1 / 72

數位學習研究的資訊議題與趨勢

數位學習研究的資訊議題與趨勢. 國立臺南大學 黃國禎 數位學習科技系 教授 資訊教育研究所 所長 理工學院 院長 2006/1/7. 數位學習的優點. 突破時空限制 提供個別化教學指導 提高學習興趣 促成同學互動 建立完整的教學管理機制. 數位學習的發展現況. 行政院在九十年 NICI( 國家資訊通信發展方案 ) 計畫中，將數位學習納入「網路社會化」的一環。 國科會通過「數位學習國家型科技計畫」的構想，預計五年內投入四十億元進行跨部會計畫 「全民數位學習」 「縮減數位落差」 「行動學習載具與輔具─多功能電子書包」 「數位學習網路科學園區」

cameo
Download Presentation

數位學習研究的資訊議題與趨勢

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. 數位學習研究的資訊議題與趨勢 國立臺南大學 黃國禎 數位學習科技系 教授 資訊教育研究所 所長 理工學院 院長 2006/1/7 數位學習研究趨勢

  2. 數位學習的優點 • 突破時空限制 • 提供個別化教學指導 • 提高學習興趣 • 促成同學互動 • 建立完整的教學管理機制 數位學習研究趨勢

  3. 數位學習的發展現況 • 行政院在九十年NICI(國家資訊通信發展方案)計畫中，將數位學習納入「網路社會化」的一環。 • 國科會通過「數位學習國家型科技計畫」的構想，預計五年內投入四十億元進行跨部會計畫 • 「全民數位學習」 • 「縮減數位落差」 • 「行動學習載具與輔具─多功能電子書包」 • 「數位學習網路科學園區」 • 「前瞻數位學習技術研發」 • 「數位學習之學習與認知基礎研究」 • 「政策引導與人才培育」 數位學習研究趨勢

  4. 數位學習研究的三大趨勢 • 導入資訊技術(人工智慧、演算法、資料探勘、物件導向、代理人)提昇系統平台功能的研究 • 對數位學習標準工具及延伸的探討 • 對行動學習(M-Learning)與普化學習(U-Learning)平台與教學策略的探討 數位學習研究趨勢

  5. PART A – 資訊技術扮演的角色 • 將在數位學習的研究由應用層次(Alessi, S.M. & Trollip, S.R. 1991)提昇到功能層次 • 應用層次 • 利用網路進行教材及試題的共享 • 利用電腦作為呈現教材及考試的媒介 • 功能層次 • 被動(Chou, C. 1996)：記錄學習過程，統計、保存、報告學習評量之結果 • 主動(Kumar, D.D., Helgeson, S.L. & White, A.L. 1994)：分析學習狀態、偵查學習迷思、引導學習、實施有規劃的教學、測驗與評量 數位學習研究趨勢

  6. 資訊技術應用1： 個人化教學系統 • 個人化學習路徑之規劃 • 線上學習過程之記錄 • 線上學習行為之分析 數位學習研究趨勢

  7. 個人化學習路徑之規劃 數位學習研究趨勢

  8. 線上學習行為分析 藉由模糊專家系統來推論出此學習者的學習狀態，並給予適當的幫助 • 學習效率(Efficiency of Learning) • 學習意願(Willingness) • 耐心度(Patience) • 專心度(Concentration) • 閒置(Idleness) • 理解度(Comprehension) • 聊天(Chat) 數位學習研究趨勢

  9. (1) 學習意願分析 • 學生用心學習的意願 • 分析依據：有效登入時間/登入時間 模糊推理法則 If willingness is low Then insert INT(T×0.5) corresponding willingness frames. If willingness is average Then insert INT(T×0.25) corresponding willingness frames If willingness is high Then keep the current status. 數位學習研究趨勢

  10. (2) 耐心度分析 • 學生瀏覽一個畫面的持續度 • 分析依據：畫面學習時間/預估學習時間 模糊推理法則 If patience is low Then record this status and warn the student. If patience is average Then keep the current status. If patience is high Then keep the current status. 數位學習研究趨勢

  11. 學生集中精神於瀏覽教材的程度 分析依據：回應時間 (3) 專心度分析 模糊推理法則 If concentration is low Then insert a corresponding concentration frame. If concentration is high Then keep the current status. If concentration is average Then keep the current status. 數位學習研究趨勢

  12. (4) 聊天狀態分析 • 學生利用線上討論區來閒聊而不是討論課程 • 分析依據：學習相關比率 模糊推理法則 If chat is high Then record this status and warn the student. If chat is average Then keep the current status. If chat is low Then keep the current status 數位學習研究趨勢

  13. Gwo-Jen Hwang (1998), “A tutoring strategy supporting system for distance learning on computer networks”, IEEE Transactions on Education, Vol. 41, No. 4, pp. 343. (SCI & EI) 數位學習研究趨勢

  14. 資訊技術應用2 ： 多專家教學知識擷取與整合系統 數位學習研究趨勢

  15. Li K Fr F Cl I 模糊專家系統知識擷取 Step 1: Elicit all of the elements (concepts to be learned) from the domain expert. 數位學習研究趨勢

  16. Li K Fr F Cl I boiling point LOW;MIDDLE;HIGH atom radius NARROW;NORMAL;WIDE metalloid WEAK;NORMAL;STRONG negative charge WEAK;MIDDLE;STRONG Step 2: Elicit attributes ( properties or fuzzy variables). 數位學習研究趨勢

  17. Li K Fr F Cl I boiling point -1/N 0/N 1/N 1/S 2/S 3/S LOW;MIDDLE;HIGH atom radius -2/S -1/S 1/N 1/S 2/S 3/S NARROW;NORMAL;WIDE metalloid 1/S 2/S 3/S -3/S -3/S -3/S WEAK;NORMAL;STRONG negative charge -3/S -3/S -3/S 3/S 2/S 1/S WEAK;MIDDLE;STRONG Step 3: Fill all of the [concept, attribute] entries of the grid. A 7-scale (-3 to +3) rating and the degree of certainty(“S”,”N”). Consider the ratings of fuzzy variable ‘boiling point’: 3 means VERY HIGH, 2 means HIGH, 1 means MORE OR LESS HIGH, 0 means MIDDLE, -1 means MORE OR LESS LOW, -2 means LOW, -3 means VERY LOW ‘S’ means ‘VERY SURE’, ‘N’ means ‘NOT VERY SURE’ 數位學習研究趨勢

  18. Step 4: the first column of the above fuzzy table is translated to the following rule: IF boiling point is MORE OR LESS LOW, and atom radius is NARROW, and metalloid is MORE OR LESS STRONG, and negative chargeis VERY WEAK THEN the element could be Li TRUTH = 0.8 TRUTH = 數位學習研究趨勢

  19. 多專家知識擷取與整合的問題 While elicit expertise from multiple experts, some problems may arise: ? same element or attribute different vocabularies ?? Unified vocabularies Different ratings ??? Conflict happen Insist their ratings 數位學習研究趨勢

  20. A. Interactive knowledge elicitation unit 數位學習研究趨勢

  21. (1) Interactive user interface • NEGOTIATOR:Please give a set of elements(concepts to be learned or decisions to be made). • EDUCATOR: Li, K, Fr, F, Cl, I. . . . • NEGOTIATOR: Select a set of fuzzy values for fuzzy variable “boiling point”: 1. LOW/MIDDLE/HIGH 2. SHORT/MIDDLE/TALL 3. LIGHT/NORMAL/HEAVY 4. SMALL/MIDDLE/BIG 0. Other (user-defined) • EDUCATOR: 1 • NEGOTIATOR: Select a set of fuzzy values for fuzzy variable “atom radius”: 1. LOW/MIDDLE/HIGH 2. SHORT/MIDDLE/TALL 3. LIGHT/NORMAL/HEAVY 4. SMALL/MIDDLE/BIG 0. Other (user-defined) • EDUCATOR: 0 • NEGOTIATOR:Indicate the lower bound of the fuzzy values. • EDUCATOR: NARROW • NEGOTIATOR:Indicate the middle of the fuzzy values. • EDUCATOR: NORMAL • NEGOTIATOR:Indicates the upper bound of the fuzzy values. • EDUCATOR: WIDE 數位學習研究趨勢

  22. (2) Fuzzy table editor 數位學習研究趨勢

  23. (3) Membership function builder 數位學習研究趨勢

  24. (4) Knowledge analysis unit • Check if conflict occurs andintegrate tutoring strategies. • The contents of a fuzzy table is represented asFuzzy_value(Educator_ID, Object_name, Fuzzy_variable) and Certainty_Degree (Educator_ID, Object_name, Fuzzy_variable) for examples, the fuzzy table below can represented as Fuzzy_value(Educator1, Li, boiling point) = -1 Certainty_Degree(Educator1, Li, boiling point) = “N” ... Li K Fr F Cl I boiling point -1/N 0/N 1/N 1/S 2/S 3/S LOW;MIDDLE;HIGH atom radius -2/S -1/S 1/N 1/S 2/S 3/S NARROW;NORMAL;WIDE metalloid 1/S 2/S 3/S -3/S -3/S -3/S WEAK;NORMAL;STRONG negative charge -3/S -3/S -3/S 3/S 2/S 1/S WEAK;MIDDLE;STRONG 數位學習研究趨勢

  25. Knowledge analysis rule: Rule_analysis_02 IF (1) Current_Phase is Knowledge_Analysis and (2) Fuzzy_value(Expi, Gk, Vs)Fuzzy_value(Expj, Gk, Vs) < 0 and (3) Certainty_Degree (Expi, Gk, Vs) is "S" and (4) Certainty_Degree(Expj, Gk, Vs) is ”N” and THEN (a) Set Suggested_Fuzzy_Value be Fuzzy_value(Expi, Gk, Vs) and (b) Set Suggested_Certainty_Degree be ”N" and (c)Set Current_Phase be Knowledge_Negotiation 數位學習研究趨勢

  26. Knowledge analysis rule: Rule_analysis_04 IF (1) Current_Phase is Knowledge_Analysis and (2) Fuzzy_value(Expi, Gk, Vs)Fuzzy_value(Expj, Gk, Vs) 0 and (3) Certainty_Degree (Expi, Gk, Vs) is "S" and (4) Certainty_Degree(Expj, Gk, Vs) is "S” and (5) Fuzzy_value(Expi, Gk, Vs)  Fuzzy_value(Expj, Gk, Vs) 0 THEN (a) Set Suggested_Fuzzy_Value be Fuzzy_value(Expi, Gk, Vs) and (b) Set Suggested_Certainty_Degree be "S" and (c)Set Current_Phase be Knowledge_Negotiation 數位學習研究趨勢

  27. Knowledge analysis rule: Rule_analysis_03 IF (1) Current_Phase is Knowledge_Analysis and (2) Fuzzy_value(Expi, Gk, Vs)Fuzzy_value(Expj, Gk, Vs)< 0 and (3) Certainty_Degree (Expi, Gk, Vs) is "S" and (4) Certainty_Degree(Expj, Gk, Vs) is "S” and THEN (a) Set Suggested_Fuzzy_Value be “Conflict” and (b)Set Current_Phase be Knowledge_Negotiation 數位學習研究趨勢

  28. B. Tutoring Strategy Negotiation unit • Present suggestions by knowledge analysis unit • When a conflict occurs, experts are asked to give suggestions. “ over-general” happen An example invoke Object_Specialization procedure Bear American gray bearbear of North Pole 數位學習研究趨勢

  29. Perform knowledge integration procedure 數位學習研究趨勢

  30. Check conflict values and decide if Object_Specialization procedure should be invoked • Generate fuzzy rules. 數位學習研究趨勢

  31. C. Knowledge base generator(export to CLIPS) (deffacts initial-state (is boiling-point MORE-OR-LESS LOW) (is atom-radius NARROW) (is metalloid MORE-OR-LESS STRONG) (is negative-charge VERY WEAK)) (defrule Rule1 ?x1 <- (is ?X1 MORE-OR-LESS LOW) ?x2 <- (is ?X2 NARROW) ?x3 <- (is ?X3 MORE-OR-LESS STRONG) ?x4 <- (is ?X4 VERY WEAK) => (retract ?x1 ?x2 ?x3 ?x4) (assert (is Li -1-21-3)) (assert (CF 0.8)) (printout t ”Li is -1-21-3 with CF=0.8" crlf)) 數位學習研究趨勢

  32. Experiment (1): Time for knowledge elicitation and integration six topics of “Introduction to Computer Science” 數位學習研究趨勢

  33. Experiment (2): Frequency for correct inferences 250 test cases Gwo-Jen Hwang (2002), “On the Development of a Cooperative Tutoring Environment on Computer Networks”, IEEE Transactions on System, Man and Cybernetic Part C, Vol. 32, No. 3, 2002, pp. 272-278. (SCI, EI) 數位學習研究趨勢

  34. 資訊技術應用3： 多目標最佳化配題機制 • 從大量試題中，選取符合出題條件（題數、測試時間、概念最低配題比重．．．等）且鑑別度最大的試卷 • 指定測驗時間範圍的試題配置問題模型(Dedicated Range of Assessment Time Problem-DRAT) • 符合期望測驗時間最高界限和最低界限的多目標配題機制。 • 固定題數的試題配置問題模型(Fixed Number of Test Items Problem – FNTI) • 符合固定試卷試題數量的多目標配題機制。 數位學習研究趨勢

  35. Fuzzy Art and Dynamic Programming (2) Dynamic Programming: Find optimal test item composition (1) Fuzzy Art: Classify test items into groups Gwo-Jen Hwang (2003), “A Test Sheet Generating Algorithm for Multiple Assessment Requirements”, IEEE Transactions on Education, Vol. 46, No. 3, pp. 329-337. (SCI and EI) 數位學習研究趨勢

  36. Heuristic Algorithms • FTF (Feasible Time First) Algorithm • Find a solution to meet range of assessment time • Replace test items to find feasible solutions and to maximize average discrimination degree • FNTF (Feasible Number of Test Item First) Algorithm • Find a solution to meet the number of test items • Replace test items to find feasible solutions and to maximize average discrimination degree Gwo-Jen Hwang, Tsung-Liang Lin, Bertrand M.T. Lin (2006), “An Effective Approach for Test-Sheet Composition from Large-Scale Item Banks”, accepted by Computers & Education, Vol. 46, No. 2, pp. 122-139. (SSCI, SCI, EI) 數位學習研究趨勢

  37. Genetic Algorithms • 源自於John Holland在1975 年出版的著作Adaptation in Nature and Artificial Systems • 仿效自然界生物進化過程 • 透過基因的選擇(selection)交換(crossover)及突變(mutation)產生更好的下一代 • 選擇(selection)過程 • 較高合適值(fitness value)就有較大機會獲得保留 • 較低合適值的解答，可能會遭到淘汰 • 較不易陷入local optimal 數位學習研究趨勢

  38. Genetic Algorithm • Population (族體): • Encoding (編碼): • Crossover (交配): • Mutation (突變): • Selection (適者生存): • Fitness Function (適合度公式): 數位學習研究趨勢

  39. Crossover randomly selects one-cut-point and exchanges the right parts of two parents to generate offspring. 基因演算法交配運算 Mutation alters one or more genes with a probability equal to the mutation rate. 基因演算法突變運算 基因演算法流程圖 數位學習研究趨勢

  40. 指定測驗時間範圍的試題配置問題 (Dedicated Range of Assessment Time) x1 x2 x3 x4 x98 x99 x100 … 0 1 1 0 0 1 0 • DRAT目標函式： Maximize Z = • DRAT限制式： 第i題告第j個概念的關係 Xi = 0 or 1, i = 1, 2, …, nXi代表第i題是否被選擇 指定鑑別度最大化 指定概念的最小出題比重 指定測驗時間的下限 指定測驗時間的上限 數位學習研究趨勢

  41. DRAT的試題配置基因演算法 (1/4) • 概念程度下限先決基因演算法(Concept Lower-bound First Genetic approach – CLFG) • CLFG建立的母體(Encoding) • X 為染色體，包含有 n 個基因 • X = [x1 , x2 , …, xn]X = [0, 0, 1, …, 0] • 第i個試題被選取時，xi為1；否則，為0； 數位學習研究趨勢

  42. DRAT的試題配置基因演算法 (2/4) • 適配等級(Fitness ranking) • 適配函數　v(Sk) = • R = • Kj = 0 if , Kj = 1 else •  = w dtl ipt_l • w = ( indi xi) / average(u, l) • dtl = • = w dtu ipt_u • dtu = 數位學習研究趨勢

  43. DRAT的試題配置基因演算法 (3/4) Cut point • 交配(Crossover) A[1110011001] A’[1110011011] B[0100100011] B’[0100100001] Procedure: crossover Begin k = 0 while (k ≤ c / 2) do flag = 0 while flag = 0 do Generate random numbers R1 and R2 from discrete interval [1,K]. If R1 ≠ R2 then flag=1 end while crossover function(R1,R2) end while End 數位學習研究趨勢

  44. DRAT的試題配置基因演算法 (4/4) • 突變(Mutation) A[1110011001] A’[1110011011] P = ( 1 / n ) Procedure: mutation Begin for(i=1, i ≤ nk, i++){ Generate random number yi from discrete interval [0, 1]. Mutation function(P, yi) } End 重覆2~5步驟，直到連續10代解無進步或已產生了1500代 數位學習研究趨勢

  45. 固定題數的試題配置問題 (Fixed Number of Test Items) x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 12 18 9 45 82 6 2 34 65 71 • FNTI目標函式： Maximize Z = • FNTI限制式： xi ≥ 1 Xi 代表題庫中的題號，最小題號為1 xi ≤ n 最大題號為n 1 ≤ i ≤ q_num– 1 共選出 q_num 題 指定鑑別度最大化 指定概念的最小出題比重 數位學習研究趨勢

  46. FNTI的試題配置基因演算法(1/3) • 試題數目先決基因演算法(Feasible Item First Genetic approach – FIFG) • FIFG的進行步驟 • 建立母體 • X 為染色體，包含有 q_num 個基因 • X = [x1 , x2 , …, xq_num] X = [25, 118, …., 803] • 基因值代表著一題試題的編號 • xi ≠ xj，且 i ≠ j和 1 ≤ i, j ≤ q_num 數位學習研究趨勢

  47. FNTI的試題配置基因演算法(2/3) Cut point • 交配(Crossover) A[12,15, 96,112,193,243]A’[12,15,96,185,256,356] B[3,56,108,185,256,356]B’[3,56,108,112,193,243] • 有兩相同基因值時，隨機更換其中一值，直到沒有相同基因值為止 • 試卷中不可有二題相同的試題 數位學習研究趨勢

  48. FNTI的試題配置基因演算法(3/3) • 突變(Mutation) A[3,8,56,66,256,515] A’[3,8,56,66,346,515] P = ( 1 / n ) Procedure: mutation Begin for (m = 1, m ≤ q_numk, m++){ Generate random number rm from discrete interval [0, 1] Generate random number RC from discrete interval [1, n] mutation function(P, rm, RC) } End 重覆2~5步驟，直到連續10代解無進步或已產生了1500代 數位學習研究趨勢

  49. 實驗題庫樣本資料 • 每一個情況進行二十次實驗處理後，採用平均求解時間和平均鑑別度建立 實驗樣本 數位學習研究趨勢

  50. CLFG實驗結果及分析 (1/3) l = 30 數位學習研究趨勢

More Related