The term frequency normalisation smoothes the dependence between the WithinDocumentTermFrequency and the DocumentLength. (Singhal et. al., 1996) gives the following two reasons for the need for the term frequency normalisation:

As a consequence, an information retrieval system without term frequency normalisation produces biased results and favours long documents.

last edited 2005-05-01 15:58:03 by CraigMacdonald