3347329100d3019e246b0c2f867a2f6b.ppt
- Количество слайдов: 42
CATAR-文獻內容探勘 具 Ø簡介 Ø安裝 Ø使用 Ø解讀 Ø案例 曾元顯 國立臺灣師範大學 2016/02/16
文獻內容分析 -簡介 • 相關學科: – Bibliometrics、 Scientometrics、 Infometrics – Content analysis in social science • 相關期刊 – JASIST, Scientometrics, Journal of Infometrics • 相關會議 – ISSI: International Society for Scientometrics and Infometrics – STI: Science and Technology Indicators 2
文獻內容分析 -動機 • 專利的前案分析,希望能在 半天 內完成 – 引自類比 IC設計製造公司副總經理說法 • 鑑往知來、避開重複、促進創新 • 擷取重點、評估形勢、規劃策略 • 找出特定作者、機構 – 邀請投稿、演講、審查、合作、求助 • 科技政策分析與決策輔助 – 作者、機構、國家生產力分析 – 績效評鑑、經費分配 3
文獻內容自動分析 • 長期目標: – 自動掃描相關文獻,分析、組織、呈現 – 提供探索、指引,以供後續驗證、決策 • 相關研究 – Structured Abstract in library science (1987) – Automated structured abstract in biology (2007) – 專利文獻自動分析 (2004, NTCIR) – 研究文獻中的情緒語意分析 (2010, STI) – 科教領域 4
文獻內容自動分析 - 具現況 (1/2) • Cite. Space – 陳超美 , Drexel University (2003) – http: //cluster. cis. drexel. edu/~cchen/citespace/ – 擷取科學論文中自然呈現的聚類 – 顯示 典範移轉中的轉折點 -類別間的橋樑 • VOSviewer – Nees Jan van Eck and Ludo Waltman (2007) – CWTS of Leiden University – http: //www. vosviewer. com/ 5
文獻內容自動分析 - 具現況 (2/2) • Science Mapping Software Tools: Review, Analysis, and Cooperative Study Among Tools – Cobo, et al, JASIST 2011 paper – 比較九種 具(免費、付費) • Bibexcel, Vantage. Point, Sci 2 Tool, … – 沒有一種 具可以涵蓋其他 具的所有功能 • 這類分析 (幾乎 )有標準流程 (Börner et al 2003) • CATAR released in 2010 (since 2004) 6
CATAR簡介 • Content Analysis Toolkit for Academic Research • 曾元顯 , 2004 -2015 – http: //web. ntnu. edu. tw/~samtseng/CATAR/ • CATAR技術細節: – Yuen-Hsien Tseng, Chi-Jen Lin, and Yu-I Lin, "Text Mining Techniques for Patent Analysis", Information Processing and Management, Vol. 43, No. 5, 2007, pp. 1216 -1247. – Journal clustering of Library and Information Science for subfield delineation using the bibliometric analysis toolkit: CATAR", Scientometrics, Vol. 95, No. 2, pp. 503 -528, May 2013. – 曾元顯、林瑜一(2011)。 內容探勘技術在教育評鑑研究發展趨勢 分析之應用 。教育科學研究期刊, 56(1), 129 -166。 – 曾元顯 , "文獻內容探 勘 具 - CATAR 之發展和應 用 ", 圖書館學與 資訊科學 半年刊 , 第 37 卷 第 1 期,頁 31 -49, 2011年 04月. 7
CATAR分析功能 • 概觀分析(overview) • 主題歸類分析( clustering) – 書目對分析(bibliographic coupling) – 共現字分析(co-word analysis) 8
CATAR安裝 • 安裝 Perl – http: //strawberryperl. com/, at least Version 5. 12. 0 • 下載 CATAR – http: //web. ntnu. edu. tw/~samtseng/CATAR/ – 參考上述網頁說明,安裝其他 Perl modules • 解壓縮到 C: ,以便產生C: CATAR目錄 – C: CATARsrc:程式放置處,命令由此目錄下 達 – C: CATARSource:待分析資料放置處 – C: CATARResult:執行結果資料放置處 9 – C: CATARdoc:執行過程資料放置處
待分析資料的準備 • 待分析資料的界定 (最重要的步驟、第二有價值的 部分 ) – 關鍵詞搜尋的結果 – 核心期刊的論文 – 綜合查詢的結果(期刊 +關鍵詞 +年代限制) – 一篇篇專家確認過的文獻 • Wo. K資料的搜尋 – 搜尋技巧與資料下載 • 參考: Wo. S_Record_Download. ppt 10
ISI Wo. S Publication Record Only the fields in red color are used. Cited References are used in the bibliographic coupling for topic clustering and citation tracking FN ISI Export Format VR 1. 0 PT J AU Tseng, SC Tsai, CC AF Tseng, Sheng-Chau Tsai, Chin-Chung TI On-line peer assessment and the role of the peer feedback: A study of high school computer course SO COMPUTERS & EDUCATION LA English DT Article DE interactive learning environments; secondary education; learning communities; improving classroom teaching; peer assessment ID WORLD-WIDE-WEB; ASSESSMENT SYSTEM; HIGHEREDUCATION; STUDENTS; THINKING; SCIENCE; SELF AB The purposes of this study were to explore the effects and the validity of on-line peer assessment in high schools and … C 1 Natl Chiao Tung Univ, Inst Educ, Hsinchu 300, Taiwan. Natl Chiao Tung Univ, Ctr Teacher Educ, Hsinchu 300, Taiwan. RP Tsai, CC, Natl Chiao Tung Univ, Inst Educ, 1001 Ta Hsueh Rd, Hsinchu 300, Taiwan. EM cctsai@mail. nctu. edu. tw CR ROTH WM, 1997, SCI EDUC, V 6, P 373 DOCHY F, 1999, STUD HIGH EDUC, V 24, P 331 … NR 23 TC 2 PU PERGAMON-ELSEVIER SCIENCE LTD PI OXFORD PA THE BOULEVARD, LANGFORD LANE, KIDLINGTON, OXFORD OX 5 1 GB, ENGLAND SN 0360 -1315 J 9 COMPUT EDUC JI Comput. Educ. PD DEC PY 2007 VL 49 IS 4 BP 1161 EP 1174 DI 10. 1016/j. compedu. 2006. 01. 007 PG 14 SC Computer Science, Interdisciplinary Applications; Education & Educational Research GA 218 OF UT ISI: 000250024100013 ER 11
Wo. S的重要欄位 AU:作者欄, Kainz, H; Hofstetter, H 例: TI:論文標題, Adaption of the main waste water treatment plant … 例: SO:期刊全名, WATER SCIENCE AND TECHNOLOGY。 例: DE:作者給定的關鍵詞, 例: large wastewater treatment plant; 。 ID:論文描述詞, WATER; CONTAMINATION; PLANT; 。 例: AB:論文摘要,約100 -300個英文字。 C 1:作者所屬機構之國家。 CR:參考文獻, 例: BALDI F, 1988, WATER AIR SOIL POLL, V 38, P 111 NR:參考文獻篇數,例: 3。 TC:被引用的次數,例: 1。 PY:論文出版年,例: 1996。 SC:論文所屬領域別 , Environmental Sciences; Water Resources 例: UT: Web of Science 之論文主鍵欄 , ISI: A 1996 VF 74600009 例: 12
概觀分析 • 將 資料解剖,置於資料庫內 ,以便於管理 • 從資料庫中,讀取各欄位,進行 交叉統計 • 趨勢分析 – 年代篇數序列 的 線性回歸線斜率 作為 趨勢指標 – Yuen-Hsien Tseng, Yu-I Lin, Yi-Yang Lee, Wen-Chi Hung, and Chun. Hsiang Lee, " A Comparison of Methods for Detecting Hot Topics", Scientometrics, Vol. 81, No. 1, Oct. 2009, pp. 73 -90. • 執行命令 (範例 ): – C: CATARsrc>perl –s automc. pl -OOA SE. . Source_DataSEdata 命令選項 分析結果之目錄名稱 待分析資料之路徑名稱 13
命令提示字元 (DOS)命令 • 開啟命令提示字元 – 開始 ->所有程式 ->附屬應用程式 ->命令提示 字元 • 變換到磁碟機 C: C: • 變換目錄到 CATAR: CATAR cd • 變換到上一層目錄: cd. . • 絕對路徑: C: CATARSource_DataSEdata • 相對路徑:若已經在目錄 CATARsrc下, 則為. . Source_DataSEdata 14
概觀分析範例 結果參考:C: CATARResultSE_SE_by_field. xls #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12 篇數 54 640 238 187 249 365 326 144 422 144 143 2, 912 查詢條件 SO=(Journal of the Learning Sciences) SO=(Computers & Education) SO=(Science Education) SO=(Journal of Computer Assisted Learning) SO=(Journal of Research in Science Teaching) SO=(British Journal of Educational Technology) SO=(Educational Technology & Society) SO=(ETR&D-Educational Technology Research And Development) SO=(International Journal of Science Education) SO=(Research in Science Education) SO=(Innovations in Education and Teaching International) #1 or #2 or #3 or #4 or #5 or #6 or #7 or #8 or #9 or #10 or #11 Document Type=(Article) Databases=SCI-EXPANDED, SSCI, A&HCI Timespan=2005 -2009 15
Year Production: Top 8 Countries AUSTRALIA CANADA UK TAIWAN 2004 12 3 1 6 1 0 4 0 2005 138 69 36 38 16 14 29 15 2006 139 63 31 25 18 19 18 13 2007 173 70 61 43 28 18 20 21 2008 204 72 108 44 34 31 19 18 2009 198 71 84 42 44 34 24 29 2010 6 4 7 0 0 2 total 870 352 328 198 141 116 114 98 16 TURKEY NETHER SPAIN LANDS USA
Most Productive Authors: Top 10 AU NC TC IF FC FTC 17. 6 104. 9 FIF Tsai, CC 6. 14 Roth, WM 18 61 3. 39 7. 7 25. 7 Koper, R 15 60 4. 00 3. 8 21. 4 Hwang, GJ 14 94 6. 71 3. 7 Valcke, M 13 165 12. 69 4. 3 Lee, O 12 93 7. 75 3. 2 5. 63 AU Tseng, SC Tsai, CC 27. 38 Tseng, SC : 0. 5 53. 4 12. 42 Tsai, CC : 0. 5 23. 0 7. 19 Chang, CY 11 49 4. 45 5. 2 25. 6 4. 92 Huang, YM 11 42 3. 82 3. 6 12. 8 3. 56 Sadler, TD 11 110 10. 00 4. 7 48. 6 10. 34 Chang, KE AU Tseng, SC Tsai, CC Tseng, SC : 1 Tsai, CC : 1 37 227 11 3. 3 16. 6 56 5. 09 5. 96 3. 34 5. 03 NC=Normal Count: each co-author is counted as a single author FC=Fractional Count: all the co-authors are counted as a single author IF =TC/NC, FIF=FTC/FC 17
Most Productive Institutes: Top 15 IN NC TC IF FC FTC FIF Natl Taiwan Normal Univ 61 220 3. 61 45. 6 157. 4 3. 45 Nanyang Technol Univ 52 217 4. 17 37 149. 2 4. 03 Open Univ 50 265 5. 30 41. 3 234. 4 5. 68 Natl Cent Univ 46 276 6. 00 29. 2 164. 1 5. 62 Indiana Univ 39 315 8. 08 22. 8 171. 0 7. 50 Natl Taiwan Univ Sci & Technol 35 212 6. 06 22 117. 8 5. 35 Natl Cheng Kung Univ 34 108 3. 18 27. 4 90. 3 3. 30 Middle E Tech Univ 33 87 2. 64 24. 3 70. 3 2. 89 Florida State Univ 32 145 4. 53 21. 2 75. 0 3. 54 Curtin Univ Technol 31 85 2. 74 18. 9 51. 2 2. 71 Univ Georgia 31 138 4. 45 19. 3 81. 7 4. 23 Natl Chiao Tung Univ 29 150 5. 17 18. 6 93. 8 5. 04 Univ London 29 168 5. 79 20. 9 83. 6 4. 00 Arizona State Univ 28 104 3. 71 18. 4 62. 8 3. 41 Weizmann Inst Sci 27 153 5. 67 20. 7 121. 3 5. 86 Data are from the C 1 field of each record: C 1 Natl Chiao Tung Univ, Inst Educ, Hsinchu 300, Taiwan 18
Most Cited References *NAT RES COUNC, 1996, NAT SCI ED STAND 245 LEDERMAN NG, 1992, J RES SCI TEACH, V 29, P 331 63 LAVE J, 1991, SITUATED LEARNING LE 157 *NRC, 1996, NAT SCI ED STAND 63 VYGOTSKY LS, 1978, MIND SOC DEV HIGHER 131 DRIVER R, 2000, SCI EDUC, V 84, P 287 61 BROWN JS, 1989, EDUC RES, V 18, P 32 113 DRIVER R, 1996, YOUNG PEOPLES IMAGES 59 WENGER E, 1998, COMMUNITIES PRACTICE 109 MILLAR R, 1998, 2000 SCI ED FUTURE 59 *AM ASS ADV SCI, 1993, BENCHM SCI LIT 93 LEMKE JL, 1990, TALKING SCI LANGUAGE 59 POSNER GJ, 1982, SCI EDUC, V 66, P 211 78 *NAT RES COUNC, 2000, INQ NAT SCI ED STAND 57 SHULMAN LS, 1986, EDUC RES, V 15, P 4 76 LINCOLN YS, 1985, NATURALISTIC INQUIRY 52 COHEN J, 1988, STAT POWER ANAL BEHA 70 BROWN AL, 1992, J LEARN SCI, V 2, P 141 52 SHULMAN LS, 1987, HARVARD EDUC REV, V 57, P 1 67 COLLINS A, 1989, KNOWING LEARNING INS, P 453 52 Data are from the CR field of each record: CR ROTH WM, 1997, SCI EDUC, V 6, P 373 19
Most Cited Authors AU NC Rank AU NC 1 ROTH WM 411 11 LEDERMAN NG 230 2 *NAT RES COUNC 397 12 BANDURA A 226 3 DRIVER R 395 13 VOSNIADOU S 214 4 JONASSEN DH 336 14 KUHN D 213 5 MAYER RE 323 15 TABER KS 196 6 VYGOTSKY LS 259 16 OSBORNE J 195 7 TSAI CC 250 17 BROWN AL 184 8 CHI MTH 249 18 SHULMAN LS 180 9 *AM ASS ADV SCI 246 19 AIKENHEAD GS 178 10 LAVE J 242 20 TOBIN K Rank Data are from the CR field of each record: CR ROTH WM, 1997, SCI EDUC, V 6, P 373 20 176
Most Cited Journals rank J 9 DF 1 J RES SCI TEACH 4707 11 COMPUT HUM BEHAV 622 2 SCI EDUC 3368 12 LEARN INSTR 622 3 INT J SCI EDUC 2927 13 EDUC RES 618 4 COMPUT EDUC 1668 14 COGNITION INSTRUCT 581 5 J LEARN SCI 899 15 J EDUC COMPUT RES 562 6 J EDUC PSYCHOL 877 16 EDUC PSYCHOL 523 7 ETR&D-EDUC TECH RES 829 17 STUDIES SCI ED 468 8 REV EDUC RES 825 18 RES SCI EDUC 446 9 J COMPUT ASSIST LEAR 737 19 J CHEM EDUC 443 10 BRIT J EDUC TECHNOL 717 20 INSTR SCI 433 Data are from the CR field of each record: CR ROTH WM, 1997, SCI EDUC, V 6, P 373 21
主題歸類分析 • • 索引建立 相似度計算 文件歸類 - 建立 主題樹 類別標題詞擷取 多階段歸類 - 建立 高階主題樹 多維縮放 (MDS) - 建立 主題地圖 主題與各項資料的交叉分析 22
索引建立 • 書目對分析: – 建立書目對強度矩陣 – 計數並正規化引用次數 • 共現字 (與任何文字 )分析: – 刪除停用詞(the、 for、 and, at, …) of、 on、 – 正規化詞彙(消除單複數、被動、進行式的差 異) – 擷取關鍵片語( 已專利之技術 [Tseng, 2002, JASIST]) – 建立詞彙到文件的反向索引資訊檔案 23
相似度之計算 T=2529 for 318 EEPA papers 詞彙 T 文獻 M Sim(A, B) = 詞彙 2 2 x|S(A)∩S(B)| ----------|S(A)|+|S(B)| 詞彙 1 文 件 A 文獻 2 文 件 B 文獻 1 文 件 A 共現字 相似性 D 1 D 2 M=9957 for 318 EEPA papers 文 件 B 書目對 相似性 Dn D 1 D 1 D 2 Dn Dn 24 D 2 Dn
主題樹 • 根據相似度 (距離 )矩陣,進行凝聚階層歸 類 agglomerative hierarchical clustering (AHC) – Complete link criterion 0. 0 門檻: 0. 075 結果: 6類 0. 1 – Dendrogram 主題樹 0. 2 0. 3 D 1 D 2 D 3 D 4 D 5 D 6 D 7 D 8 D 9 D 10 D 11 D 12 D 13 D 14 D 15 D 16 D 17 25
主題樹範例 (電影新聞資料 ) • 1(7): 161 : 7 Docs. : 0. 3478 (美國 : 9. 4) – 2 : 4 Docs. : 1. 0000 (美國 : 4. 1) 類別序號 與篇數 • • 13 : 101765 : 2006 -01 -01: 納尼亞傳奇 美國片 55 : 113371 : 2006 -03 -19: V怪客 美國片 48 : 109839 : 2006 -03 -12: 北國性騷擾 美國片 1 : 98663 : 2006 -01 -08: 惡狼 ID 美國片 相似度 類別標題詞 – 32 : 3 Docs. : 0. 7245 (影迷 : 7. 0, 美國 : 2. 4) • 14 : 2 Docs. : 0. 9340 (影迷 : 4. 0, 絕命終結站 : 3. 5, 絕命 : 3. 5, 飛車 : 2. 8, 雲霄飛車 : 2. 8) 類別編號 – 11 : 101543 : 2006 -01 -15: 奪魂鋸 2美國片 (下一階使用) – 27 : 104778 : 2006 -02 -26: 絕命終結站 3雲霄飛車驚魂 與篇數 • 16 : 102575 : 2006 -01 -08: 偷穿高跟鞋 美國片 • 9(3): 28 : 3 Docs. : 0. 7614 (傑克 : 10. 0, 李安 : 8. 9, 傑克基倫霍 : 7. 0, 希斯萊傑 : 3. 2) – 17 : 2 Docs. : 0. 9141 (李安 : 11. 0, 傑克 : 5. 7, 斷背山 : 4. 9, 希斯萊傑 : 4. 0, 傑克基倫霍 : 3. 2) • 3 : 98770 : 2006 -01 -22: 李安靠 斷背山重拾熱情 • 7 : 100886 : 2006 -01 -22: 斷背山 美國片 – 21 : 104156 : 2006 -02 -26: 鍋蓋頭 美國片 • 12(3): 74 : 3 Docs. : 0. 5263 (奶油 : 7. 3, 絕配 : 6. 0, 料理 : 5. 1, 凱特 : 4. 9, 尼克 : 3. 2) – 58 : 2 Docs. : 0. 6041 (番紅花 : 6. 3, 凱特 : 6. 0, 番紅花醬汁 : 4. 9, 尼克 : 4. 0, 鮮奶 : 4. 0) • 68 : 397612 : 2007 -08 -25: 料理絕配 跟著男主角做義國菜 26 • 71 : 403973 : 2007 -08 -25: 料理絕配 跟著女主角做法國菜
類別標題詞自動擷取 • 歸類後,自動擷取類別特徵詞,作為類別標題 – 結合「相關係數」及「詞頻」排序詞彙,可獲顯著成效 Yuen-Hsien Tseng, " Generic Title Labeling for Clustered Documents", Expert Systems With Applications, Vol. 37, No. 3, 15 March 2010, pp. 2247 -2254. 27
多階段歸類示意圖 每一階段 均為 獨立的 凝聚階層歸類(AHC) Topics 第二階段 Concepts 第一階段 Docs. Outliers: 低於門檻,無法歸入類別者 28
主題地圖 • MDS (Multi-Dimensional Scaling, 多維縮放 ) – 將 n個物件,總共 n(n-1)/2個相似度關係,投 影到 2或 3維空間上,以便於視覺化其關係 5. Material 2. Electronics and Semi-conductors 1. Chemistry 4. Communication and computers 3. Generality 6. Biomedicine 29 NSC美國專利文獻主題地圖
主題樹與主題地圖 Carbon Nanotube專利文獻分析 30
書目對分析 • 執行命令 (範例 ): – C: CATARsrc>perl -s automc. pl -OBC SE. . Source_DataSESE. mdb • 結果: – C: CATARResultSE_BC • • *. html:主題樹 *all*. html:主題樹以及各項資料之交叉分析 *. xls:各主題與各項資料之交叉分析 *titles*. html:每個主題包含之篇名 31
共現字分析 • 執行命令 (範例 ): – C: CATARsrc>perl -s automc. pl -OCW SE. . Source_DataSESE. mdb • 結果: – C: CATARResultSE_CW • • *. html:主題樹 *all*. html:主題樹以及各項資料之交叉分析 *. xls:各主題與各項資料之交叉分析 *titles*. html:每個主題包含之篇名 32
分析範例 -曾元顯著作 -書目對 • 1(6): 34 : 6 Docs. : 0. 020000 (cluster: 5. 1, map: 3. 0, min: 3. 0, text: 2. 1) – 12 : 4 Docs. : 0. 142857 (cluster: 7. 0, patent: 5. 2, text: 3. 7, generic: 2. 6, title: 2. 6) • • 門檻=0. 0 合理度: 100% 5 : 3 Docs. : 0. 224490 (cluster: 5. 0, generic: 3. 1, title: 3. 1, text: 2. 4, document: 2. 3) – 1 : 2 Docs. : 0. 692308 (generic: 4. 0, title: 4. 0, cluster: 3. 2, document: 3. 1, correlation coefficient: 2. 0) » 2 : ISI: 000241690200012 : 2006: Toward generic title generation for clustered documents 6 : ISI: 000272846500049 : 2010: Generic title labeling for clustered documents – 3 : ISI: 000246869800006 : 2007: Text mining techniques for patent analysis 4 : ISI: 000251991600006 : 2007: Patent surrogate extraction and evaluation in the context of patent mapping – 18 : 2 Docs. : 0. 052632 (education: 4. 0, content analysi: 2. 0, content: 2. 0, media: 2. 0) • • • 7 : ISI: 000277110400017 : 2010: Mining concept maps from news stories for measuring civic scientific literacy in media 8 : ISI: 000279714800001 : 2010: Trends of Science Education Research: An Automatic Content Analysis 2(3): 15 : 3 Docs. : 0. 095238 (neural network: 3. 1, quadratic: 2. 3, sort: 2. 3, perceptron: 1. 7) – 2 : 2 Docs. : 0. 333333 (quadratic: 3. 0, sort: 3. 0, perceptron: 2. 3, winner-take-all: 1. 4, constant-time: 1. 4) • • 13 : ISI: A 1995 QT 09700011 : 1995: ON A CONSTANT-TIME, LOW-COMPLEXITY WINNER-TAKE-ALL NEURAL-NETWORK 9 : ISI: A 1992 HU 15600007 : 1992: SOLVING SORTING AND RELATED PROBLEMS BY QUADRATIC PERCEPTRONS – 10 : ISI: A 1992 HY 58100028 : 1992: CONSTRUCTING ASSOCIATIVE MEMORIES USING HIGH-ORDER NEURAL NETWORKS • 3(2): 14 : 2 Docs. : 0. 113208 (automatic: 3. 1, chinese: 1. 4, text: 1. 4, thesauru: 1. 4) – 0 : ISI: 000167255500002 : 2001: Automatic cataloguing and searching for retrospective data by use of OCR text – 1 : ISI: 000178776600007 : 2002: Automatic thesaurus generation for Chinese documents • 4(2): 3 : 2 Docs. : 0. 285714 (code: 4. 0, decoder: 1. 4, fast: 1. 4, reed-muller: 1. 4) – 11 : ISI: A 1993 MA 58300001 : 1993: DECODING REED-MULLER CODES BY MULTILAYER PERCEPTRONS – 12 : ISI: A 1993 MA 58300002 : 1993: FAST NEURAL DECODERS FOR SOME CYCLIC CODES • • 5(1): 36 : 1 Docs. : 0 (hot: 2. 0, detect: 2. 0, comparison: 2. 0, topic: 1. 1, scientometric: 0. 7) 5 : ISI: 000270841800006 : 2009: A comparison of methods for detecting hot topics 33
分析範例 -曾元顯著作 -書目對 第二階 ) ( 門檻=0. 0 合理度: 100% • 1(2): 1 : 5 Docs. : 0. 100000 (neural: 4. 0, perceptron: 3. 0, code: 2. 4, decoder: 1. 8, network: 1. 8) 第一階的 類別編號 與篇數 – 1 : 15 : 3 Docs. : 0. 095238(neural network: 3. 1, quadratic: 2. 3, sort: 2. 3, perceptron: 1. 7) – 3 : 2 Docs. : 0. 285714(code: 4. 0, decoder: 1. 4, fast: 1. 4, reed-muller: 1. 4) • 2(2): 2 : 8 Docs. : 0. 022556 (automatic: 5. 0, document: 4. 0, text: 4. 0, generation: 3. 0, cluster: 1. 8) – 0 : 34 : 6 Docs. : 0. 020000(cluster: 5. 1, map: 3. 0, min: 3. 0, text: 2. 1) – 2 : 14 : 2 Docs. : 0. 113208(automatic: 3. 1, chinese: 1. 4, text: 1. 4, thesauru: 1. 4) • 3(1): 4 : 1 Docs. : 0 (hot: 2. 0, detect: 2. 0, comparison: 2. 0, topic: 2. 0, scientometric: 1. 0) – 4 : 36 : 1 Docs. : 0(hot: 2. 0, detect: 2. 0, comparison: 2. 0, topic: 1. 1, scientometric: 0. 7) 34
分析範例 -曾元顯著作 -書目對 第二階 ) 35 (
分析範例 -曾元顯著作 -共現字 • 1(5): 29 : 5 Docs. : 0. 0940 (term: 19. 0, document: 6. 7, algorithm: 4. 0) 合理度: 60%-80% – 7 : 3 Docs. : 0. 5403 (document: 12. 2, generic: 7. 7, cluster: 7. 6, term: 7. 4, algorithm: 6. 0) 有共同的Map或 • 2 : 2 Docs. : 0. 9610 (cluster: 10. 8, generic: 10. 0, label: 7. 0, title: 7. 0, document: 5. 6) Mapping之詞彙 – 2 : ISI: 000272846500049 : 2010: Generic title labeling for clustered documents – 6 : ISI: 000241690200012 : 2006: Toward generic title generation for clustered documents 與概念,但其他 • 7 : ISI: 000178776600007 : 2002: Automatic thesaurus generation for Chinese documents 概念則不相同 – 3 : 2 Docs. : 0. 7090 (map: 7. 7, patent: 5. 4, term: 4. 1, scientific: 4. 0, new: 4. 0) • • • 1 : ISI: 000277110400017 : 2010: Mining concept maps from news stories for measuring civic scientific literacy in media 4 : ISI: 000251991600006 : 2007: Patent surrogate extraction and evaluation in the context of patent mapping 2(3): 19 : 3 Docs. : 0. 2776 (automatic: 7. 3, text: 6. 9, analysi: 4. 9, approach: 4. 6, topic: 1. 9) – 4 : 2 Docs. : 0. 6881 (science: 7. 4, analysi: 6. 9, education: 5. 4, science education: 5. 4, research: 5. 4) • • 0 : ISI: 000279714800001 : 2010: Trends of Science Education Research: An Automatic Content Analysis 5 : ISI: 000246869800006 : 2007: Text mining techniques for patent analysis – 8 : ISI: 000167255500002 : 2001: Automatic cataloguing and searching for retrospective data by use of OCR text • 3(2): 1 : 2 Docs. : 1. 00 (network: 7. 7, memory: 4. 0, associative memory: 2. 7, winner-take-all: 2. 0) – 12 : ISI: A 1992 HY 58100028 : 1992: CONSTRUCTING ASSOCIATIVE MEMORIES USING HIGH-ORDER NEURAL NETWORKS – 9 : ISI: A 1995 QT 09700011 : 1995: ON A CONSTANT-TIME, LOW-COMPLEXITY WINNER-TAKE-ALL NEURAL-NETWORK • 4(1): 30 : 1 Docs. : 0 (trend: 6. 7, different: 5. 0, better: 3. 0, trend observation: 3. 0, choice: 3. 0) – 3 : ISI: 000270841800006 : 2009: A comparison of methods for detecting hot topics 36
主題趨勢分析 • 列出主題之年代篇數序列 • 計算其線性回歸線的斜率 • 按此斜率排序主題,可分析 – 趨勢漸增之主題 – 趨勢漸減之主題 37
Breakdown Trends of ICT in Edu. Cluster 4 Cluster 5 Cluster 6 68 : 993筆 104 : 464筆 22 : 237筆 85 : 139筆 97 : 55筆 51 : 83筆 1990 38 1 9 1 7 1 1991 53 1 8 2 6 0 1992 55 4 4 1 11 0 1993 50 3 9 1 7 1 1994 42 2 18 5 5 1 1995 50 17 23 0 6 0 47 12 19 2 5 0 1997 57 27 17 10 3 0 1998 66 29 11 5 2 0 1999 52 28 14 7 0 1 2000 69 33 15 8 0 0 2001 43 43 13 11 0 1 2002 44 44 9 10 0 2 2003 53 34 8 14 1 11 2004 56 59 7 12 0 11 2005 71 43 21 15 1 14 2006 69 37 10 10 1 27 2007 78 47 22 25 0 13 Promising topics (not yet mature) Topic with periodic attraction Cluster 3 Dying out topics Hot topics during that period Cluster 2 1996 Main stream topic Cluster 1
解讀( 1/2) • 最有價值的部分 • Access 檔案:原始資料結構化後放置處 – 可人 修改,再進行書目對與共現字之分析 • Excel 檔案:各種交叉分析結果,可製作圖表 • HTML 檔案:主題樹、篇名與交叉分析結果 • 結果檔案,在 C: CATARResult 目錄裡: – 第 n 階的 類別 ,其主題樹 在第 n 階的結果目錄 裡,其中有上述的各種 html, xls, png 等檔案 – 第 n 階的 類別 ,其主題地圖 在第 n+1 階的結果 裡,而其中的顏色,代表各個文件所屬的類別 39
解讀( 2/2) • 多用不同參數、方式探索 • 解讀出有意義的資訊為準 • 有意義的資訊,可能散落在不同參數所產生的結 果中 • 每一種參數、方式,都有可解讀之處,越多者,越 佳 • 配合領域專家、有經驗者 (科學計量專家 )一起解讀 • 參考: – 陳超美 (2010)。如何選取 Cite. Space的參數。 http: //www. sciencenet. cn/m/user_content. aspx? id=378974 40
分析案例 • 曾元顯 、林瑜一( 2011)。 內容探勘技術在教育評 鑑研究發展趨勢分析之應用 。教育科學研究期 刊,56(1), 1 -38。 • Yueh-Hsia Chang, Chun-Yen Chang, Yuen-Hsien Tseng, "Trends of Science Education Research: An Automatic Content Analysis", Journal of Science Education and Technology, Vol. 19, No. 4, 2010, pp. 315 -331. • 曾元顯 , "文獻內容探勘 具 - CATAR 之發展和應用 ", 圖 書館學與資訊科學 半年刊 , 第 37 卷 第 1 期,頁 31 -49, 2011年 04月. 41
注意事項 • 先從概觀分析開始 – 才能將 Wo. K資料剖析到資料庫中 – 其後才能進行書目對與共現字分析 • 非 Wo. S資料之分析 – 參考: • C: CATARSource_Datamovie. mdb • C: CATARSource_Dataeport. mdb (感謝本校資訊中心 汪耀華 整理資 料) – 將自己的資料,依照上述檔案範例放入資料庫中,若無 CR欄位 則直接進行共現字分析 – 若某欄位要置放多個項目,則每個項目之間以「 ; 」隔開, 如「 Chang, YH; Chang, CY; Tseng, YH」 • 將 Wo. S的 SC對應到自己定義的類別 – 編輯: C: CATARbinISI_SC 2 C. txt – 如果沒有此項需求,可以不用理會這一點 42
3347329100d3019e246b0c2f867a2f6b.ppt