation (identifying outliers far from the centroid), and after clustering (removing clusters with 1-2 elements).

Analysis quality depends on data quality — garbage in, garbage out. Noise in SEO data includes: competitor branded keywords, foreign-language queries, spammy phrases, and lexical duplicates like 'probate' and 'probate ' (with space). Automated noise cleaning includes: deduplication, lowercasing, removing zero-volume phrases, and language filtering.