5-1227msg3.txt

运用贝叶斯网编的一个识别垃圾邮件的程序

Subject: more datum on - body / - one

i can't refuse jane anything . here be some more datum , just in from the british national corpus . ( well , actually only from a very small 1 % sample of it ) corpus size : 625 , 0 write and 182 , 0 speak word ( as define by claws ) proportional raw frequency for some common - one / - body word : write speak total anyone : anybody 57 : 31 17 : 32 74 : 63 someone : somebody 72 : 38 36 : 64 108 : 101 no-one : nobody 18 : 57 2 : 18 20 : 75 everyone : everybody 81 : 43 12 : 23 93 : 66 what i find interest here be the * absence * of any strong evidence for preference for - one or - body form ( except the pathological case of " no-one " ) when you look at the whole corpus , rather than separate out the speak and write component part . i ' m optimistically interpret that as evidence that in the corpus as a whole we ' ve get a good spread of textual variety . i ' ll repeat the analysis on the whole bnc when we ' ve finish build its index . . . lou burnard p . s . i look for " no one " too - - it appear only once , in a sentence begin " no one person would . . . "