參數調校模擬於高效率的色情網頁分類機制之應用

ഏઝᄎ᎖ܗૠ྽ᒳᇆΚNSC 93-2213-E-155-035

ୖኧፓਠኳᔕܭଯਏ౗ޑՅ௃ᆛ।ϩᜪᐒڋϐᔈҔ!

׵ᓪᇬ

P 1 P

Ǵഌ܍ד

P 2 P

Ǵ໳࠶ᑉ

P 3 P

ϡඵεᏢၗૻᆅ౛ࣴز܌!

P 1 P

longhow.lee@gmail.com

P 2 P

imcjluh@saturn.yzu.edu.tw

P 3 P

s937714@mail.yzu.edu.tw

ᄔा!

ӧᆛၡጲࠁว৖ޑϞВǴό྽ၗૻޑୀෳᆶٛ ݯς࿶ԋࣁቶݱ૸ፕޑ᝼ᚒǶӧӃ߻ޑࣴزύǴך ॺගрၮҔьБࣁ୷ᘵޑ಍ीᄽᆉݤܭՅ௃ᆛ। ϩᜪǴҁጇЎകଞჹ܌ගБݤޑس಍ୖኧ೽ϩǴග р΋ঁس಍ኳᔕޑፓਠᐒڋǴаයӧԋҁਏ੻ޑԵ ໆϐΠǴၲډଯਏ౗ޑس಍ਏૈǶ!

ᜢᗖӷ

;!ୖኧፓਠǴس಍ኳᔕǴԋҁਏ੻ǴьБ ϩଛǴᆛ।ϩᜪǴό྽ၗૻၸᘠ!

1

߻ق

ᒿ๱ᆛሞᆛၡޑጲࠁว৖Ǵ٬ளၗૻޑࢬ೯ߚ தِೲᆶቶݱǴय़ჹόᘐౢғޑӭኬϯၗૻǴӵՖ ଞჹᆛၡϣ৒଺፾྽ޑϩભᆶᆅ౛[02,04] ΋ޔࢂ ΋ঁ೏ቶݱ૸ፕޑ᝼ᚒǴЀځࢂό྽ၗૻޑୀෳᆶ ٛݯ׳ڙډ੝ձᜢݙǶό྽ޑᆛၡၗૻх֖Յ௃ǵ ኪΚǵ֎ࢥǵ።റ฻ሦୱǴߙϿԃϷٽูӧᎦԋ௲ ػਔௗڙډό྽ၗૻޑቹៜǴ৒ܰ܉্يЈޑ଼ӄ ว৖ǴᏤठΓ਱ޑୃৡǴ೷ԋ౲ӭޗ཮ᆶҍ࿾ୢᚒ ฻Ƕ!

World Wide Web Consortium [14]ගрPlatform for Internet Content Selection [10]Ǵڋۓᆛ।ϣ৒ϩ ભޑ኱ྗೕ਱Ǵϣ৒ගٮޣёа٩Ԝೕ਱ڋۓޑ኱ ᠸ(labels)ჹϣ৒ԾךуຏϩભǴ܈ࢂҗڐΚቷ୘(3P rd P party)ჹᆛၡ΢ණթޑϣ৒଺ϩભǶҗܭ཰ޣޑ୘཰ ճ੻ԵໆǴҗᆛၡϣ৒ගٮޣԾࡓޑ଺ݤԋਏό ᄆǴਥᏵཥуڵᏢޣLee et al. (2002&2003)!ޑፓ ࢗǴᆛၡϣ৒௦ҔPICSεऊ໻՞ 11.0%Ƕ܌аҗᆛ ၡᓎቨගٮޣǴٯӵISP཰ޣǵࡹ۬ǵᐒᜢიᡏ฻Ǵ ٰ଺ୀෳᆶٛጄ՟ЯࢂҞ߻ၨࣁёՉޑБԄǶਥᏵ Չࡹଣཥᆪֽ܌ࣴᔕޑᆛઠϣ৒ϩભ௢ቶीฝ \16^ǴЬाჹྣ୯ሞThe Internet Content Rating Association [13]!܌ڋۓຒ༼኱ྗǴஒᆛၡϣ৒٩ឦ ܄୔ϩࣁᇟقǵ܄ᆶᇘ៛ǵኪΚϷځд฻ϩᜪǴӆ КྣႝຎϩભޑբݤǴϩࣁද೯ભǵᇶᏤભǵߥៈ ભکज़ڋભ\01^Ǵܴुр౜ӧᆛઠ΢ޑϣ৒ᔈჹᔈ ՖᅿભձǴٯӵӧᇟق΢р౜ܴᡉᆶ܄࣬ᜢӷ౳Ǵ ஒ೏ϩᜪࣁज़ڋભǶ! ҁࣴزЬाଞჹό྽ᆛၡϣ৒ύനࣁҌᔲޑ ᆛၡՅ௃Ǵගр΋ঁୖኧፓਠኳᔕޑᐒڋǴҔܭь Бࣁ୷ᘵޑ಍ीᄽᆉݤϐՅ௃ᆛ।ϩᜪǶჴᡍ่݀ ᡉҢǴךॺ܌ගрޑୖኧፓਠኳᔕᐒڋӧԋҁਏ੻ Եໆ΢ǴёаԖਏ౗ගϲՅ௃ᆛ।ୀෳޑᆒዴ౗ (precision rate)ǴଛӝԾ୏ϯޑᆛ।ᇆ໣ǵၸᘠᆶϩ ᜪǴךॺёаזೲᇆ໣εໆޑՅ௃໵ӜൂǶ!

2

ьБՅ௃ᆛ।ϩᜪ

ךॺӧӃ߻ޑࣴزύǴගрၮҔьБࣁ୷ᘵޑ ಍ीᄽᆉݤܭՅ௃ᆛ।ϩᜪ\09^Ǵᙖҗჹᆛၡϣ৒ ύޑЎӷ೽ϩǴၮҔ಍ी௢ፕύޑьБ(chi-square) ϩଛ੝܄Ǵჹ؂΋ঁᆛ।ीᆉр΋ঁϟܭ 0 ډ 1 ϐ ໔ޑՅ௃ࡰ኱ॶ(Indicator Value, I value )Ǵӆ٩ᖏࣚ ॶ(threshold)ஒ I ॶጄൎϩ႖ԋΟঁ໔ᘐ(distinct)ޑ ୔໔Ǵ؂ঁ୔໔፟ϒ΋ঁᜪձǴϩձࢂՅ௃(Porn)ǵ ҂ዴۓ(Unsure)ᆶߚՅ௃(Non-Porn)Ƕ!

3/2!س಍ཷᢀ!

٬Ҕ Web Crawler வᆛሞᆛၡ΢׬ڗၟՅ௃ᜢᗖӷ ࣬ᜢޑᆛ।Ǵᒧڗ೽ҽᆛ।଺ࣁ૽ግၗ਑(training data set)Ǵ࿶җ Training ౢғ Token DatabaseǹځᎩ ᆛ।߾଺ࣁෳ၂ၗ਑(testing data set)Ƕךॺ೸ၸ΋ ঁ಍ीᄽᆉݤޑьБϩଛ(chi-square)੝܄Ǵჹ؂ঁ ෳ၂ᆛ।؃р΋ঁՅ௃ࡰ኱ॶ(Indicator Value)Ǵа ਥᏵԜՅ௃ࡰ኱ॶஒᆛ।୔ϩࣁՅ௃(Porn)ǵ҂ዴ ۓ(Unsure)аϷߚՅ௃(Non-Porn)ΟঁᅿᜪǴӆஒځ ύޑՅ௃ᆛ।уΕ URL Black List ύǶ!

!

კ 1!س಍ࢎᄬკ!

3/3 Յ௃ࡰ኱ॶ(Indicator Value)!

๏ۓ΋ঁෳ၂ᆛ।Ǵךॺ२Ӄѐନ HTML tagsǴฅࡕ٩ Token Database ଺ᘐຒǴפډр౜ӧ

Token Databaseύൂ΋ঁձӷຒǴаϷ؂ঁӷຒჹ ᔈޑՅ௃໼ӛॶ(Porn Tendency)ǴҗܭՅ௃໼ӛॶ ࢂ࣬ჹК౗ޑཷۺǴ܌аՅ௃໼ӛॶࣁϟܭ 0 ډ 1 ϐ໔ޑჴኧॶǶௗ๱Ǵךॺ௦Ҕ Gary Robinson ග рޑ Spam Filtering БݤύޑьБϩଛ[03,12]ޑ੝ ܄่ӝ p-value ౢғՅ௃ࡰ኱ॶǴҔډޑ൳ঁБำ ԄӵΠ܌ҢǶ!

( )

1

2* ln

, 2

n

R

=

C

−

⎛

_⎜

−

f w

⎝

∏

n

⎠

⎞

⎟

(1)

( )

(

)

1

2 * ln

1

, 2

n

G

=

C

−

⎛

_⎜

−

−

f

⎝

∏

w

n

⎞

⎟

⎠

(2)

1

2

R G

I

=

+ −

(3) !

ځύf(w)ࣁঁձӷຒޑՅ௃໼ӛॶǴn

ࣁ ӷ ຒ ኧ Ҟ Ǵ ࣁ ന ε ཷ ՟ ՗ ी ໆ Ǵ n ࣁ߈՟Ծҗࡋ 2nޑьБϩଛ![03]ǴCP -1 P

ࣁInverse Chi-square FunctionǴ೸ၸCP

-1

P

ᆉрٰޑॶ ߾ࣁp-valueǶБำԄ(1)ύǴ௦Ҕ಍ी௢ፕύޑ Likelihood Ratio Test![03]Ǵ၀ଷ೛ᔠۓӵკ 8 ܌ҢǴ RࣁoঁӷຒՅ௃໼ӛॶճҔьБϩଛ੝܄่ӝ؃ рޑp-valueǴж߄ӧ຀คଷ೛(null hypothesis )Ǻᇡ ࣁ೭٤f(w)۶ԜᐱҥǴԶև౜ߚ֡Ϭϩଛޑచҹ ΠǴԖӭϿᐒ౗ᡉ๱௢ፕளޕ೭ঁ຀คଷ೛όԋ ҥǶӧჴሞ௃ݩύǴҔٰඔॊՅ௃ᆛ।ϣ৒ޑӷຒ ࢂ࿶தՔᒿр౜Ǵ٠ߚ۶ԜᐱҥǴ܌аךॺႣයଷ ೛ᔠۓޑ่݀ࣁܔ๊຀คଷ೛Ƕᜪ՟ޑБำԄ(2)Ǵ ߾ஒf(w)ڗжԋ 1-f(w)߄ҢߚՅ௃໼ӛॶǴGࣁoঁ ঁձӷຒޑߚՅ௃໼ӛॶ؃рޑp-valueǶ྽RǵG ٿޣ؃рϐࡕǴջё೸ၸБำԄ(3)ٰ؃ளՅ௃ࡰ኱ ॶ(Indicator Value)Ƕ! 3/4!س಍ୖኧ! ӧ᏾ঁьБՅ௃ϩᜪޑࣴزύǴԖٿঁख़ाޑ س಍ୖኧǴϩձࢂߐᘖॶ(Threshold Pair)ک Effective TokensޑኧҞǶߐᘖॶҔٰ،ۓӵՖஒ I-value ϟܭ 0ډ 1 ޑჴኧ୔໔Ǵ୔႖ԋ”Յ௃”ǵ”҂ዴۓ”ک”ߚ Յ௃”Οঁᅿᜪჹᔈޑη୔໔ǶќѦǴӧ๏ۓ΋ঁෳ ၂ᆛ।Ǵӧᘐֹຒפډр౜ӧ၀ᆛ।ޑൂ΋ঁձӷ ຒаϷჹᔈޑՅ௃໼ӛॶϐࡕǴᔈ྽่ӝӭϿኧҞ ޑӷຒ(Effective Tokens)೸ၸьБϩଛޑ੝܄ीᆉ рՅ௃኱ॶǴஒ཮ࢂ೭ঁьБࣁ୷ᘵޑ಍ीᄽᆉݤ ૈցԖਏ౗ၮբޑᜢᗖǶ!

3

ୖኧፓਠኳᔕ

ךॺගр΋ঁୖኧፓ௲ޑኳᔕᐒڋǴଞჹ೭ٿ ঁख़ाޑس಍ୖኧǴ೸ၸኳᔕޑ่݀פډന٫ޑୖ ኧ೛ۓǴаΠஒଞჹ᏾ঁኳᔕޑၸำၟ่݀଺၁ಒ ޑϟಏǶ

!

3.1 ኳᔕᕉნ ךॺ٬Ҕ MATLAB 6.5 ӧ Windows ᕉნΠ଺س ಍ኳᔕǴ᏾ঁኳᔕޑᕉნ೛ۓӵΠǺଷ೛ςޕԖ“Յ

∏

n w f( )

∏

−2ln f(w)

௃”ǵ “҂ዴۓ”аϷ“ߚՅ௃”೭Οঁᅿᜪޑᆛ।Ӛ 3000฽ǴᕴӅ 9000 ฽ᆛ।຾Չኳᔕෳ၂ǹќѦǴ ଷ೛“Յ௃”ᆛ।೽ϩӧ࿶ၸᘐຒϐࡕளډޑ Tokens வ 0.2 ډ 1.0 ޑ֡ϬϩଛύᒿᐒౢғՅ௃໼ӛॶ (f(w))ǹӕኬӦǴ“ߚՅ௃”ᆛ।வ 0 ډ 0.8 ޑ֡Ϭϩ ଛύᒿᐒౢғ Token ޑՅ௃໼ӛॶǴԶ“҂ዴۓ”೭ ঁ೽ϩ߾ࢂவ 0.2 ډ 0.8 ޑ֡Ϭϩଛύᒿᐒ๏ϒՅ ௃໼ӛॶǶࣁΑफ़ե໶ኧౢғᏔ೷ԋޑୃৡǴךॺ ჹ؂΋ঁՅ௃໼ӛॶ(f(w))଺ 5 ԛޑ໶ኧౢғǴฅࡕ ӆڗځѳ֡྽բՅ௃௃ӛॶǶ 3.2 ߐᘖॶ(Threshold Pair) ߐᘖॶޑ೛ۓஒቹៜډس಍ޑᆒዴ౗(precision rate)Ǵ೭္ޑᆒዴ౗ࣁس಍҅ዴղᘐϩᜪޑᆛ।ኧ Ҟନаςޕ๏ۓϩᜪޑᆛ।ኧҞǶس಍ޑߐᘖॶ (Threshold Pair)җٿঁ೽ϩಔԋǴϩձࢂߐᘖΠࣚ (Lower Bound, L)کߐᘖ΢ࣚ(Upper Bound, U)ǴΨ ൩ࢂᇥჹܭ؂΋ಔThreshold Pair (TBiB)ǴTBi=(LB BiB, UBi B)Ƕ ኳᔕύޑLBiBҗ 0.5 ډ 1.0Ǵ؂ԛሀቚ 0.5ǹUBiB߾җ 1.0 ډ 0.5 Ǵ ؂ ԛ ሀ ෧ 0.5 Ƕ ᖐ ٯ ٰ ᇥ TB5B=(LB5B,UB5B)=(0.25,0.75)Ǵ߄ 1 ࣁኳᔕޑ่݀Ǵ؂ঁಒ ਱(cell)ж߄ޑࢂӧ၀੝ۓ௃ݩΠޑᆒዴ౗Ǵനࡕ΋ ӈࣁӧ၀Threshold PairڰۓΠǴჹόӕޑEffective TokensኧҞޑѳ֡ᆒዴ౗Ƕҗኳᔕ่݀ёޕǴӧTB7B= (0.35,0.65)ਔǴ᏾ᡏޑѳ֡ᆒዴ౗ࣁ 99.95Ǵࢂ܌а όӕޑߐᘖॶ೛ۓύჹس಍ᆒዴ౗Զقനӳޑ่ ݀Ƕ ߄ 1 ኳᔕ่݀ Threshold Pair Number of Effective Tokens TB₁B TB₂B TB₃B TB₄B TB₅B TB₆B TB₇B TB₈B TB₉B 50 72.41 84.25 91.01 94.71 97.03 98.36 99.14 98.88 93.05 100 88.32 94.96 97.79 98.94 99.50 99.81 99.94 99.97 99.85 150 95.20 98.49 99.41 99.76 99.89 99.97 100 100 100 200 98.22 99.51 99.83 99.95 99.98 99.99 100 100 100 250 99.41 99.87 99.97 99.99 100 100 100 100 100 300 99.76 99.94 99.99 100 100 100 100 100 100 350 99.93 99.98 100 100 100 100 100 100 100 400 99.98 100 100 100 100 100 100 100 100 450 99.99 100 100 100 100 100 100 100 100 500 99.99 100 100 100 100 100 100 100 100 550 100 100 100 100 100 100 100 100 100 600 100 100 100 100 100 100 100 100 100 650 100 100 100 100 100 100 100 100 100 700 100 100 100 100 100 100 100 100 100 750 100 100 100 100 100 100 100 100 100 800 100 100 100 100 100 100 100 100 100 850 100 100 100 100 100 100 100 100 100 900 100 100 100 100 100 100 100 100 100 950 100 100 100 100 100 100 100 100 100 1000 100 100 100 100 100 100 100 100 100 Mean 97.66 98.85 99.4 99.67 99.82 99.91 99.95 99.94 99.65

3.3 Effective TokensޑኧҞʳ

ךॺஒ߄ 1 ύޑ Effective Tokens ޑኧҞӧόӕ ޑ Threshold Pair ܌ளޑᆒዴ౗଺ѳ֡Ǵёаளډ؂ ঁ Effective Tokens ޑኧҞჹѳ֡ᆒዴ౗ޑϩթᖿ ༈Ǵӵკ 3 ܌ҢǴךॺёаว౜ѳ֡ᆒዴ౗ᒿ๱ Effective TokensޑኧҞԋሀቚޑᖿ༈Ǵ྽ Effective TokensޑኧҞຬၸ 150 ਔǴѳ֡ᆒዴ౗ςຬၸ 99%Ƕ! კ!3!ᆒዴ౗ϩթᖿ༈კ! ԜѦǴךॺΨว౜྽ၨӭޑEffective TokensҔ ٰीᆉՅ௃ࡰ኱ॶǴஒ઻຤ၨӭޑीᆉਔ໔ǶࣁΑ ёаԖਏ៾ᑽس಍ᆒዴ౗کीᆉਔ໔Ǵךॺ௦Ҕа ΠޑБำԄीᆉCost-EffectivenessΔځύoࣁuplfot ঁኧǴAPRBnBж߄ӧ܌ԖޑThreshold PairΠޑѳ֡ᆒ

ዴ౗ǹACTBnB߾ࢂѳ֡ޑीᆉਔ໔Ƕ n n n

APR

Cost

Effectiveness

ACT

−

=

(4) ߄ 2 ࣁ Cost-Effectiveness ޑीᆉ่݀Ǵ྽ Effective Tokens ޑ ኧ Ҟ ࣁ 150 ਔ Ǵ ё а ว ౜ Cost-Effectiveness ฻ܭ 7.2467Ǵ೭ࢂ܌Ԗ௃׎ύന ӳޑ่݀Ǵᡉฅޑ೭ࢂኳᔕӧคݤளޕՖᒏന٫่ ݀(optimal solution)௃׎ΠǴளډޑ߈՟ന٫่݀ (near optimal solution)Ƕ

3.4ʳ ന٫ୖኧ೛ۓʳ

ਥ Ᏽ ΢ ॊ ޑ ኳ ᔕ ่ ݀ Ǵ ך ॺ ё а ว ౜ ྽ Effective Tokens ޑኧҞࣁ 150 ਔЪ Threshold Pair

ࣁ(0.35,0.65)ਔǴӧѳ֡ᆒዴ౗کीᆉਔ໔઻຤ϐ ໔Ǵٿ࣬៾ᑽϐΠࣁനڀԋҁਏ੻ޑୖኧ೛ۓǶ! კ 3 ࣁ Effective Tokens ޑኧҞ 150 ਔޑኳᔕޔ БკǶځύ”Յ௃”೽ϩޑ 3000 ฽ᆛ।Ǵ࿶ၸीᆉϐ ࡕޑՅ௃ࡰ኱ॶ(I-value)Ǵ๊ε೽ϩ߈՟ 1(εܭ 0.8)ǹӕኬӦǴ”ߚՅ௃”ᆛ।ޑ 3000 ฽ǴՅ௃ࡰ኱ ॶ(I-value)߾ε೽ϩ᎞߈ 0(λܭ 0.2)ǹԶ”҂ዴۓ” ޑ 3000 ฽ᆛ।߾Տܭ 0.5Ƕऩஒ Threshold Pair ೛ۓ ࣁ(0.35,0.65)Ǵ߾س಍ޑᆒዴ౗ࣁ 100%Ƕ! კ 4!Effective Tokens ޑኧҞࣁ 150 ޑኳᔕޔБკ!

4

س಍ჴբᆶෳ၂

ךॺӧ Linux Fedora Core 2 ᕉნΠǴ௦Ҕ MySQL 4.0.20ǵApache 1.3.31ǵ PHP 5.0.1 ک GNU

Wget 1.9 [06] ஒьБՅ௃ϩᜪޑس಍ჴբǴ٠ஒख़ ा ޑس಍ୖኧ Threshold Pair ೛ ࣁ (0.35,0.65) ǹ Effective TokensޑኧҞ೛ۓࣁ 150ǴΨ൩ࢂᇥ྽๏ ۓ΋ঁෳ၂ᆛ।Ǵऩᘐຒϐࡕޑൂ΋ӷຒόຬၸ 150 ঁਔǴ߾ஒځჹᔈޑՅ௃໼ӛॶӄ೽Ҕٰीᆉ Յ௃ࡰ኱ॶǹऩᘐຒϐࡕຬၸ 150 ঁൂ΋ӷຒǴ߾ ஒჹᔈޑՅ௃໼ӛॶҗεठλ௨ׇǴڗ߻ 150 ঁ่ ӝьБϩଛीᆉ I-valueǶ ךॺஒ٣Ӄᇆ໣ޑՅ௃ᜢᗖӷǴҔำԄԾ୏ᒡ Εཛྷ൨ЇᔏǴஒཛྷ൨่݀ύᆛ֟ޑᆛ।׬ڗӣٰǴ ځύ೽ϩ྽բ૽ግၗ਑Ǵќ΋೽ϩ྽բෳ၂ၗ਑Ƕ ךॺӧෳ၂ၗ਑ύǴᒧڗύमЎᆛ।Ӛ 500 ฽Ǵ٣ ӃΓπᔠࢗ๏ۓ҅ዴϩᜪǴฅࡕ๏س಍բϩᜪղ ᘐǶჴᡍ่݀ӵ߄ 3 ܌ҢǶ

߄ 3 ϩᜪᆒዴ౗ ᇟق ᆛ।ኧҞ ղᘐ ҅ዴ ղᘐ ᒱᇤ ᆒዴ౗ मЎ 500 486 14 97.2% ύЎ 500 482 18 96.4% ӧमЎ೽ϩԖ 597 ฽ղᘐ҅ዴǴ14 ฽ղᘐᒱ ᇤǴᆒዴ౗ࣁ 97.2%ǹύЎ߾Ԗ 482 ฽ղᘐ҅ዴǴ 18 ฽ղᘐᒱᇤǴᆒዴ౗ࣁ 96.4%Ƕղᘐᒱᇤޑ 32 ฽ᆛ।Ǵ࿶ၸ၁ಒᔠࢗว౜Ǵᆛ।ϣ৒ЬाаკТ БԄև౜Ǵཱུλ೽ϩޑЎӷၗૻǴس಍คݤԖਏղ ᘐϩᜪǴ೭೽ҽёૈሡा٬ҔቹႽೀ౛ޑמೌᇶշ ϩ݋Ƕၟ Lee!et al. (2002 & 2003)!аᜪઓ࿶ᆛၡ଺ Ўӷϣ৒ϩ݋ޑࣴز࣬КၨǴӧᆒዴ౗΢ගϲऊ

2.2% (मЎ೽ϩ)ǴӆޣǴךॺΨёаೀ౛ύЎᆛ।Ƕ

߄!2 Cost-Effectiveness ޑीᆉ่݀

Number of Effective Tokens

Average Precision Rate of Each Threshold Pair

Average Computational Time (Seconds) Cost Effectiveness 50 92.09 13.0656 7.0483 100 97.68 13.5624 7.2023 150 99.19 13.6876 7.2467 200 99.72 14.1438 7.0504 250 99.92 14.2312 7.0504 300 99.97 14.5000 6.8949 350 99.99 14.7720 6.7689 400 99.99 14.8970 6.7121 450 99.99 14.5844 6.8560 500 99.99 14.6780 6.8122 550 100 15.4312 6.4804 600 100 32.9344 3.0363 650 100 41.6688 2.3999 700 100 42.9876 2.3263 750 100 43.4344 2.3023 800 100 43.8158 2.2823 850 100 44.5750 2.2434 900 100 44.5970 2.2423 950 100 48.6126 2.0571 1000 100 57.7188 1.7325

5

่ፕ

ךॺගр΋ঁୖኧፓਠኳᔕޑᐒڋǴҔܭӃ߻ ගрޑаьБՅ௃ᆛ।ϩᜪس಍Ǵჴᡍ่݀ᡉҢ࿶ җኳᔕ่݀ளډޑ߈՟ന٫ୖኧ೛ۓॶǴӧԋҁਏ ੻ޑԵໆΠǴёа٬᏾ঁس಍ၲډଯਏ౗Ǵس಍ޑ ᆒዴ౗ଯၲ 96%а΢Ƕ ҂ٰךॺीฝೀ౛όӕሦୱޑό྽ၗૻх֖። റǵኪΚǵҍ࿾฻฻ǶќѦǴନΑύЎǵमЎϐѦǴ Ѻᆉჹځдᇟس଺ೀ౛Ǵх֖ܿ٥ᇟس(ВЎǵᗬ Ў)ǵՋኻᇟس(ቺЎǵՋ੤УЎǵကεճЎ฻฻)Ǵ

Ҿკᇆ໣׳ֹ᏾ޑό྽ၗૻ໵ӜൂǴёа଺ࣁᆛၡ ϣ৒ϩભᆶၗૻၸᘠޑ٩ᏵǶ

ठᖴ

ҁ ࣴ ز җ ୯ ࣽ ཮ ஑ ᚒ ࣴ ز ी ฝ ! NSC 93-2213-E-155-035ံշǴ੝ԜTठTᖴǶ!

ୖԵЎ᝘!

1. B. J. Bushman, and J. Cantor, “Media Ratings for Violence and Sex,” American Psychologist, Vol. 58, No. 2, pp. 130-141.

2. J. M. Balkin, B. S. Noveck, and K. Roosevelt, “Filtering the Internet: A Best Practices Model,” Information Society Project at Yale Law School, September 1999, pp. 1-38.

3. G.. Casella, and R. L. Berger, (2001), Statistical Inference (2P

nd

P

edition), TWadsworth Pub. Co.T

4. S. Goodwin, and R. Vidgen, “Content, Content, Everywhere…..Time to Stop and Think? The Process of Web Content Management,” Computing & Control Engineering Journal , Vol. 13, No. 2, 2002, pp.66-70.

5. Government Information Office, Republic of China, Project For Promoting Internet Content Rating System, available online at

TU

http://info.gio.gov.tw/public/Attachment/451214 545571.docUT

6. GNU Wget Project, available online at

TU

http://www.gnu.org/software/wget/wget.htmlUT

7. P. Y. Lee, S.C. Hui, and A.C.M. Fong, "Neural Networks for Web Content Filtering," IEEE Intelligent Systems, Vol. 17, No. 5, 2002, pp.48-57.

8. P. Y. Lee, S.C. Hui, and A.C.M. Fong, "A Structural and Content-Based Analysis for Web

Filtering," Internet Research: Electronic Networking Applications and Policy, Vol. 13, No. 1, 2003, pp. 27-37.

9. L. H. Lee, ”A Web Content Classification for Pornographic Blacklist Generation,” Master thesis, Department of Information Management, Yuan Ze University, 2005.

10. Platform for Internet Content Selection (PICS), available online at TUhttp://www.w3c.org/PICS/UT .

11. G.. Robinson, “A Statistical Approach to the Spam Problem,” Linux journal, Vol. 2003, issue 107. pp.1-9.

12. G.. Robinson (May 3, 2004), “Why Chi? Motivations for the Use of Fisher’s Inverse Chi-Square Procedure in Spam Classification, Version 0.93,” available online at

TU

http://www.garyrobinson.net/2004/05/why_chi.h tmlUT

13. The Internet Contenting Rating Association (ICRA), available online atTUhttp://www.icra.org/UT

14. World Wide Web Consortium (W3C), available online at TUhttp://www.w3c.org/UT .