Ϣ
INFORMATION AND CONTROL
1999 28 3 Vol.28 No.3 1999



ϢԶϵͳŻ㷨
ٻɡŻݻ
ժҪУùԶ෽˽ܣڶԶ෽ĸĽϣһȫµĻԶŻ㷨ַ˹ۣרҵְҵ飬ģֹ˼ά̣ԹȥƳɵϵͳ˸ĽŻַ˼롢ϵṹŻصܣ
ؼʣϢԶ࣬磬ѧϰ磬Ż㷨
ͼţTP14ױʶ룺B
NEURAL NETWORK OPTIMIZING ALGORITHM OF CHINESE
INFORMATION AUTO-CLASSIFICATION
DIAO Qian
(97300BA, Shanghai Jiaotong University, Shanghai 200030)
WANG Yongcheng
(Imtitute of Computer Application Technocogy, Shanghai Jiaotong University, Shanghai 200030)
ZHNAG Huihui
(Library of Shanghai Jiaotong University, Shanghai 200030)
AbstractAccording to two old methods of auto-classification of Chinese document, we provide an optimizing method based on the neural network algorithm. This method uses the theory of neural network, combines with the professional experiences of experts, imitates the process of manual classification, and optimizes the last two classification methods. The theory, architecture, and optimization of the new method are showed in the paper.
Key wordschinese information, automatic classification, neural network, competitive learning network, optimizing algorithm
1
׵ķ׵Ŀٲңһ׼ֳn࣬ƽnĲٶȣо˷࣬ÿƪױӦķţֹһִкǿԡԵķʱǣԱ60ͿʼԶоҹǽ£
2ѿԶϵͳ
Ǵ1986ͿʼԶоȿһĿƼףࣩԶϵͳϵͳBayes׼򣬻ǣÿһƪDi(ti1,ti2tim)ӦʣtimʾDiеһؼʣP(Cj|tikʾCjйؼtikֵƵʣʳһֵʱͽDiCj࣮׿ɷΪnࣺ{Ci}ΪDi(t1,t2,,tm)ΪCiĸΪ
P(Ci|D)(ʹʽĵʽ)
=P(D|Ci)P(Ci)/P(D)(P(Ci)P(D)ֱΪCiDĸ)
=P(1,t2,,tm)|Ci)P(Ci)/P(D)(tiΪУ)
=P(tj|Ci)P(Ci)/P(D)(øȫɼԹʽ)
=P(Ci|tj)P(tj)/P(D)(P(ti)P(t))
=P(t)/P(D)[P(Ci|tj)ݡ(ʹʽ)
ԣP(Ck/tj)=max[P(Ci|tj)]ɰDi(t1,t2,,tm)鵽Ck࣬ӦǱȽϺģȻP(Ci|tj)ڻĳһֵʱӦD鵽CiҲǺʵģǸϵͳԶݣϵͳȱȫģûժҪױؼʴʵ¿ԶϵͳϵͳĺɱؼŵḶ́ϵͳڴͳƻõνȱױʵۼֵ׵ķţѵ뱡ڡȱĻãһҪ㹻ɿҲܵƣڶͳƼ󽵵ķٶȣ۹ʽڼٶtiΪָʶȵ»õģʵٶģ⣬ڸϵͳУ׻Ŀȷò洢ڼеģܸûȴ洢ĿţṹȽϴ֣һͼҲãר˵಻ϸ£дߣ
1995꣬ķѧרΪ׵ĿվѧԺʦϺڶҽѧԺйרҺһѧרҵΪԶϵͳϵͳΪģ飮ԶȡױؼʣȻԶţкڵڶģ飮أڵõ׵ʻƷʺйרұƵġԶùؼʷתɶӦʵķţȻ󰴷йع涨ԾȨֵӦķΪѡţȨֵӦķŲֹһʱͬʱѡţȻڴ˻ϣʲͼйĿ涨γշţϵͳеġԶùؼʷͨ䷨ģרΪȷʵȨֵСԼУרҵְҵΪеݣһϵзôʷֳɷ棬ٸһȨֵӶˡԶùؼʷϵͳȽϴ󣬸׼ڶרҵ׵ķ࣮Ҳһȱ㣬ڷʵȨֵСԼоΪȷרҵľ鲢㵽Ҳһƫͬʱ޷ԸϵͳнһŻ
3Ż㷨Զϵͳ
Զϵͳ˼Լšȱ㣬һµġŻ㷨Զϵͳһ˵ϵͳԭ࣬ΪҵA1ũҵA2Ϊ޶ҲͨС࣬ʷֱΪеB1ӣB2B3ԴB4ˮB5ֱСÿһำֵֵʾÿСĹȣWij(j=1,,M; i=1,,N)ͼ1ʾ.


ͼ1ϵͼ
ĳһCгȡһױؼ{C1,C2,Cn}ʱͨͼ2ʾһģ飬õյ׷ţ
ͼ2ģУ1995ƵѧԶϵͳȨֵȫרΪöרȷֵ۶ȨֵֵŻȷôʵȨֵΣ


ͼ2ģͼ
ʵϵͳУABɷΪνṹֻǴСΣҿһݷôʿSʹø÷ϵͳʱSڵĿݷôʸΪ㣮ʹһʱ󣬿ͨʹƵʣһЩ÷ôʿSЩôʵţǶ1Ҳ1Զࣩһױؼ{C}гִôʣֱӸ׸ʵķţﵽ˵Ŀģ
ͼ2ʾģ飬ȵľŻ£ͼ1еľѧϰ磮ɱΪͼ3ʾ


ͼ3ѧϰ
Ϊ;㣮ΪNԪֱB1еB2ӣB3B4ԴB5ˮС࣮MԪֱA1ҵA2ũҵ࣮ȨȣΪ{ Wij}i=1,2,,N; j=1,2,,M(N=5,M=2)
(1)һѾֳɷŵPѡһC磬ȨֵŻױؼʼΪ{C}ͨɨôʱ{A}{B}󣬵֪{C}гB2B3X=(01100Ϊ1Ϊ0Ҳ˵ױг˵ӡʣ趨WijֵΪ

Щֵ趨йרҸݾ飬Դ׼趨֮ھѧϰķȨֵĳʼֵйأЩȨֵĳʼֵרƾְҵ趨׽ȷٷ࣬ͬʱѧϰŻܣ㾺ԪֵSj
(2)
(3)
Xcjʾ״̬
Xc1=1, Xc2=0, A1=1, A2=0, A=(1,0).
ϰʤΪԭSj(j=1,,M)ֵӦԪΪʤߣ״̬Ϊ1Ԫ״̬Ϊ0Sj=Si߶ӦԪΪʤߣ״̬Ӧ10Ϊ׷ص֮һͬһ׿вͬţҲ˵ͬһ׿Թֻ࣬ҪǺطͿˣ
ʤԪĸȨֵʽ(4)Ȩֲֵ䣮
(4)

mΪXԪֵ1ĸΪѧϰϵΪ0.01һΪ0.010.03йأ.ΪA1ʤA1=1A2=0ֻA1йصȨֵʽ(4)ɵ

ԣȨֵΪ

ѡȡPеظϲ裬ֱȨֵĵúСΪֹѧϰ·
ʽ4֪ʤԪjԪi֮ȨXiΪ1ʱWijʽ1ԼWij1ΪȨֵķ仯XiΪ0ʱΪȨֵСķ仯Ȩֵʼսڣ01֮仯ͼ4ʾУģʽXĵ23ԪΪ1뾺jԪȨֵ仯ԪֵΪ0ԪȨֵСķ仯
ͬһѧϰģʽṩѧϰһģʽǰӦľʤԪֵS󣬼ʤߵĵλңһѧϰģʽǳƵģʽṩʱҲʹͬһԪھлʤʱͿԸѧϰģʽʽ3ģʽڽ࣬ԾʤԪʾķʾͼͼ(4)ʾ⣬ȨֵεսԼʽ1֤ʽ(5)ʾ
(5)
ΪôȨֵֵϵͳ۽ѵľȨ̵ֵǰеͻ裬ѧϰõϢֲ洢Ԫͻϣ֮ѵõϢֲ洢СڴȣȨֵϣڶôʱֻж㣩Ųʱɽѵɴβ㣬¶ϣֱﵽķΪֹ

ͼ4ѧϰ̷ʾͼ
4
Ϣϵͳĳһר,ۺϢ(InternetϵϢ)ķҪһٽз,ܴﵽԤЧ.ҲǽҪео͸Ľ.,ŵѧ֪ʶķٷչѧƵ¼ʵĲӿ֣ԼѧרҵķչԶôʱʿⶼ費ϸ¡ƣһµԶϵͳѧϰȹܣʹ֮ڼΪԤ£ʵϵͳҸºƣӶϸߵܻˮƽչ
߼飺ٻŮ25꣬ʿоΪϢģʽʶȷо
ߵλٻϺͨѧ97300BA200030
ɣϺͨѧӦüо200030
ŻݻݣϺͨѧͼݡ200030
ο
1ɣϢϺϺͨѧ磬1990191193
2ɣš׵Զо[]ϺϺͨѧOAң1997
3ΰ˹ԭӦãպѧ磬1995133136
1998-10-26ո
