オープンアクセスを想定した日本語学術論文ファイルの自動判定

全文

(1)2006−FI−82（8） 2006−DD−54（8） 2006／3／22. 社団法人情報処理学会研究報告 IPSJ SIG Technical Report. ỼὊἩὅỴἁἍἋửे‫ܭ‬ẲẺଐஜᛖ‫ܖ‬ᘐᛯ૨ἧỳỶἽỉᐯѣЙ‫ܭ‬ᴾ ‫࢟ܤ‬ᴾ ᠗ᴾᵆʣኬʣ‫ܖٻ‬ᵇ*ᴾ ჽဋ௿፦ᵆᬮඕӨ‫ܖٻ‬ᵇᴾ ᴾ ൷ϋᴾ คᴾᵆ‫ٻ‬ி૨҄‫ܖٻ‬ᵇᴾ ʁ᣼᭗࣓ᵆ˺ૼ‫ܖ‬ᨈ‫ܖٻ‬ᵇᴾ᣼஛ᢊ‫܇‬ᵆᤧᢊዮӳ২ᘐᄂᆮ৑ᵇᴾ ɥဋ̲ɟᵆঅড፯‫ܖٻؽ‬ᵇᴾ *e-mail㧦[email protected] ࠝ࡯ࡊࡦࠕࠢ࠮ࠬⅣႺ߇ㅴዷߔࠆߦߟࠇ‫ߩࠣࡦࡆࠗࠞ࡯ࠕࡈ࡞࠮ޔ‬ᒻᑼߢ⥄ࠄߩ⎇ⓥᚑ ᨐࠍ౏㐿ߔࠆ⎇ⓥ⠪߇ᕆჇߒߡ޿ࠆ‫ߥ߁ࠃߩߘޕ‬ᚑᨐߪ‫ޔ‬ᓥ᧪ߩߔߴߡߩ࠙ࠚࡉࠍኻ⽎ߣ ߔࠆᬌ⚝ࠛࡦࠫࡦ߆ࠄ߽ࠕࠢ࠮ࠬ߇น⢻ߢ޽ࠆ߇‫ޔ‬ᬌ⚝⚿ᨐਛߩઁߩ߽ߩߦၒᴚߒߡߒ߹ ߁ߎߣ߇ᄙ޿‫⎇ᧄޔߢߎߘޕ‬ⓥߢߪ࠙ࠚࡉࠦࡦ࠹ࡦ࠷ਛ߆ࠄߩቇⴚ⺰ᢥ‫⺰ߪ޿ࠆ޽ޔ‬ᢥߦ Ḱߕࠆࠦࡦ࠹ࡦ࠷ࠍ್ቯߔࠆࠪࠬ࠹ࡓ᭴▽ࠍ⋡ᜰߒ‫ޔ‬SVM ߥߤ‫ޔ‬ᄙߊߩᚻᴺࠍ↪޿ߡ⥄േ ್ቯታ㛎ࠍⴕߞߚ‫⥄ޕ‬േ್ቯߩᚻ߇߆ࠅߣߥࠆዻᕈ⟲ߣߒߡߪࡈࠔࠗ࡞ਛߦ಴⃻ߔࠆ⺆ߣ ⚻㛎⊛ߥ࡞࡯࡞⟲ࠍ↪޿ߚ‫ޕ‬ታ㛎ߢߪ SVM ߢߪ㜞޿♖ᐲ‫ߪߢ࠭ࠗࡌࡉ࡯ࠗ࠽ޔ‬㜞޿ౣ⃻₸ ߇ᓧࠄࠇࠆߥߤ‫ޔ‬ฦᚻᴺߩ․ᕈ߇᣿ࠄ߆ߣߥߞߚ‫ޕ‬. Automatic identification of academic articles in Japanese PDF files towards open-access Teru AGATA (Asia University) * Michiko NOZUE (Railway Technical Research Institute) Atushi IKEUCHI (Daito Bunka Univ.) Takashi KUNO (Sakushingakuin University) Emi ISHIDA (Surugadai University) Shuichi UEDA (Keio University) *email: [email protected] As open-access becomes common, many researchers deposit their research products in a publicly accessible web (i.e. self-archiving). Although they are accessible from general search engines, massive other contents tend to hide them. The purpose of this research is to identify academic articles or quasi-articles from the entire web automatically. In this paper we conduct experiments on the performance of various classifiers and compare in terms of precision, recall, F-value. The classifiers used such attributes as terms appeared in PDF files and empirical rules. The diverse performance of each classifier discloses its characteristics.. 㸇䈲䈛䉄䈮 A. 䉥䊷䊒䊮䉝䉪䉶䉴䈫䈲ㄭᐕ䇮ቇⴚᖱႎᵹㅢ䈮䈍䈇䈩䇮䉥䊷䊒䊮䉝䉪䉶䉴䈮ኻ䈜䉎㑐ᔃ䈏㜞䉁䈦䈩䈇䉎䇯᭽䇱䈭࿅ ૕䉇୘ੱ䈏䇮⚵❱⊛䊶ᛛⴚ⊛஥㕙䈎䉌䇮䉥䊷䊒䊮䉝䉪䉶䉴䈱᥉෸䈫ᝄ⥝䈮ነਈ䈜䉎䉋䈉䈮䈭䈦䈩䈇䉎㪈䋩䇯2001 ᐕ䈮㐿௅䈘䉏䈢䉥䊷䊒䊮䉝䉪䉶䉴䈮㑐䈜䉎ળ ⼏䈱ᚑᨐ 䈪䈅䉎䇸Budapest Open Access Initiative䋨BOAI䋩䇹䈮䉋䉎⴫⃻ 䉕୫䉍䉎䈭䉌䈳䇮䉥䊷䊒䊮䉝䉪䉶䉴䈫䈲䇮"ቢో 䈮ήఘ䈪೙⚂䈱䈭䈇䉝䉪䉶䉴䈮䉋䈦䈩䇮ቇⴚ ᢥ₂䉕਎⇇ⷙᮨ䈪㔚ሶ⊛䈮ឭଏ䈜䉎䈖䈫 " 㪉㪀䈫䈘䉏䈩䈇䉎䇯䈢䈣䈚䇮䉥䊷䊒䊮䉝䉪䉶䉴⺰⠪䈱㑆䈪䈲䇮ᩏ⺒ઃ䈐ቇⴚ⺰ᢥ䈮㒢ቯ䈜䉎䈎䈬䈉䈎䇮㔀⹹಴ ᓟ䈱₈੍ᦼ㑆䉕⹺䉄䉎䈎䈬䈉䈎䈭䈬䈱ὐ䈮䈍䈇䈩ቯ⟵䈲৻᭽䈪䈲䈭䈇䇯. ৻⥸䈮䇮⎇ⓥ⠪䈏䊒䊧䊒䊥䊮䊃䈅䉎䈇䈲䊘䉴䊃䊒䊥䊮䊃䉕⥄ಽ䈱䉡䉢䊑䉰䉟䊃䈮౏㐿䈜䉎䇸䉶䊦䊐䉝䊷䉦䉟䊎䊮䉫䇹䇮ᯏ㑐䊥䊘䉳䊃䊥䈭䈬৻⥸ 䈱ੱ䈏䉝䉪䉶䉴䈪䈐䉎䉡䉢䊑䉰䉟䊃䈮⊓㍳䈜䉎 ᣇᑼ䇮⺰ᢥ䈱⪺⠪䈏಴ ⾌↪䉕ᡰᛄ䈇䇮⺒⠪ 䈲 ήᢱ䈪⺒䉃䈖䈫䈏䈪䈐䉎䇸䉥䊷䊒䊮䉝䉪䉶䉴䉳䊞䊷䊅䊦䇹䈱ឭଏ䈫䈇䈦䈢ታ⃻ᣇᴺ䈏⹺䉄䉌䉏䈩䈇䉎䇯ቇⴚ⺰ᢥ䈏䉥䊷䊒䊮䉝䉪䉶䉴䈫䈇䈉 ᒻ䈪ឭଏ䈘䉏䉎䈖䈫䈲䇮⎇ⓥ⠪䈮䈫䈦䈩䈲䇮⪺ ⠪䈫䈚䈩⥄䉌䈱⎇ⓥᚑᨐ䉕ᐢ▸࿐䈮ᵹㅢ䈘䈞䉎䈢䉄䈱೙ᐲ⊛ၮ⋚ 䈏⏕┙䈘䉏䉎䈖䈫䈪䈅䉍䇮⺒⠪䈮䈫䈦䈩䈲⎇ⓥ⾗Ḯ䉕ኈᤃ䈮೑↪䈪䈐䉎䈖䈫䈮䈧䈭䈏䉎䇯䈠䈱⚿ᨐ䇮䉥䊷䊒䊮䉝䉪䉶䉴䈪ឭଏ䈘䉏䈩䈇䉎⺰ᢥ䈲䈠䈉䈪䈭䈇⺰ᢥ䉋䉍䉅䇮 ⵍᒁ↪₸䈏㜞䈇䈫䈇䈉⎇ⓥᚑᨐ䈏ႎ๔䈘䉏䈩䈇䉎䉋䈉䈮䈠䈱ല↪䈏⹺䉄䉌䉏䈧䈧䈅䉎䇯䈭䈍䇮2006 ᐕ 2 ᦬ᤨὐ䈪䈲䇮9 ഀએ਄䈱ਥ. −55− - 1-.

(2) ⷐ䈭ቇⴚ㔀⹹䈏䇮䊘䉴䊃䊒䊥䊮䊃䈅䉎䈇䈲䊒䊧䊒䊥䊮䊃䉕䉶䊦䊐䉝䊷䉦䉟䊎䊮䉫䈜䉎䈖䈫䉕⸵น䈚䈩䈇䉎㪊㪀䇯 B. 䉥䊷䊒䊮䉝䉪䉶䉴䈭⺰ᢥ䈻䈱䉝䉪䉶䉴ᚻᲑ 䉥䊷䊒䊮䉝䉪䉶䉴䈻䈱ᦼᓙ䈫䈠䈱⿲൓䈲᣿ 䉌䈎䈪䈅䉎䈏䇮⃻⁁䈪䈲䇮䈅䉌䉉䉎ቇⴚ⺰ᢥ䈏䊈䉾䊃䊪䊷䉪਄䈮䈍䈇䈩ήఘ䈪೑↪䈪䈐䉎⸶䈪䈭䈇䇯䈫䈒䈮䇮⸒⺆㑆䇮෸䈶䇮ಽ㊁㑆䈱ᩰᏅ䈲㗼⪺䈪䈅䉎䇯వㅴ⊛䈭⹜䉂䈫䈚䈩⚫੺䈘䉏䉎䉅䈱䈲䈇䈝䉏䉅ᶏᄖ䈱੐଀䈪䈅䉍䇮ᚒ䈏࿖䈪䈲䇮 ⑼ቇᛛⴚᝄ⥝ᯏ᭴(JST) 㪋㪀䈱䇸⑼ቇᛛⴚᖱႎ ⊒ା䊶ᵹㅢ✚ว䉲䉴䊁䊛䇹(J-STAGE) 㪌㪀䈏䉥䊷䊒䊮䉝䉪䉶䉴䉳䊞䊷䊅䊦䉕 100 䉺䉟䊃䊦 એ਄ឭ ଏ䈚䈩䈇䉎䈏䇮ో૕䈫䈚䈩䉂䉏䈳ᣣᧄ䈱ቇⴚ 㔀⹹䈱㔚ሶൻ⁁ᴫ䈲ᭂ䉄䈩ㆃ䉏䈩䈍䉍䇮․䈮 ੱᢥ␠ળ⑼ቇಽ㊁䈪䈲 ᦠ⹹䊂䊷䉺䊔䊷䉴䈜䉌ቢ஻䈘䉏䈩䈇䈭䈇㗔ၞ䉅ዋ䈭䈒䈭䈇㪍㪀䇯 ৻ᣇ䇮઒䈮䇮ᄙ䈒䈱ቇⴚ⺰ᢥ䈏䉥䊷䊒䊮䉝䉪䉶䉴䈮䈭䈦䈢䈫䈚䈩䉅䇮䈠䈱ត⚝䇮౉ᚻ䈱໧㗴䈲 ଐὼ䈫䈚䈩ᱷ䈘䉏䉎䇯䈖䉏䈮ኻ䈚䈩䈲䇮ᣢ䈮䇸 Open Archive Initiative Protocol for Metadata Harvesting䋨OAI-PMH䋩䇹㪎㪀䈱䉋䈉䈭ⷙᩰ䉅⠨᩺䈘䉏䇮OAI-PMH 䈮Ḱ᜚䈚䈢ᯏ 㑐䊥䊘䉳䊃䊥䈱ᮮᢿᬌ⚝䉰䊷䊎䉴䉕ឭଏ䈜䉎 OAIster 䈭䈬䉅䈅䉎䇯 ⃻ ࿷䇮 604 ᯏ 㑐䈎䉌 6,509,564 ઙ䈏⊓㍳䈘䉏䈩䈇䉎䈏䇮䈠䉏䈏ታ 㓙䈮⎇ⓥ⠪䈮೑↪䈘䉏䈩䈇䉎䈎䈬䈉䈎䈲ᧂ⍮ ᢙ䈪䈅䉎䇯䉥䊷䊒䊮䉝䉪䉶䉴䉳䊞䊷䊅䊦䈱႐ว䇮䈠䉏䈏೑↪⠪䈮䈫䈦䈩ᣢ⍮䈱ᖱႎḮ䈪䈅䉏䈳䉝䉪䉶䉴䈲ኈᤃ䈭䉅䈱䈫䈭䉎䈏䇮⪺⠪䈱䉡䉢䊑䉰䉟䊃䈭䈬䈮䉋䈦䈩ឭଏ䈘䉏䉎⾗ᢱ䈮䈧䈇䈩䈲䇮 Google 䈭䈬䈱৻⥸⊛䈭䉰䊷䉼䉣䊮䉳䊮䉕↪䈇䈩ᬌ⚝䈜䉎䈾䈎ᚻ┙䈩䈲䈭䈇䇯䈚䈎䈚䈭䈏䉌䇮 ታ㓙䈮․ቯ䈱⪺⠪䇮⺰㗴䈱ቇⴚ⺰ᢥ䉕䉰䊷䉼䉣䊮䉳䊮䈪ត⚝䈚䉋䈉䈫䈜䉎႐ว䇮⤘ᄢ䈭ᬌ ⚝䊉䉟䉵䈮ㆣㆄ䈜䉎น⢻ᕈ䈏㜞䈇䈖䈫䈎䉌䇮䈭䉖䉌䈎䈱⛔৻䈘䉏䈢䉟䊮䉺䊷䊐䉢䊷䉴䈱ሽ࿷䈏 ᦼᓙ䈘䉏䉎䇯 ቇⴚ⺰ᢥ䉕ኻ⽎䈫䈚䈢䉰䊷䉼䉣䊮䉳䊮䈫䈚䈩䈲䇮⧷⺆࿤䈮䈲䇮CiteSeer.IST 㪏㪀䉇 Google Scholar㪐㪀╬䇮ઍ⴫⊛䈭䉅䈱䈏䈇䈒䈧䈎ሽ࿷䈜䉎䇯CiteCeer.IST 䈲䇮⸘▚ᯏ⑼ቇಽ㊁䉕ਛᔃ 䈫䈚䈢㒢ቯ⊛䈭෼㓸䈪䇮ⷙᮨ䈲䈠䉏䈾䈬ᄢ䈐䈒䈭䈇䇯Google Scholar 䈲ಽ㊁䉕㒢ቯ䈲䈚䈩䈇䈭䈇䈏䇮ᬌ⚝⚿ᨐ䈮䈲䇮ቇදળ䇮ቇⴚ಴ ␠䇮 ᄢቇ࿑ᦠ㙚䊂䊷䉺䊔䊷䉴䊔䊮䉻䈭䈬䈖䉏䉁䈪 ቇⴚᖱႎᵹㅢ䉕ᜂ䈦䈩䈐䈢䉰䉟䊃䈚䈎฽䉁䉏䈝䇮 ෼㓸ኻ⽎䈏㒢ቯ䈘䉏䈩䈇䉎䈫ផ᷹䈘䉏䉎㪈㪇㪀䇯⸒. ⪲䉕឵䈋䉏䈳䇮Google Scholar 䈪䈲䇮⪺⠪䈱䉡䉢䊑䉰䉟䊃䈮౏㐿䈘䉏䈢䉥䊷䊒䊮䉝䉪䉶䉴⺰ ᢥ䈲ᬌ⚝䈘䉏䈭䈇䇯䉁䈫䉄䉎䈫䇮ቇⴚᖱႎ䉕ኻ⽎䈫䈚䈢ᄢඨ䈱ᬌ ⚝䉰䊷䊎䉴䈲䇮(1)వㅴ⊛䈭䉰䊷䊎䉴䈲⧷⺆࿤ 䈏ਛᔃ䈪䈅䉍䇮䈠䈱䉋䈉䈭䉰䊷䊎䉴䈪䈜䉌䇮(2) ቇⴚ㔀⹹䉕ਛᔃ䈫䈚䈢ቇⴚᖱႎᵹㅢ䉕ᔒะ䈚䈩䈍䉍䇮⎇ⓥ⠪䈱୘ੱ䈱䉡䉢䊑䉰䉟䊃䉁䈪෼㓸 ኻ⽎䉕ᐢ䈕䈩䈲䈇䈭䈇䇯චಽ䈭䉝䉪䉶䉴ᚻᲑ 䈏ឭଏ䈘䉏䈩䈇䈭䈇⃻⁁䈮䈍䈇䈩䇮ో䉡䉢䊑䈎䉌䉥䊷䊒䊮䉝䉪䉶䉴䈭ቇⴚ⺰ᢥ䉕චಽ䈭♖ ᐲ䈪⼂೎䈪䈐䉎䈭䉌䈳䇮ታ↪⊛䈭ቇⴚᖱႎᬌ ⚝䉲䉴䊁䊛䈱᭴▽䈏น⢻䈫䈭䉎䈲䈝䈪䈅䉎䇯 એ਄䈱⁁ᴫ䉕〯䉁䈋䇮ᧄ⎇ⓥ䈲䇮ᣣᧄ⺆࿤ 䉕ኻ⽎䈫䈚䈩䈱䉡䉢䊑䉮䊮䊁䊮䉿䈎䉌䈱ቇⴚ⺰ ᢥ䈱⥄േ್ቯ䈮䈧䈇䈩 ᭽䇱䈭್ቯᚻᴺ䉕↪ 䈇䈢Ყセታ㛎䈎䉌ᬌ⸛䈚䈩䈇䈒䇯 C. 䊐䉤䊷䊙䊦䋯䉟䊮䊐䉤䊷䊙䊦ቇⴚᖱႎᵹㅢ Google Scholar 䈭䈬ቇⴚᖱႎ䈮․ൻ䈚䈢 ᬌ⚝䉣䊮䉳䊮䈲䇮䊐䉤䊷䊙䊦䈭ᵹㅢ⚻〝䈮ਸ਼ 䉎ቇⴚᖱႎ䉕ᔒะ䈚䈩䈇䉎䇯䈚䈎䈚䇮䊐䉤䊷䊙䊦䈭ቇⴚᖱႎ䈪䈲䈭䈒䈫䉅䇮⎇ⓥ⠪ห჻䈪੤ ឵䈘䉏䉎䊒䊧䊒䊥䊮䊃䇮ౝㇱ䈪૞ᚑ䈘䉏䉎⎇ⓥ ႎ๔䈱⾗ᢱ䈭䈬䇮䉟䊮䊐䉤䊷䊙䊦䉮䊚䊠䊆䉬䊷䉲䊢䊮䈱㊀ⷐᕈ䈲䇮ቇⴚᖱႎᵹㅢ䈱㗔ၞ䈪䈢䈶䈢䈶ᜰ៰䈘䉏䈩䈐䈢㪈㪈㪀䇯䈖䈖䈪䈲ቇⴚ⺰ᢥ䈱⥄േ್ቯ䈮䈧䈇䈩ᬌ⸛ 䈚䈩䈇䈒㓙䈮䇮䉟䊮䊐䉤䊷䊙䊦䉮䊚䊠䊆䉬䊷䉲䊢䊮䉕䉅൮฽䈚䈢ᒻ䈪䈱⥄േ್ቯ䉕⋡ᜰ䈚䈩䈇䈒䇯䈢䈣䈚䇮䉟䊮䊐䉤䊷䊙䊦䉮䊚䊠䊆䉬䊷䉲䊢䊮䈱ᒻ 䈲䈘䉁䈙䉁䈪䈅䉎䈢䉄䇮╙৻Ბ㓏䈫䈚䈩䇮䈖䈖䈪䈲⎇ⓥႎ๔䈭䈬䇮ቇⴚ⺰ᢥ䈮Ḱ䈛䉎䉮䊮䊁䊮䉿䋨એਅ䇮Ḱ⺰ᢥ䋩䉕್ቯኻ⽎䈮ട䈋䉎䇯. - 2−56−. ࿑䋱㩷䉡䉢䊑ਛ䈱ቇⴚ⺰ᢥ㩷.

(3) D. PDF 䊐䉜䉟䊦䉕ኻ⽎䈫䈚䈢⥄േ್ቯ ⃻࿷䇮ቇⴚ⺰ᢥ䈱ోᢥ䉕䊐䉜䉟䊦䈪ឭଏ䈜䉎႐ว 䇮䊐䉜䉟䊦 ᒻᑼ 䈮䈲 PDF䇮HTML䇮 XML䇮TeX䇮MS Word 䈭䈬䈏䈅䉎䇯䈭䈎䈪䉅䇮 PDF ᒻᑼ䈲ઁ䈱ᒻᑼ䈫Ყ䈼ᢥᦠ䈱䊧䉟䉝䉡䊃䉇䊂䉱䉟䊮䉕⛽ᜬ䈚䈢䉁䉁㑛ⷩ䈪䈐䇮㑛ⷩ᧦ઙ 䉕⸳ቯ䈜䉎䈖䈫䉅 น⢻䈪䈅䉎䈢䉄䇮ᦨ䉅৻⥸⊛ 䈭㈩Ꮣᒻᑼ䈫䈭䉍䇮࿑䋱䈱䉋䈉䈮䈾䈫䉖䈬䈱ቇⴚ ⺰ᢥ䈅䉎䈇䈲䈠䉏䈮Ḱ䈛䉎䉮䊮䊁䊮䉿䈲 PDF 䈫⠨䈋䉌䉏䉎䇯䈠䈖䈪䇮䉡䉢䊑䉮䊮䊁䊮䉿ਛ䈱 PDF 䊐䉜䉟䊦⟲䈎䉌䈱ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䈱್ ቯ䉕╙৻䈱⺖㗴䈫䈚䈩⠨䈋䈢䇯㩷. 䈱⛔৻䉕࿑䉍䇮䈪䈐䉎䈎䈑䉍್ቯ䈮ំ䉏䈏䈭䈇䉋䈉䈮䈚䈢䇯. 㸈 ታ㛎㓸ว䈱૞ᚑ. 䉟䊮䉺䊷䊈䉾䊃㩷㪧㪛㪝䊐䉜䉟䊦䈱᛽಴㩷. 㪧㪛㪝䊐䉜䉟䊦⟲㩷 ੱᚻ䈮䉋䉎ಽ㘃㩷. ቇⴚ㩷 A. 䌐䌄䌆䊐䉜䉟䊦䈱෼㓸 Ḱ⺰ᢥ㩷㕖⺰ᢥ㩷⺰ᢥ㩷 PDF 䊐䉜䉟䊦㓸ว䈱૞ᚑ䈮䈧䈇䈩䈲䇮2005 ᐕ 5 ᦬䈫ඨᐕᓟ䈱 2005 ᐕ 11 ᦬䈫䈱 2 ᐲ䈮䉒 ⥄േ್ቯ㩷䈢䈦䈩ⴕ䈦䈢䇯䉁䈝䇮ipadic2.5.1 䈱 6 䈧䈱ฬ ⹖ㄉᦠ䊐䉜䉟䊦䋨⸘ 213,020 ⺆䋩䈎䉌䇮䈠䉏䈡 ࿑㩷㩷㩷㩷 ᚻ㗅㩷 ࿑䋲㩷 ್ቯታ㛎䈱ᚻ㗅㩷䉏䇮9,750 ⺆䋨╙䋱࿁⋡䋩䇮10,250 ⺆䋨╙䋲࿁ ቇⴚ⺰ᢥ䈱್ቯⷙḰ䈫䈚䈩䇮(1)⺰ᢥ䈱ᒻᘒ ⋡䋩䉕ή૞ὑ᛽಴䈚䇮ฦ䇱䈱⺆䈮䈧䈇䈩䇮䉰䊷䉕䈫䈦䈩䈇䉎䇮(2) 䉺䉟䊃䊦䇮⪺⠪ฬ䇮ᚲዻᯏ㑐䉼䉣䊮䉳䊮䉕↪䈇䈩ᬌ⚝䉕ⴕ䈦䈢䇯䈠䈱㓙䇮ᬌ 䈏᣿⸥䈘䉏䈩䈇䉎䇮(3) ᒁ↪䇮ෳ⠨ᢥ₂䈏䈅䉎䇮 ⚝ኻ⽎䉕䇸PDF 䊐䉜䉟䊦䇹䋫䇸ᣣᧄ⺆䇹䈮㒢ቯ (4)1 ⺰ᢥ䈏 1 䊐䉜䉟䊦䈪᭴ᚑ䈘䉏䈩䈇䉎䇮(5)䋲䈜䉎䈫䈫䉅䈮䇮ฦᬌ⚝⺆䈱ᦨᄢ෼㓸ઙᢙ䈲 ਄ 䊕䊷䉳એ਄䈪䈅䉎䈭䈬䉕↪䈇䈢䇯Ḱ⺰ᢥ䈱ၮ ૏ 100 ઙ䉁䈪䈫䈚䈢䇯಴ജ⚿ᨐ䈱㊀ⶄ㒰෰ᓟ Ḱ䈲 ቇⴚ⺰ᢥ䈱್ቯၮḰ 䈮৻ㇱḩ䈢䈭䈇䉅䈱⇣䈭䉍 URL ઙᢙ䈲䇮䈠䉏䈡䉏䇮307,514 ઙ 䈱䇮ౝㇱะ䈔䈱䉟䊮䊐䉤䊷䊙䊦䈭⎇ⓥႎ๔䉕䋨╙䋱࿁⋡䋩䈫 441,598 ઙ䋨╙䋲࿁⋡䋩䈫䈭䉍䇮 ฽䉄䈢䇯ౕ૕⊛䈮䈲䇸⎇ⓥ䊉䊷䊃䈭䈬䈱♿ⷐ⺰ ฦ䇱䈱 URL䈮ኻ䈚䈩 PDF 䊐䉜䉟䊦䈱䉻䉡䊮䊨 ᢥ䇹䇸䋨ቇⴚ㔀⹹એᄖ䈪䈱䋩⎇ⓥႎ๔䇹䇸ญ㗡⊒ 䊷䊄䉕⹜䉂䈢䇯䉻䉡䊮䊨䊷䊄䈏ਇน⢻䈪䈅䈦䈢 ⴫ේⓂ䇹䇸ⶄᢙ䈱⺰ᢥ䈱㓸ว૕䇹䇸⺰ᢥ䊶ቇⴚ 䉅䈱䇮෸䈶䇮0 䊋䉟䊃䊶䊐䉜䉟䊦䇮⎕៊䊐䉜䉟䊦䇮 ᦠ䈱ᢿ 䇹䇸තᬺ⎇ⓥ䊶ୃ჻⺰ᢥ䇹䇸᝼ᬺᢎ᧚䇹 ᥧภൻ䊐䉜䉟䊦䇮PDF 䊐䉜䉟䊦䈪䈭䈇䉅䈱╬䉕䈭䈬䈪䈅䉎䇯㒰෰䈚䈢⚿ᨐ䇮╙䋱࿁෼㓸䈪䈲 248,314 ઙ䇮䈭䈍䇮ᣣᧄ⺆䈱䊐䉜䉟䊦䉕ኻ⽎䈮䈚䈩䈇䉎䈢 ╙䋲࿁෼㓸䈪䈲 349,971 ઙ䈱 PDF 䊐䉜䉟䊦㓸䉄䇮⺋್ቯ䈎䉌ᄖ࿖⺆䈱䊐䉜䉟䊦䉅৻ㇱ฽䉁䉏 ว䈏ᓧ䉌䉏䈢䇯䈘䉌䈮䇮╙䋱࿁⋡䈫╙䋲࿁⋡䈱䈩䈇䈢䈢䉄䇮䈖䉏䉌䈲㕖⺰ᢥ䈫್ቯ䈚䈢䇯㊀ⶄ䉕㒰෰䈚䈢䈫䈖䉐 599,673 ઙ䈫䈭䈦䈢䇯 C. ታ㛎㓸ว䈱․ᕈ B. ቇⴚ⺰ᢥ䇮Ḱ⺰ᢥ䈱್ቯ (1) ታ㛎㓸ว䈱ၮᧄ⊛䈭ዻᕈ ో૕䈱 PDF 䊐䉜䉟䊦㓸ว䈎䉌䇮20,000 ઙ䉕 ቇⴚ⺰ᢥ䈫⺰ᢥએᄖ䋨એਅ䇮䇸㕖⺰ᢥ䇹䈫␜ ή૞ὑ䈮᛽಴䈚䇮6 ੱ䈱್ቯ⠪䈏ฦ PDF 䊐䉜䈜䋩䈱䊐䉜䉟䊦ᢙ䇮䊐䉜䉟䊦䉰䉟䉵䇮䊕䊷䉳ᢙ䇮䉟䊦䈮䈧䈇䈩 ቇⴚ⺰ᢥ䇮Ḱ⺰ᢥ䇮㕖⺰ᢥ䈪䈅䉎䈎䉕್ቯ䈚䈢䇯12,000 ઙ䉕್ቯ䈚䈢ᤨὐ䈪䇮 ❑ဳ䈱ഀว䉕⴫䋱䈮␜䈜䇯⴫䋱䈎䉌䇮PDF 䊐䉜䉟䊦㓸ว 20,000 ઙਛ䈱⺰ᢥ䈱ഀว䈲 1.63% ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䈫್ቯ䈘䉏䈢 565 ઙ䈱䊐䉜䈫ૐ䈒䇮Ḱ⺰ᢥ䉕฽䉄䈩䉅 5%䈮ḩ䈢䈭䈇䈖䈫䈏䉟䊦䉕ᡷ䉄䈩 6 ੱోຬ䈏ౣ್ቯ䈚䇮್ቯၮḰ 䉒䈎䉎䇯⥄േ್ቯ䉕ⴕ䈉 ⴫䋱㩷ታ㛎㓸ว䈱ၮᧄ⊛䈭ዻᕈ ዻᕈ䈮㑐䈚䈩䈲䇮㕖Ᏹ䈮 ቇⴚ⺰ᢥ Ḱ⺰ᢥ 㕖⺰ᢥ ઙᢙ 㪊㪉㪍㪍㪉㪋㪈㪐㪃㪇㪌㪇 ஍䈦䈢㓸ว䈪䈅䉎䈫䈇䈋 ᐔဋ䊐䉜䉟䊦䉰䉟䉵㪋㪐㪎㪃㪍㪉㪉㪅㪎㪹㫐㫋㪼㫊㪋㪊㪍㪃㪎㪊㪍㪅㪋㪹㫐㫋㪼㫊㪉㪐㪌㪃㪈㪈㪈㪅㪐㪹㫐㫋㪼㫊䉎䇯ᐔဋ䊐䉜䉟䊦䉰䉟䉵䈲 ᐔဋ䊕䊷䉳ᢙ 㪈㪇㪅㪐㪋㫇㪸㪾㪼㫊㪈㪊㪅㪏㪍㫇㪸㪾㪼㫊㪍㪅㪏㪏㫇㪸㪾㪼㫊⺰ᢥ䇮Ḱ⺰ᢥ䈱ᣇ䈏㕖⺰ ❑ᒻ䈱ഀว 㪈㪇㪇㪅㪇㪇㩼㪐㪏㪅㪌㪋㩼㪐㪉㪅㪌㪇㩼 ᢥ䉋䉍䉅 ᄙ䈇䇯ᐔဋ䊕䊷 −57− - 3-.

(4) 䉳ᢙ䈲⺰ᢥ䉋䉍䉅Ḱ⺰ᢥ䈫್ቯ䈘䉏䈢䉅䈱䈱ᣇ 䈏ᄙ䈒䇮㕖⺰ᢥ䈲䈣䈇䈹ዋ䈭䈇䉅䈱䈏ᄙ䈇䈖䈫䈏䉒䈎䉎䇯❑ဳ䈱ഀว䈲䊕䊷䉳䈱❑ᮮᲧ䉕ข 䈦䈢䈫䈐䈮䇮❑㐳䈪䈅䉎 ഀว䈪䈅䉎䇯⺰ᢥ䈪䈲䈜䈼䈩䈱䊐䉜䉟䊦䈏❑㐳䈪䈅䉎䈖䈫䈏䉒䈎䉎䇯 (2) 䊄䊜䉟䊮䈱ಽᏓ 䊄䊜䉟䊮㪸㪺㪾㫆㪺㫆㫆㫉㫅㪼䈠䈱ઁ. ⴫䋲㩷㪡㪧䉰䊑䊄䊜䉟䊮䈱ಽᏓ ቇⴚ⺰ᢥ Ḱ⺰ᢥ 㕖⺰ᢥ 㪈㪎㪉㪌㪉㪅㪍㪇㩼㪉㪍㪐㪋㪊㪅㪊㪉㩼㪈㪃㪎㪋㪐㪐㪅㪊㪇㩼㪌㪉㪈㪌㪅㪐㪇㩼㪈㪇㪐㪈㪎㪅㪌㪌㩼㪈㪃㪏㪏㪐㪈㪇㪅㪇㪌㩼㪉㪐㪏㪅㪏㪎㩼㪋㪎㪎㪅㪌㪎㩼㪊㪃㪇㪉㪊㪈㪍㪅㪇㪏㩼㪉㪋㪎㪅㪊㪋㩼㪌㪐㪐㪅㪌㪇㩼㪉㪃㪈㪉㪎㪈㪈㪅㪊㪉㩼㪌㪈㪅㪌㪊㩼㪉㪇㪊㪅㪉㪉㩼㪈㪃㪊㪉㪉㪎㪅㪇㪊㩼㪋㪌㪈㪊㪅㪎㪍㩼㪈㪈㪎㪈㪏㪅㪏㪋㩼㪏㪃㪍㪏㪎㪋㪍㪅㪉㪈㩼㩷. 㩷㩷 ⴫䋲䈲෼㓸䈘䉏䈢 URL 䈱䊃䉾䊒䊧䊔䊦䊄䊜䉟䊮䈪ᄢ൓䉕භ䉄䈢 jp 䊄䊜䉟䊮䈮䈍䈔䉎䉰䊑䊄䊜䉟䊮䈱਄૏ 5 ૏䉁䈪䉕␜䈚䈢䉅䈱䈪䈅䉎䇯⴫䋲䈎䉌䇮ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䈱䉰䊑䊄䊜䉟䊮䈲 ac.jp 䈏ᄙ䈇䈖䈫䈏䉒䈎䉎䇯৻ᣇ䇮㕖⺰ᢥ䈱䊄䊜䉟䊮䇮䉰䊑䊄䊜䉟䊮䈲ᄙ⒳ᄙ᭽䈪䈅䉍䇮ਛ䈪䈲 co.jp 䈎䉌䈱䉅䈱䈏ᦨ䉅ᄙ䈇䇯 (3)⺰ᢥ䈱ਥ㗴ಽ㊁ ⴫䋳㩷⺰ᢥ䈱ਥ㗴ಽ㊁⺰ᢥ 㪥㪛㪚㪇㪇㩷✚⸥ 㪊㪊㪈㪇㪅㪈㩼㪈㪇㩷ືቇ 㪉㪇㪍㪅㪈㩼㪉㪇㩷ᱧผ 㪈㪉㪊㪅㪎㩼㪊㪇㩷␠ળ⑼ቇ 㪍㪋㪈㪐㪅㪍㩼㪋㪇㩷⥄ὼ⑼ቇ 㪍㪋㪈㪐㪅㪍㩼㪌㪇㩷ᛛⴚ䋯Ꮏቇ 㪏㪏㪉㪎㪅㪇㩼㪍㪇㩷↥ᬺ 㪉㪉㪍㪅㪎㩼㪎㪇㩷⧓ⴚ䋯⟤ⴚ 㪋㪈㪅㪉㩼㪏㪇㩷⸒⺆㪈㪇㪊㪅㪈㩼㪐㪇㩷ᢥቇ 㪐㪉㪅㪏㩼. Ḱ⺰ᢥ 㪉㪉㪊㪅㪌㩼㪈㪌㪉㪅㪋㩼㪉㪉㪊㪅㪌㩼㪈㪍㪈㪉㪌㪅㪏㩼㪈㪎㪌㪉㪏㪅㪇㩼㪈㪋㪌㪉㪊㪅㪉㩼㪍㪇㪐㪅㪍㩼㪈㪋㪉㪅㪉㩼㪌㪇㪅㪏㩼㪌㪇㪅㪏㩼. 䊦䈎䉌䊁䉨䉴䊃䊂䊷䉺䉕᛽಴䈜䉎ᔅⷐ䈏䈅䉎䇯 ᧄታ㛎䈪䈲䇮Xpdf 3.01pl2㩷㪈㪉㪀䉕↪䈇䈩䇮PDF 䊐䉜䉟䊦䈎䉌䊁䉨䉴䊃䊂䊷䉺䈱᛽಴䉕ⴕ䈦䈢䇯 PDF 䊐䉜䉟䊦䈲⴫␜䊶ශ೚ᤨ䈮䊧䉟䉝䉡䊃䈏 ౣ⃻น⢻䈭䊂䊷䉺ᒻᑼ䈪䈅䉍䇮ౝㇱ⊛䈮䈲ᢥ ᦠ᭴ㅧ䈱ᖱႎ䉕䉅଻ᜬ䈜䉎䈖䈫䈏น⢻䈪䈅䉎䇯䈚䈎䈚䈭䈏䉌䇮ᄙ䈒䈱 PDF 䊐䉜䉟䊦䈲න䈮䊧䉟䉝䉡䊃ᖱႎ䈚䈎ᜬ䈢䈭䈇䇯䈠䈱䈢䉄䇮䊁䉨䉴䊃䊂䊷䉺䈱᛽಴䉕ⴕ䈉䈫䇮Xpdf 䈲䊧䉟䉝䉡䊃䈱ᜰቯ 䈏䈘䉏䈩䈇䉎▎ᚲ䉕ᡷⴕ䊶ⓨ⊕䈻䈫ᄌ឵䈜䉎䈖䈫䈏ᄙ䈇䇯ᗧ࿑䈘䉏䈢ᡷⴕ䊶ⓨ⊕䈫 Xpdf 䈮䉋䈦䈩ᄌ឵䈘䉏䈢ᡷⴕ䊶ⓨ⊕䉕್೎䈜䉎䈖䈫䈲 ࿎㔍䈪䈅䉎䈢䉄䇮䈖䈖䈪䈲ᡷⴕ䊶ⓨ⊕䈱㒰෰╬䈱․ ೎䈭ᓟಣℂ䈲 ⴕ䉒䈝䇮ᄌ឵䈚䈢䊐䉜䉟䊦䉕䈠䈱䉁䉁↪䈇䈢䇯 ᣣᧄ⺆䈲 ⤔⌕⺆䈪䈅䉎䈢䉄䊁䉨䉴䊃䊂䊷䉺䉕䇮䊃䊷䉪䊮䋨ᢥሼ೉䉇න⺆䋩䈮ಽഀ䈜䉎ᔅⷐ 䈏䈅䉎䇯䊃䊷䉪䊮ൻ䈮䈲䇮ᒻᘒ⚛⸃ᨆ䉲䉴䊁䊛 MeCab 0.81㪈㪊㪀䈫 bigram 䉕↪䈇䈢䇯એਅ䈪䈲 ೨ ⠪ 䈮䉋䈦䈩 ᒻ ᘒ ⚛ 䈮 ಽ ഀ 䈘䉏䈢䉅䈱䈲 mecab䇮bigram 䈮䉋䈦䈩ಽഀ䈘䉏䈢䉅䈱䈲 bigram 䈫䈚䈩ෳᾖ䈜䉎䇯ಾ䉍಴䈚䈢䊃䊷䉪䊮䈎䉌䈱ㆬᛯ䈲ⴕ䉒䈝䇮䈜䈼䈩↪䈇䈢䇯. 㸉 ታ㛎ⅣႺ. ⺰ᢥ䊐䉜䉟䊦䈱ಽ㊁䈱ಽᏓ䉕⴫䋳䈮␜䈜䇯 ⴫䋳䉕䉂䉎䈫䇮⺰ᢥ䈱ಽ㊁䈮䈲ᛛⴚᎿቇ䈏ᄙ䈒䇮䈧䈇䈪⥄ὼ⑼ቇ䇮␠ળ⑼ቇ䈱㗅䈮䈭䈦䈩䈇䉎䇯䉡䉢䊑਄䈪౏㐿䈘䉏䈩䈇䉎⺰ᢥ䈮䇮⥄ὼ⑼ቇ䊶 ᛛⴚᎿቇ䈏ᄙ䈇䈱䈲 ੍ᗐ䈘䉏䉎䈏䇮৻ᣇ䈪䈠䉏䉌䈮඘ᢜ䈜䉎䈾䈬䈱ੱᢥ␠ળ⑼ቇಽ㊁䈱⺰ ᢥ䉅౏㐿䈘䉏䈩䈇䉎䈖䈫䈏䉒䈎䉎䇯⥄ὼ⑼ቇಽ ㊁䈫Ყセ䈚䈩䇮ోᢥ䊂䊷䉺䊔䊷䉴䈱ឭଏ䈏ㆃ 䉏䈩䈇䉎ੱᢥ␠ળ⑼ቇಽ㊁䈪䉅䉡䉢䊑਄䈪䈱 ⎇ⓥᚑᨐ䈱౏⴫䈲Ⓧᭂ⊛䈮ⴕ䉒䉏䈩䈇䉎䈫⠨ 䈋䉌䉏䉎䇯 (4) PDF 䊐䉜䉟䊦䈱䊁䉨䉴䊃ൻ䈫䊃䊷䉪䊮ൻ ್ቯታ㛎䈮↪䈇䉎್ቯེ䈲䇮PDF 䊐䉜䉟䊦䉕䇮⋥ធ䇮ᛒ䈉䈖䈫䈲䈪䈐䈭䈇䈢䉄䇮PDF 䊐䉜䉟. A. ್ቯ䈮↪䈇䈢ዻᕈ ቇⴚ⺰ᢥ䈱್ቯ䈲䊁䉨䉴䊃ಽ㘃䈱⺖㗴䈱৻ 䈧䈪䈅䉍䇮䉁䈝䊁䉨䉴䊃䈱ౝኈ䉕⠨ᘦ䈚䈢႐ว 䈮䇮PDF 䊐䉜䉟䊦ਛ䈮಴⃻䈜䉎⺆䉕ᚻ䈏䈎䉍䈫䈭䉎ዻᕈ䈫䈚䈩↪䈇䉎䈖䈫䈏⠨䈋䉌䉏䉎䇯䈖䈱಴ ⃻⺆䈮䉋䉎䉝䊒䊨䊷䉼䈲䇮ᓥ᧪䈱⎇ⓥᚑᨐ䉅 ᄙ䈒䇮ታ❣䈱䈅䉎䊁䉨䉴䊃ಽ㘃䈱್ቯེ䉕↪䈇䉎䈖䈫䈏䈪䈐䉎䇯䈢䈣䈚䇮಴⃻⺆䉝䊒䊨䊷䉼䈪䈲䇮 ዻᕈᢙ䈏Ყセ⊛ዋ䈭䈇 mecab 䈪䉅 77,814 ઙ 䈫䈭䉍䇮ዻᕈ䈱ㆬᛯ╬䉕ⴕ䉒䈭䈇႐ว䇮ᔕ↪น ⢻䈭್ቯེ䈲㒢ቯ䈘䉏䈩䈚䉁䈉䇯 ╩⠪䉌䈮䉋䉎వⴕ⎇ⓥ䈪䈲䇮䉡䉢䊑䉮䊮䊁䊮䉿䈎䉌䈱⺰ᢥ್ቯ䈲䇮⤘ᄢ䈭㕖⺰ᢥਛ䈎䉌䈱㕖Ᏹ䈮㒢䉌䉏䈢ᢙ䈱⺰ᢥ䉕᛽಴䈜䉎㔍ᐲ䈱㜞䈇್ቯⴕὑ䈪䈅䉍䇮಴⃻⺆䈣䈔䈎䉌䈱䉝䊒䊨䊷䉼䈣䈔䈪䈲ਇචಽ䈭䈖䈫䈏␜ໂ䈘䉏䈢㪈㪋㪀䇯䈠䈖䈪䇮಴⃻⺆䈮䉋䉎䉝䊒䊨䊷䉼䈣䈔䈪䈭䈒䇮ઁ䈱 ዻᕈ䉕↪䈇䈢䉝䊒䊨䊷䉼䉅ⴕ䈦䈢䇯 ⃻ᤨὐ䈪䈲ੱ䈏ᢥᦠ⟲䈎䉌⺰ᢥ䉕⺰ᢥ䈫䈚䈩್ᢿ䈜䉎ⴕὑ䈮ኻ䈜䉎૕♽⊛䈭⎇ⓥ䈏䈭䈇䈢䉄䇮⥄േ್ቯ 䈱ᚻ䈏䈎䉍䈫䈭䉎ዻᕈ⟲䉕᳿ ቯ䈜䉎䈖䈫䈲䈪䈐䈭䈇䇯䈚䈎䈚䇮⺰ᢥ䈱ౝኈ䈣䈔䈪䈭䈒䇮⺰ᢥ䈱䊧䉟䉝䉡䊃䇮᭴ㅧ⊛䈭․ᕈ䇮౉. −58− - 4-.

(5) ᚻర╬䈱䈘䉁䈙䉁䈭ⷐ⚛䉕✚ว⊛䈮↪䈇䉎䈲䈝䈪䈅䉎䇯䈖䈖䈪䈲䇮⚻㛎⊛䈮⺰ᢥ䈫㑐ଥ䈜䉎䈫⠨䈋䉌䉏䉎䈖䈫䇮䌐䌄䌆䊐䉜䉟䊦䈎䉌౉ᚻน⢻ 䈎䈧್ቯེ䈮ᛩ౉น⢻䈭䈖䈫䈱ੑὐ䉕᧦ઙ䈫䈚䈩એਅ䈱⴫䋴䈮䈅䉎䋴䉦䊁䉯䊥䇮䋱䋹ዻᕈ䋨䊦䊷䊦䋩䉕ណ↪䈚䈢䇯. 䳦. ⴫䋴㩷䊦䊷䊦䊔䊷䉴䈱್ቯ䈪↪䈇䈢ዻᕈ 䉦䊁䉯䊥 ዻᕈ 䊐䉜䉟䊦䉰䉟䉵 ᭴ 䊕䊷䉳ᢙ ㅧ 䊕䊷䉳䈱ᒻ䋨❑ဳ䈎ᮮဳ䈎䋩㪬㪩㪣䈏㪸㪺㪅㫁㫇䈪䈅䉎䈎 ౉ᚻర 㪬㪩㪣䈏㪾㫆㪅㫁㫇䈎䉌䈪䈅䉎䈎 ᢥᧃ䈏䈪䈅䉎⺞䈎䈪䈜䉁䈜⺞䈎 ળ⹤䈏಴䈩䈒䉎䈎 ᢥ ૕ 䋨ᢥᧃ䈮䇸䈰䇯䇹䇮䇸䇹䈏૶䉒䉏䈩䈇䉎䈎䋩䈵䉌䈏䈭䈏಴⃻䈜䉎䈎䋨ᄖ࿖⺆䈎䋩䇸⎇ⓥ䇹䇸ᢥ₂䇹 ಴ 䇸ⵍ㛎⠪䇹 ⃻ 䇸⺞ᩏ䇹䇸ಽᨆ䇹䇸ታ㛎䇹 ࠠ 䇸♿ⷐ䇹䇸⎇ⓥႎ๔䇹䇸⎇ⓥ䊉䊷䊃䇹䇸࿑䇹䇸⴫䇹 ࡢ 䇸ᧄⓂ䇹䇸ᧄ⎇ⓥ䇹䇸ᧄ⺰ᢥ䇹䇸⎇ⓥᚑᨐ䇹䇸⎇ⓥ⚿ᨐ䇹 ࠼ 䇸⠨ኤ䇹䇸⠨ᘦ䇹䇸ᒁ↪ᢥ₂䇹䇸ෳᾖᢥ₂䇹䇸ෳ⠨ᢥ₂䇹䇸ᄢቇ䇹䇸⎇ⓥᚲ䇹䇸⎇ⓥ䉶䊮䉺䊷䇹䳦. 䈖䉏䉌䈱ዻᕈ⟲䉕↪䈇䈢⥄േ್ቯ䈲䇮એਅ 䈪䈲䇸䊦䊷䊦䊔䊷䉴䉝䊒䊨䊷䉼䇹䈫䈚䈩䇮䇸಴⃻ ⺆䉝䊒䊨䊷䉼䇹䈫඙೎䈜䉎䇯䈖䈱䉋䈉䈭 ዻᕈ䉕↪䈇䉎䈢䉄䈮䈲䈅䉌䈎䈛䉄 ቇⴚ⺰ᢥ䈮㑐䈜䉎⍮⼂䈏ᔅⷐ䈪䈅䉎䈢䉄䇮᳢ ↪ᕈ䈮䈲ᰳ䈔䉎䈏䇮಴⃻⺆䉝䊒䊨䊷䉼䈫Ყセ䈚䈩䇮㕖Ᏹ䈮ዋ䈭䈇ዻᕈ䈎䉌䈱್ቯ䈪䈅䉎䈢䉄䇮㒢䉌䉏䈢ᯏ᪾⾗Ḯ䇮ᤨ㑆䈫䈇䈉ὐ䈎䉌䈲᦭ ೑䈭䉝䊒䊨䊷䉼䈫䈭䉎䇯 B. ್ቯᚻᴺ䈫䈠䈱ታⵝ ್ቯᕈ⢻䈱ะ਄䉕࿑䉍䇮ฦ್ቯᚻᴺ䈱․ᕈ Ყセ䈱䈢䉄䇮䈪䈐䉎䈣䈔᏷ᐢ䈇ⷰὐ䈎䉌ណ↪ 䈜䉎್ቯᚻᴺ䉕ᬌ⸛䈚䈢䇯⚿ᨐ䈫䈚䈩䇮಴⃻⺆䉝䊒䊨䊷䉼䈪䈲䇮䊁䉨䉴䊃ಽ㘃䈮䈍䈇䈩⹏ଔ䈱㜞䈇䇮SVM䇮AdaBoost䇮䈠䈚䈩䇮䉴䊌䊛䊐䉞䊦䉺䈫䈚䈩ᐢ䈒૶䉒䉏䈩䈇䉎䊔䉟䉳䉝䊮䊐䉞䊦䉺䈱 ਃ⒳㘃䈱್ቯᚻᴺ䈫䈚䈩↪䈇䈢䇯䊦䊷䊦䊔䊷䉴䉝䊒䊨䊷䉼䈪䈲䇮SVM䇮AdaBoost 䈮ട䈋䈩䇮䊅䉟䊷䊑䊔䉟䉵䇮᳿ቯᧁ(C4.5) 䇮䊜䉺್ቯ ᚻᴺ䈫䈚䈩 Voting 䈎䉌䈱್ቯ䉕ⴕ䈦䈢䇯 ฦ್ቯᚻᴺ䉕ታⵝ䈚䈢䉲䉴䊁䊛䈫䈚䈩䇮䊦䊷䊦䊔䊷䉴䉝䊒䊨䊷䉼䈪䈲䇮 Weka(Waikato. Environment for Knowledge Analysis)䉕 ↪䈇䈢䇯Weka 㪈㪌㪀䈲 Waikato ᄢቇ(䊆䊠䊷䉳䊷䊤䊮䊄)䈱ᯏ᪾ቇ⠌䉶䊮䉺䊷䉕ਛᔃ䈮 Java ⸒ ⺆䈪㐿⊒䈏ⴕ䉒䉏䈩䈇䉎䊂䊷䉺䊙䉟䊆䊮䉫䉿䊷䊦䈪䈅䉍䇮ᢙᄙ䈒䈱ᯏ᪾ቇ⠌䈮ၮ䈨䈒್ቯ ེ䉕ታⵝ䈚䈩䈇䉎䇯ේೣ⊛䈮䈲 Weka 3.4.7䇮 ᔅⷐ䈭႐㕙䈪䈲㐿⊒ 䈪䈅䉎 Weka 3.5.2 䉕 ↪䈇䈢䇯Weka 䈲಴⃻⺆䈮䉋䉎䉝䊒䊨䊷䉼䈪䈲 ૶↪䈚䈭䈎䈦䈢䇯䈖䉏䈲䇮೨ㅀ䈱䉋䈉䈮䇮ታ㛎㓸ว䈮䈍䈔䉎಴⃻⺆䈲㜞ᰴర䈱ዻᕈ⟲䈪䈅䉍䇮 Weka 䈪䈲ᛒ䈉䈖䈫䈏䈪䈐䈭䈎䈦䈢䈢䉄䈪䈅䉎䇯䈠䈱䈢䉄䇮಴⃻⺆䈮䉋䉎䉝䊒䊨䊷䉼䈪䈲ฦ್ ቯᚻᴺ䈮䈧䈇䈩⇣䈭䉎ታⵝ䉕↪䈇䈩䈇䉎䇯 (1) 䉰䊘䊷䊃䊔䉪䉺䊷䊙䉲䊮䉰䊘䊷䊃䊔䉪䉺䊷䊙䉲䊮䋨એਅ䇮SVM䋩䈲䇮 Vladimir N. Vapnik 䈮䉋䈦䈩ឭ᩺䈘䉏䈢 2 䉪䊤䉴ಽ㘃ེ䈱৻⒳䈪䈅䉎㪈㪍㪀䇯SVM 䈲ᱜ䈱଀䈫⽶䈱଀䉕ಽ㔌䈜䉎ᐔ㕙䉕᭴ᚑ䈚䇮䈠䈱ಽ㔌ᐔ 㕙䈮ᦨ䉅ㄭ䈇଀䋨䉰䊘䊷䊃䊔䉪䉺䊷䋩ห჻䈱䊙䊷䉳䊮䋨䉰䊘䊷䊃䊔䉪䉺䊷䈫ಽ㔌ᐔ㕙䈱ᦨዊ〒㔌䋩䉕ᦨᄢൻ䈜䉎䈖䈫䈪 ቇ⠌䈏ⴕ䉒䉏䉎䇯䈖䉏䉕䉦䊷䊈䊦㑐ᢙ䈮䉋䉍㜞ᰴరⓨ㑆䈮౮௝䈜䉎䈖䈫䈪䇮㜞ᰴరⓨ㑆䈮䈍䈇䈩䉅✢ᒻಽ㔌䉕ⴕ䈉䉅䈱䈪䈅䉎䇯㜞䈇᳢ൻᕈ⢻䉕ᜬ䈤䇮䉦䊷䊈䊦ᴺ䈮䉋䉍㕖 Ᏹ䈮㜞ᰴర䈱䊂䊷䉺䉕ᛒ䈉䈖䈫䈏䈪䈐䉎ὐ䈏․ ᓽ䈪䈅䉍䇮ᛩ౉䈜䉎ዻᕈᢙ䈏ᄙ䈒䈭䉍䈏䈤䈭䊁䉨䉴䊃 ಽ㘃䈮䈍䈇䈩䇮ᄙ䈒䈱ᔕ↪੐଀䈏䈅䉎䇯 ᓟㅀ䈱 AdaBoost 䈫䉅䈮䊁䉨䉴䊃ಽ㘃䈮䈍䈇䈩㜞䈇ᕈ⢻䉕␜䈜䈫䈘䉏䈩䈐䈢䇯 SVM 䈱ታⵝ䈫䈚䈩䈲䇮಴⃻⺆䈮㑐䈚䈩䈲䇮 SVMlight 6.01 㪈㪎㪀䉕↪䈇䇮䊦䊷䊦䊔䊷䉴䈪䈲䇮 Weka 䈎䉌 LIBSVM 2.81㪈㪏㪀䉕๭䈶಴䈜ᒻ䈪 ↪䈇䈢䇯 (2) AdaBoost 䊑䊷䉴䊁䉞䊮䉫(Boosting)ᴺ䈲䇮䊋䉾䉩䊮䉫 (Bagging) 䈫 ห ᭽ 䈮㓸 ࿅ ቇ ⠌ (ensemble learning)䈪䈅䉍䇮♖ᐲ䈏䈠䉏䈾䈬㜞䈒䈭䈇ⶄ ᢙ䈱ᒙቇ⠌ེ䈱⚵䉂ว䉒䈞ᣇ䇮㊀䉂ઃ䈔䉕 ቇ ⠌ 䈜䉎䈖䈫䈪 ᕈ ⢻ 䉕㜞䉄䉎 ᚻᴺ 䈪䈅䉎䇯 AdaBoost 䈲ೋᦼ䈱䊑䊷䉴䊁䉞䊮䉫ᴺ䉕ᡷ⦟䈚䈢䉅䈱䈪䇮Schapire 䈫 Singer㪈㪐㪀䈮䉋䉎ታ㛎䈪䈲䇮න⺆䈱᦭ή䈮䉋䉎ᒙቇ⠌ེ䉕 AdaBoost 䈮䉋䈦䈩 ⚵䉂 ว 䉒䈞䈢 ್ ቯ ེ 䈏ᦨ ㄭ ற ᴺ (k-NN ᴺ)䉇䊅䉟䊷䊑䊔䉟䉵 ᴺ䈮䉋䉎್ቯེ䉋䉍䉅㜞䈇್ቯᕈ⢻䉕␜䈚䈩䈇䉎䇯 SVM 䈫䈱㑐ଥ䈪䈲䊙䊷䉳䊮䈱ℂ⺰䈮ၮ䈨䈒 ὐ䈪䈲㕖Ᏹ䈮ૃㅢ䈦䈩䈇䉎䈏䇮䇸⇣䈭䉎䊉䊦䊛. −59− - 5-.

(6) 䈲⇣䈭䉎䊙䊷䉳䊮䈮ኻᔕ䈚䈉䉎䇹䇸ᔅⷐ䈭⸘▚ ㊂䈏㆑䈉䇹䇸㜞ᰴర䈪䈱ត⚝䉕ല₸⊛䈮ⴕ䈉䈢䉄䈮⇣䈭䉎䉝䊒䊨䊷䉼䉕↪䈇䈩䈇䉎䇹ὐ䈏⇣䈭䈦䈩䈇䉎㪉㪇㪀䇯 ੹࿁䈲 Boosting 䈱ታⵝ䈫䈚䈩䇮಴⃻⺆䈮㑐䈚䈩䈲 BoosTexter 䈪䈲 AdaBoost.MH 䉕↪ 䈇䈢䇯BoosTexter 㪉㪈㪀䉕↪䈇䈢ℂ↱䈲䇮಴⃻⺆䈲 mecab 䈪 70 ਁએ਄䈫ᰴరᢙ䈏ᄙ䈇䈏䇮ᰴ ర❗⚂䈭䈚䈮ᛒ䈉䈖䈫䈏䈪䈐䉎 AdaBoost ታⵝ 䈪䈅䉎䈎䉌䈪䈅䉎䇯BoosTexter 䈮ታⵝ䈘䉏䈢 AdaBoost ታ ⵝ 䈲 3 ⒳ 㘃䈪䈅䉎䈏䇮 AdaBoost.MH 䈏వⴕ⎇ⓥ䈪ઁ䈱 2 䈧䈮ఝ䉎 ⚿ᨐ䉕಴䈚䈩䈇䉎䈢䉄㪉㪉㪀䇮AdaBoost.MH 䉕↪ 䈇䈢䇯䊦䊷䊦䊔䊷䉴䈪䈲 Weka 䈱䊝䉳䊠䊷䊦䉕↪䈇䈢䇯ਔᣇ䈫䉅䈮 AdaBoost 䉕ᒙቇ⠌ེ䈫䈚䈩න৻䊉䊷䊄䈎䉌᭴ᚑ䈘䉏䉎᳿ቯᧁ䋨᳿ቯ ᩣ䋺decision stumps䋩䈫⚵䉂ว䉒䈞䇮➅䉍㄰䈚 ᢙ䉕 10 ࿁䇮100 ࿁䈮䈍䈔䉎್ቯ䉕ⴕ䈦䈢䇯㩷 (3) 䊅䉟䊷䊑䊔䉟䉵䋯䊔䉟䉳䉝䊮䊐䉞䊦䉺䊅䉟䊷䊑䊔䉟䉵 ಽ㘃ེ (naive Bayesian classifier)䈲䇮䊔䉟䉵䈱⏕₸䊝䊂䊦䈮ၮ䈨䈒䇮 න⚐䈭䉲䉴䊁䊛䈪䈅䉎䇯"naive"䈫䈲ฦዻᕈห ჻䈏⁛┙䈪䈅䉎䇮ẜ࿷⊛䈭ዻᕈ䈏ᓇ㗀䈚䈭䈇䈫䈇䈉઒ቯ䈎䉌ฬ䈨䈔䉌䉏䈢䉅䈱䈪䈅䉎䈏䇮න ⚐䈪䈅䉎䈏䉉䈋䈮ℂ⺰⊛䈭᜛ᒛ䈏ኈᤃ䈪䈅䉍䇮 ᔕ↪▸࿐䉅ᐢ䈇䇯䊔䉟䉳䉝䊮䊐䉞䊦䉺(Bayesian Filter)䈲䇮䊅䉟䊷䊑䊔䉟䉵 ᴺ䈱ᔕ↪䈚䈢䉅䈱䈪䈅䉎䇯⃻࿷䈪䈲䇮ਥ䈫䈚䈩㔚ሶ䊜䊷䊦䈱ਛ䈎䉌䉴䊌䊛䊜䊷䊦䉕ᬌ಴䈜䉎䉲䉴䊁䊛䈪↪䈇䉌䉏䈩䈍䉍䇮․䈮㵰A plan for spam㵱㪉㪊㪀䈏⊒⴫䈘䉏䈩એ㒠䇮ᄙ䈒䈱䉲䉴䊁䊛䈏㐿⊒䈘䉏䈩䈇䉎䇯䊔䉟䉳䉝䊮䊐䉞䊦䉺䉕䉴䊌䊛䊜䊷䊦䈮ᔕ↪䈜䉎႐ว䇮㕖䉴䊌䊛䊜䊷䊦䈫䉴䊌䊛䊜䊷䊦䈮಴⃻䈜䉎䊃䊷䉪䊮䈮ኻ䈜䉎䉴䊌䊛⏕₸䉕ቇ⠌䈚䇮䈠䈱䉴䊌䊛⏕₸䉕䉅䈫䈮䇮 ᣂ䈢䈮ฃା䈚䈢㔚ሶ䊜䊷䊦䈮ኻ䈚䈩䇮䉴䊌䊛䊜䊷䊦䈱ᬌ಴䉕ⴕ䈉䇯䉴䊌䊛䊜䊷䊦䈲䇮ౝኈ䈎䉌䉅್ቯ䈜䉎䈖䈫䈲 น⢻䈪䈅䉎䈏䇮ౝኈ䈣䈔䈪䈭䈒䇮ઙฬ䈱ᦠ䈐ᣇ䈭䈬䈠䈱䉴䉺䉟䊦䈏್ቯ䈮᦭ ല䈪䈅䉎䈫䈇䉒䉏䈩䈇䉎䇯䊔䉟䉳䉝䊮䊐䉞䊦䉺䈱ታⵝ䈫䈚䈩䈲䇮ᣣᧄ⺆䈮䉅 ኻ ᔕ น ⢻ 䈪䈅䉎 bsfilter 㪉㪋㪀䉕↪ 䈇䈢䇯 bsfliter 䈮䈲᦭ฬ䈭 Paul Graham ᣇᑼ䉅ታ ⵝ䈘䉏䈩䈇䉎䈏䇮䉋䉍♖ᐲ䈏㜞䈇䈫䈘䉏䉎 Gary Robinson-Fisher ᣇᑼ㪉㪌㪀䉕↪䈇䈢䇯 bsfilter 䈲䇮ฦ䊐䉜䉟䊦䈮ኻ䈚䈩䉴䊌䊛⏕₸ 䉕▚಴䈜䉎䇯䉴䊌䊛䊜䊷䊦್ቯ䈮↪䈇䉎႐ว 䈮䈲䇮䈖䈱⏕₸䈏㜞䈇䈫䉴䊌䊛䊜䊷䊦䈪䈅䉎䈫. ್ቯ䈘䉏䉎䈏䇮ᧄታ㛎䈪䈲䇮䇸㕖⺰ᢥ䇹䈫䈚䈩್ ቯ䈜䉎䇯 (4) ᳿ቯᧁ(C4.5) ᳿ቯᧁ(decision tree)䈲 น⺒ᕈ䈱㜞䈇ಽ 㘃ེ䈪䈅䉍䇮ㄭᐕ䈪䈲 AdaBoost 䈭䈬䈱㓸࿅ ቇ⠌䈱ᒙቇ⠌ེ䈫䈚䈩૶䉒䉏䉎䈖䈫䈏 ᄙ䈇䇯ઍ ⴫⊛䈭᳿ቯᧁ䉝䊦䉯䊥䉵䊛䈫䈚䈩䈲䇮CART 㪉㪍㪀䇮 ID3㪉㪎㪀䇮C4.5㪉㪏㪀䈏䈅䉎䈏䇮䈖䈖䈪䈲 Weka 䈮ታ ⵝ䈘䉏䈩䈇䉎 C4.5䋨䊝䉳䊠䊷䊦ฬ䈲 J48䋩䉕ቇ ⠌⚿ᨐ䈱ಽᨆ䈱䈢䉄䈮↪䈇䈢䇯 ታ㓙䈮੤Ꮕᬌቯ↪ቇ⠌㓸ว No.䋱䈎䉌ቇ⠌ 䈘䉏䈢᳿ቯᧁ䈱৻ㇱ䉕࿑䈮䈅䈕䈢䇯. ࿑䋳 ↢ᚑ䈘䉏䈢᳿ቯᧁ䈱৻ㇱ (5) 䊜䉺್ቯེ䋨Voting䋩䊦䊷䊦䊔䊷䉴䈱್ቯ䈪䈲 ⶄᢙ䈱್ቯེ䉕 ⚵䉂ว䉒䈞䈢䊜䉺್ቯེ䉅↪䈇䈢䇯䈖䉏䈲䇮ቇ ⴚ⺰ᢥ䈱್ቯ䈫䈇䈉㔍䈚䈇⺖㗴䈮ኻ䈚䈩䇮䈪䈐䉎䈣䈔ᄙ䈒䈱ⷰὐ䈎䉌䈱੍᷹䉕↪䈇್ቯᕈ⢻ ะ਄䉕⋡ᜰ䈜䈫䈇䈉Ⓧᭂ⊛䈭ℂ↱䈫䇮䊦䊷䊦䊔䊷䉴䈱್ቯ䈪䈲ዻᕈᢙ䈏ዋ䈭䈒⸘▚㊂䈏ዋ 䈭䈇䈢䉄䇮ⶄᢙ䈱್ቯེ䉕หᤨ䈮↪䈇䈩䉅ᯏ ᪾⊛䈭⾗Ḯ䈻䈱⽶⩄䈏䈭䈇䈫䈇䈉ᶖᭂ⊛䈭ℂ ↱䈎䉌ⴕ䈦䈢䇯 ታ㛎䈪䈲 Weka 䈱 Vote 䊝䉳䊠䊷䊦䉕↪䈇䇮⺰ᢥ್ቯ䈮ᄬᢌ䈚䈢 SVM 䉕㒰䈒䊅䉟䊷䊑䊔䉟䉵䇮AdaBoost䋨100䋩䇮᳿ቯᧁ䈱䋳䈧䈱್ቯེ 䈱੍᷹୯䉕⚵䉂ว䉒䈞ⴕ䈦䈢䇯 C. ⹏ଔዤᐲ 䈖䈱⎇ⓥ䈪䈲♖ᐲ(P)䇮ౣ⃻₸(R)䇮F ୯(F1䇮 F2)䉕⹏ଔ䈱䈢䉄䈮↪䈇䈢䇯 ♖ᐲ䈲䈬䉏䈣䈔 ᱜ⏕䈮ᬌ಴䈪䈐䈢䈎䉕䇮ౣ ⃻₸䈲䈬䉏䈣䈔✂⟜⊛䈮್ቯ䈪䈐䈢䈎䉕␜䈜䇯䈢䈣䈚䇮ේೣ⊛䈮♖ᐲ䈫ౣ⃻₸䈲෻Ყ଀䈱㑐. −60− - 6-.

(7) ଥ䈮䈅䉎䈢䉄䇮♖ᐲ䈣䈔䈅䉎䈇䈲 ౣ⃻₸䈣䈔䈎䉌⹏ଔ䈜䉎䈖䈫䈲䈪䈐䈭䈇䇯䈠䈖䈪䇮✚ว⊛ 䈭ᜰᮡ䈫䈚䈩 F ୯䉕↪䈇䈢䇯F ୯䈲㱍䈱୯䈮䉋䈦䈩䇮♖ᐲ䈫ౣ⃻₸䈱㊀䉂䉕ᄌ䈋䉎䈏䇮㱍=0.5 䈫䈚䈢႐ว䇮䈧䉁䉍♖ᐲ䈫ౣ⃻₸䈱⺞๺ᐔဋ䈱 ୯䈫䈚䈢䉅䈱䈏৻⥸⊛䈮䈲↪䈇䉌䉏䉎䇯䈚䈎䈚䈭䈏䉌䇮䈖䈱䉲䉴䊁䊛䈱⋡⊛䈪䈅䉎⺰ᢥ䈅䉎䈇䈲䈠䉏䈮Ḱ䈛䉎䉮䊮䊁䊮䉿䈱⥄േ್ቯ䉕ᗐቯ 䈚䈢႐ว䇮ౣ⃻₸䈏䉋䉍㊀ⷞ䈘䉏䉎䈫⠨䈋䉌䉏䉎䇯䈠䈱䈢䉄䇮䈖䈖䈪䈲㱍=0.5 䈱႐ว䈱 F1䇮 ౣ⃻₸䉕䉋䉍㊀ⷞ䈚䈢㱍䋽0.33 䈱 F2 䈱ਔᣇ 䉕Ყセ䈮↪䈇䈩䈇䉎䇯. 䉲䉴䊁䊛䈏್ቯ䈚䈢ᱜ ⸃ઙᢙ 䉲䉴䊁䊛䈏⺰ᢥ䈫 ್ቯ 䈚䈢ઙᢙ 䉲䉴䊁䊛䈏್ቯ䈚䈢ᱜ ⸃ઙᢙ R= ో⺰ᢥઙᢙ 1 F= 1 1 α ⋅ + (1 − α ) ⋅ P R P=. ᧄታ㛎䈪䈲䇮ቇ⠌↪䊶್ቯ↪䊂䊷䉺䉕ಽഀ 䈚䇮4 ੤Ꮕᬌቯ䉕ⴕ䈦䈢䈏䇮ฦ䊂䊷䉺䉶䉾䊃䈮䈍䈇䈩䇮ฦ⹏ଔዤᐲ䈱୯䉕᳞䉄䇮䈠䉏䉌䉕ᐔ ဋ䈚䈢୯䉕▚಴䈚䈢(macro-averaging)䇯. 㸊 ್ቯ⚿ᨐ ᧄ⎇ⓥ䈱ੑ䈧䈱⋡⊛䈮ኻᔕ䈚䈩䇮⥄േ್ቯ ታ㛎䈲䇮ੱᚻ䈪ቇⴚ⺰ᢥ䈫್ቯ䈚䈢䉅䈱䈱䉂䉕⺰ᢥ䈫䈚䈩ቇ⠌䈚್ቯ䈚䈢႐ว䈫䇮Ḱ⺰ᢥ䈫 ್ቯ䈘䉏䈢䉅䈱䉅฽䉄䈩⺰ᢥ䈫䈚䈩ቇ⠌䈚್ቯ 䈚䈢႐ว䈱ੑ䈧ⴕ䈦䈢䇯. ዻᕈ ಴ ⃻ ⺆. 丶. 䊦䊦. ዻᕈ ಴ ⃻ ⺆. 丶. 䊦䊦. ⴫䋵㩷ቇⴚ⺰ᢥ䈮㑐䈜䉎⥄േ್ቯ ᚻᴺ 䊃䊷䉪䊮 ♖ᐲ ౣ⃻₸ 㫄㪼㪺㪸㪹㪎㪌㪅㪇㩼㪉㪎㪅㪎㩼㪪㪭㪤㪹㫀㪾㫉㪸㫄㪎㪉㪅㪎㩼㪉㪎㪅㪋㩼㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪀㫄㪼㪺㪸㪹㪌㪉㪅㪈㩼㪋㪇㪅㪊㩼㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪀㫄㪼㪺㪸㪹㪌㪋㪅㪐㩼㪋㪇㪅㪎㩼㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪇㪀㫄㪼㪺㪸㪹㪍㪇㪅㪌㩼㪊㪏㪅㪊㩼㫄㪼㪺㪸㪹㪈㪈㪅㪊㩼㪏㪐㪅㪌㩼䊔䉟䉳䉝䊮䊐䉞䊦䉺㪹㫀㪾㫉㪸㫄㪋㪅㪎㩼㪐㪉㪅㪊㩼㪉㪊㪅㪊㩼㪏㪐㪅㪊㩼䊅䉟䊷䊑䊔䉟䉵㪋㪊㪅㪇㩼㪉㪊㪅㪍㩼 ᳿ቯᧁ㩿㪡㪋㪏㪀㪋㪉㪅㪉㩼㪊㪊㪅㪈㩼㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪀㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪀㪋㪍㪅㪎㩼㪊㪐㪅㪊㩼㪪㪭㪤㪇㪅㪇㩼㪇㪅㪇㩼㪭㫆㫋㪼㪋㪋㪅㪋㩼㪌㪊㪅㪎㩼. 㪝㪈୯ 㪝㪉୯ 㪅㪋㪇㪋㪅㪊㪌㪇㪅㪊㪐㪏㪅㪊㪋㪍㪅㪋㪌㪌㪅㪋㪊㪍㪅㪋㪍㪎㪅㪋㪋㪌㪅㪋㪍㪐㪅㪋㪊㪎㪅㪉㪇㪇㪅㪉㪎㪇㪅㪇㪐㪇㪅㪈㪉㪏㪅㪊㪎㪇㪅㪋㪌㪐㪅㪊㪇㪌㪅㪉㪎㪏㪅㪊㪎㪈㪅㪊㪌㪎㪅㪋㪉㪎㪅㪋㪈㪌㪥㪆㪘㪥㪆㪘㪅㪋㪏㪍㪅㪌㪇㪉. ⴫䋶㩷ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䈮㑐䈜䉎⥄േ್ቯ ᚻᴺ 䊃䊷䉪䊮 ♖ᐲ ౣ⃻₸ 㪝㪈୯ 㪝㪉୯ 㫄㪼㪺㪸㪹㪎㪋㪅㪉㩼㪋㪏㪅㪉㩼㪅㪌㪏㪋㪅㪌㪋㪍㪪㪭㪤㪹㫀㪾㫉㪸㫄㪎㪋㪅㪐㩼㪋㪎㪅㪏㩼㪅㪌㪏㪋㪅㪌㪋㪋㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪀㫄㪼㪺㪸㪹㪌㪏㪅㪇㩼㪋㪊㪅㪉㩼㪅㪋㪐㪌㪅㪋㪎㪉㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪀㫄㪼㪺㪸㪹㪍㪉㪅㪋㩼㪌㪈㪅㪌㩼㪅㪌㪍㪋㪅㪌㪋㪎㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪇㪀㫄㪼㪺㪸㪹㪍㪎㪅㪌㩼㪌㪈㪅㪊㩼㪅㪌㪏㪊㪅㪌㪌㪎㫄㪼㪺㪸㪹㪈㪈㪅㪊㩼㪏㪐㪅㪌㩼㪅㪉㪇㪇㪅㪉㪎㪇䊔䉟䉳䉝䊮䊐䉞䊦䉺㪹㫀㪾㫉㪸㫄㪈㪊㪅㪍㩼㪐㪈㪅㪊㩼㪅㪉㪊㪍㪅㪊㪈㪋䊅䉟䊷䊑䊔䉟䉵㪊㪍㪅㪊㩼㪎㪉㪅㪍㩼㪅㪋㪏㪋㪅㪌㪋㪌 ᳿ቯᧁ㩿㪡㪋㪏㪀㪍㪍㪅㪉㩼㪋㪋㪅㪌㩼㪅㪌㪊㪉㪅㪌㪇㪇㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪀㪍㪋㪅㪉㩼㪋㪊㪅㪊㩼㪅㪌㪈㪎㪅㪋㪏㪍㪘㪻㪸㪙㫆㫆㫊㫋㩿㪈㪇㪇㪀㪍㪌㪅㪋㩼㪋㪊㪅㪎㩼㪅㪌㪉㪋㪅㪋㪐㪈㪪㪭㪤㪍㪏㪅㪈㩼㪋㪊㪅㪍㩼㪅㪌㪊㪉㪅㪋㪐㪌㪭㫆㫋㪼㪌㪐㪅㪉㩼㪌㪌㪅㪈㩼㪅㪌㪎㪈㪅㪌㪍㪋. −61− - 7-. A. ቇⴚ⺰ᢥ䉕ኻ⽎䈫䈚䈢್ቯ⚿ᨐ ቇⴚ⺰ᢥ䉕 ኻ⽎䈫䈚䈢⥄േ್ቯ䈱್ቯ⚿ᨐ 䉕⴫䋵䈮␜䈚䈢䇯♖ᐲ䊶 ౣ⃻₸䈱ਔᣇ䈏หᤨ䈮 50%䉕⿧䈋䉎ᚻᴺ䈏䈭䈇ὐ䈎䉌䈲ో૕⊛䈮ච ಽ䈭್ቯᕈ⢻䈏ᓧ䉌䉏䈢䈫䈲䈇䈋䈭䈇䇯䈚䈎䈚䇮 ⹏ଔዤᐲ೎䈮䈲ฦᚻᴺ 䈱․ᓽ䈏㗼⪺䈮⃻䉏䈩䈍䉍⥝๧ᷓ 䈇⚿ᨐ 䈫䈇䈋䉎䇯 ♖ᐲ䊶ౣ⃻₸ 䈱ὐ䈎䉌䈲䇮SVM䋨mecab䋩䈱 ႐ว䈲 75%એ਄䈱㜞䈇 ♖ᐲ䈪⺰ᢥ䉕ᬌ಴䈪䈐䈢䇯䉁䈢䇮92% 䈫㜞ౣ⃻ ₸䈪䈅䉎䈱䈲䊔䉟䉳䉝䊮䊐䉞䊦䉺䋨bigram䋩䈪䈅䉎䈏♖ᐲ䈏ૐ䈒䇮♖ᐲ䈫䈱䊋䊤䊮䉴䈎䉌䈲䊦䊷䊦䊔䊷䉴䈱䊅䉟䊷䊑䊔䉟䉵䈏䈅䉎⒟ᐲ䈱♖ᐲ䈲 ⏕଻䈚䈧䈧䇮㜞䈇ౣ⃻ ₸䉕␜䈚䈢䈫䈇䈋䉎䇯䊜䉺 ್ቯ䉕㒰䈐♖ᐲ䉋䉍䉅㜞䈇ౣ⃻₸䉕␜䈚䈢䈱䈲䇮.

(8) 䊔䉟䉳䉝䊮䊐䉞䊦䉺䈫䊅䉟䊷䊑䊔䉟䉵䈪䈱್ቯ 䈣䈔䈪䈅䉎䇯 F ୯䈎䉌䉂䉎䈫䇮䊜䉺್ቯ䈱 Vote 䈏ᦨ㜞୯ 䈪䈅䉎䈏䇮䈠䉏䉕㒰䈔䈳䇮F1 ୯䈪䈲䇮಴⃻⺆䉝䊒䊨䊷䉼䈱 AdaBoost(1000) 䈱 mecab 䈏.469 䈫ᦨ㜞୯䉕␜䈚䈢䇯ౣ⃻₸㊀ⷞ䈱 F2 ୯䈪䈲䊅䉟䊷䊑䊔䉟䉵䈱୯䈏㜞䈇䇯 B. ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䉕ኻ⽎䈫䈚䈢್ቯ⚿ᨐ ቇⴚ⺰ᢥ䈫Ḱ⺰ᢥ䉕ኻ⽎䈫䈚䈢⥄േ್ቯ䈱 ⚿ᨐ䉕⴫䋶䈮␜䈜䇯⥄േ್ቯ䈮䈍䈇䈩඙೎䈏䈧䈐䈮䈒䈇䈫⠨䈋䉌䉏䉎Ḱ⺰ᢥ䉕฽䉄䉎䈖䈫䈪 ቇⴚ⺰ᢥ䈱䉂䉕ኻ⽎䈫䈚䈢႐ว䉋䉍䉅 ್ቯᕈ ⢻䈲ో૕⊛䈮ะ਄䈚䈩䈇䉎䇯F1 ୯䈲䈘䉌䈮♖ ᐲ䈏਄䈏䈦䈢 SVM 䈱 mecab 䈏䇮F2 ୯䈲䊜䉺 ್ቯ䈱୯䈏㜞䈇䇯 C. ⠨ኤ ቇⴚ⺰ᢥ䈱್ቯ䈱䊦䊷䊦䊔䊷䉴䈮䈍䈇䈩䇮 SVM 䈏䋱ઙ䉅⺰ᢥ䉕ᱜ䈚䈒್ቯ䈪䈐䈭䈎䈦䈢䈖䈫䈲䇮ዻᕈ䈱ㆬᛯ䈮໧㗴䈏䈅䈦䈢䈖䈫䈏 ේ࿃䈎䉅䈚䉏䈭䈇䈏䇮৻ᣇ䈪䇮ઁ䈱ᚻᴺ䈪䈲䈅䉎⒟ᐲ 䈱ᱜ⸃ᢙ䈏ᓧ䉌䉏䈢䈖䈫䉕 ⠨ᘦ䈜䉎䈫䇮․╩䈮 ଔ䈜䉎䇯ේ࿃䈱․ቯ䈱䈢䉄䈮䈲⹦⚦䈭ಽᨆ䉕 ⴕ䈉ᔅⷐ䈏䈅䉎䈏䇮䈖䈱ታ㛎䈱㔍ᐲ䈱㜞䈘䉕␜ ໂ䈜䉎৻଀䈫䈇䈋䉎䇯䉁䈢䇮䊦䊷䊦䊔䊷䉴䉝䊒䊨䊷䉼䈪䈲䋱䋹䈱ዻ ᕈ䈚䈎↪䈇䈩䈇䈭䈇䈮䉅䈎䈎䉒䉌䈝䇮಴⃻⺆䉝䊒䊨䊷䉼䈮Ყ䈼㆖⦡䈱䈭䈇䇮䈅䉎䈇䈲䈠䉏 એ਄䈱ᕈ⢻䉕␜䈚䈩䈇䉎䇯 ੹ᓟ䈲಴⃻⺆䉝䊒䊨䊷䉼䈮ኻ䈚䈩ẜ࿷⊛ᗧ ๧䉟䊮䊂䉨䉲䊮䉫䇮ਥᚑಽಽᨆ䈭䈬䈱ᚻᴺ䉕 ㆡ↪䈚䈢ᰴర❗⚂ಣℂ䉕ⴕ䈇䇮ᰴరᢙ䉕ᷫ䉌䈜䈫䈫䉅䈮䇮䊦䊷䊦䊔䊷䉴䉝䊒䊨䊷䉼䈫䈱⛔ว 䉕࿑䉍䇮್ቯᕈ⢻䈱ะ਄䉕⋡ᜰ䈜䇯䇼ᵈ䊶ᒁ↪ᢥ₂䇽㩷 1) ᤨታ⽎৻. "䉥䊷䊒䊮䉝䉪䉶䉴䈱േะ". ᖱႎ▤ℂ. Vol.47, No.9, 2004, p.616-624. 2) Budapest Open Access Initiative. 2002. <http://www.soros.org/openaccess/read.shtml> 3) http://romeo.eprints.org/stats.php 4) “⁛┙ⴕ᡽ᴺੱ⑼ቇᛛⴚᝄ⥝ᯏ᭴” <http://www.jst.go.jp/> 5) “J-STAGE” <http://www.jstage.jst.go.jp/> 6) 㜞ᧁర. “⎇ⓥ⠪䈮䈫䈦䈩䈱䉶䊦䊐䉝䊷䉦䉟䊎䊮䉫”. ᖱႎ䈱⑼ቇ䈫ᛛⴚ. Vol.55, No.10, 2005, p.434. 7) Open Archive Initiative Protocol for Metadata Harvesting. <http://www.openarchives.org/OAI/openarchiv esprotocol.html> 8) “CiteSeer.IST”. <http://citeseer.ist.psu.edu/cs> 9) “Google Scholar Beta” <http://scholar.google.com/> 10) Google Scholar 䈲 Beta 䈪䈅䉍䇮ᱜᑼ䉰䊷䊎䉴㐿ᆎᤨ䈮䈬䈉䈭䉎䈎䈲᣿䉌䈎䈪䈲䈭䈇 11) 䉻䉟䉝䊅䊶䉪䊧䊷䊮⪺䋨ᵤ↰⦟ᚑ⋙⸶䋩. ⷗䈋䈙䉎 ᄢቇ:⑼ቇ౒ห૕䈱⍮⼂䈱વ᠞. ᧲੩, ᢘᢥၴ, 1979, 260p. 12) 㵰Xpdf㵱<http://www.foolabs.com/xpdf/> 13) “MeCab: Yet Another Part-of-Speech and Morphological Analyzer” <http:// chasen.org/~taku/software/mecab/> 14) ⍹↰ᩕ⟤䈾䈎. “ᣣᧄ⺆ PDF 䊐䉜䉟䊦䉕ኻ⽎䈫䈚䈢ቇⴚ⺰ᢥ䈱⥄േ್ቯ”. ᣣᧄ࿑ᦠ㙚ᖱႎቇળ䋬 ਃ↰࿑ᦠ㙚䊶ᖱႎቇળวห⎇ⓥᄢળ⊒⴫ⷐ✁ 2005䋬ᘮᙥ⟵Ⴖᄢቇ䋬2005-10-22/23䋬p.165-168 15) “Weka” <http://www.cs.waikato .ac.nz/~ml/weka> 16) Vladimir N. Vapnik. The nature of statistical learning theory, 2nd ed. New York, Springer, xix, 314p., 2000 17) “SVMlight” <http://svmlight.joachims.org/> 18) “LIBSVM -- A Library for Support Vector Machines” <http://www.csie .ntu.edu.tw/~cjlin/libsvm/> 19) Schapire, R.E.; Singer, Y. “BoosTexter : A Boosting-based System for Text Categorization”, Machine Learning, Vol. 39, Number 2/3, p.135-168 (2000) 20) 䊣䉝䊑䊶䊐䊨䉟䊮䊄; 䊨䊋䊷䊃䊶䉲䊞䊏䊥⪺(቟ㇱ⋥᮸ ⸶). “䊑䊷䉴䊁䉞䊮䉫౉㐷”. ੱᎿ⍮⢻ቇળ⹹, Vol. 14, No. 5, 1999, p.771-780. 21) “BoosTexter” <http://www.research.att.com/sw/tools/BoosTe xter/> 22) R. E. Schapire. ”The boosting approach to machine learning: an overview.” MSRI workshop on nonlinear estimation and classification. 2001. p. 149-172 23) 䊘䊷䊦䊶䉫䊤䊊䊛⪺(Ꮉวผᦶ⋙⸶). ╙ 8 ┨䇸䉴䊌䊛䈻䈱ኻ╷䇹. 䇺䊊䉾䉦䊷䈫↹ኅ䋺䉮䊮䊏䊠䊷䉺ᤨઍ䈱 ഃㅧ⠪䈢䈤䇻 ᧲੩, 䉥䊷䊛␠, 2005, p.127-135 24) “bsfilter / bayesian spam filter”. <http://bsfilter.org/> 25) Gary Robinson. A statistical approach to the spam problem. <http://www. linuxjournal.com/article/6467> 26) Breiman, L.; Friedman, J.; Olshen, R.; Stone, C. Classification and regression trees. Belmont, Wadsworth International Group, 1984, 358p. 27) Quinlan, R. J. "Induction of decision trees". Machine Learning. Vol.1, No.1, p.81-106(1986) 28) J. R. 䉨䊮䊤䊮⪺(ฎᎹᐽ৻⋙⸶). AI 䈮䉋䉎䊂䊷䉺 ⸃ᨆ. ᧲੩, 䊃䉾䊌䊮, 1995, 293p.. −62− - 8-E.

(9)