Mis on pööratud normaliseerimine?

Pivoted normalization is a normalization method in information retrieval that adjusts the relevance assessment of documents by taking their length into account. The idea is to avoid bias towards overly long or overly short documents.

In classical models like BM25, the length of the document influences the assessment, and pivoted normalization introduces a parameter called pivot (usually the average document length in the collection), around which normalization occurs. The formula typically looks like this:

NormalizedLength = (1 - slope) * pivot + slope * actualLength

where slope is a parameter that controls the degree of normalization.

This helps to balance the influence of document length on relevance, improving the quality of ranking in search engines.