SUBJECT: Re : [ &NAME ] &NAME / recall &NUM point for &NAME &NAME classifier &EMAIL said : <QUOTE> Thanks for your reply . My main concern with this type of thresholding is that this is no longer a &NAME classifier , because the &NAME decision rule assigns each example to the class with the highest score ( posterior probability ) , regardless of the absolute values of the scores . <END_QUOTE> In fact , it is still a Naive &NAME classifier , though with an altered class prior . For example , if the class &NUM likelihood is twice the class &NUM likelihood , then an even prior ( &NUM for both classes ) will yield &CHAR ' &NUM ' label ( since the posterior for the class &NUM will be larger ) . But , a prior that weights class &NUM sufficiently heavily ( e.g. &NUM for class &NUM , &NUM for class &NUM ) will yield &CHAR ' &NUM ' label for the document . If you look at log-probabilities , &NAME &NAME is simply a linear classifier and the class prior is the ' bias ' or ' threshold ' term . Section &NUM of this paper might help to clarify things a bit : &WEBSITE &NAME