Application of Automatically Constructed Concept Map of Learning to Conceptual Diagnosis of e-learning Expert Systems with Application 36 (2009) Chun-Hsiung Lee, Gwo-Guang Lee, Yungho Leu Presenter : Liew Keng Hou LOG O Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 2 What is Concept Map? A B Epistemological order of concept map 3 Types Concept Map for Learning Completely manual Semi-automatic Automatic 4 Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 5 Purpose in This Study 1. Develop the intelligent Concept Diagnostic System(ICDS) of an automatically constructed concept map of learning by the algorithm of Apriori for Concept Map 2. Teachers were provided with the constructed concept map of learners to diagnose the learning barriers and misconception of learners. 3. Remedial-Instruction Path(RIP) was constructed through the analyst of the concepts and weight in the concept map to offer remedial learning. 4. Statistical methods were used to analyze whether the learning performance of learners can be significantly enhanced after they have been guided by the RIP. 6 Flow Chart of Concept Diagnosis 7 Remedial-Instruction Path A Remedial-Instruction Path B C D E Relationships of the epistemological order 8 Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 9 Presetting Conceptual Weight Question Concept C1 C2 C3 C4 C5 Q1 1 0 0 0 0 Q2 0 1 0 0 0 Q3 0.5 0 0.5 0 0 Q4 0.3 0.4 0 0.3 0 Q5 0 0 0 0 1 ‘0’: not relevant ‘1’: strongly relevant 10 Recording Test Portfolio of Testees Question Testees C1 C2 C3 C4 C5 Total Q1 1 1 1 0 0 3 Q2 1 1 1 1 0 4 Q3 1 1 1 1 1 5 Q4 0 0 1 1 1 3 Q5 0 0 0 1 1 2 ‘0’: Student answered correctly the test item ‘1’: Student failed to answer correctly the test item 11 Find Out All Large Item sets Comparison Chart Itemset S1 Q1,Q2,Q3 Q1 3 S2 Q1,Q2,Q3 Q2 4 S3 Q1,Q2,Q3,Q4 Q3 5 S4 Q2, Q3, Q4, Q5 Q4 3 S5 Q3, Q4, Q5 Q5 2 12 No. of supports Find Out All Large Item sets(Cont.) Using association rules of data mining Sets Min support(MS) = 0.4 (Depends on teacher) Number of testees = 5 Questions with wrong answers given by testees has to be ≥ MS x N (0.4 x 5 = 2) 13 Find Out All Large Item sets(Cont.) Itemset Q1, Q2 Q1, Q3 Q1, Q4 Q2, Q3 Q2, Q4 Q3, Q4 Q4, Q5 No. of support Itemset 3 3 1 MS ≥ 0.4 0.4 x 5 ≥ 2 4 2 3 2 14 No. of support Q1, Q2 3 Q1, Q3 3 Q2, Q3 4 Q2, Q4 2 Q3, Q4 3 Q4, Q5 2 Ruling the Test Question Association The confidence level of the test question association rule Q𝑖 → Q𝑗 is the concept of conditional probability. It implies that a testee gives a wrong answer to Question Q𝑖 , there is a probability for the testee to give a wrong answer to Question Q𝑗, too The estimated confidence level formula is 15 Using Association Rules of Data Mining Question Testees C1 C2 C3 C4 C5 Total Q1 1 1 1 0 0 3 Q2 1 1 1 1 0 4 Q3 3 1Confidence(Q1 1 1 13 = 100% 5 Confidence(Q1 →1Q2)=P(Q2|Q1)= → Q2)=P(Q1|Q2)= = 75% 3 Q4 0 Q5 0 0 1 → Q2)=P(Q3|Q2)= 1 1 Confidence(Q3 0 0 1 1 Note: ‘0’: Student answered correctly the test item ‘1’: Student failed to answer correctly the test item 16 4 4 5 3 = 80% 2 Using Association Rules of Data Mining Let the minimum confidence(MC) level be below 70% Rule 1. Confidence (Q1 → Q2) = 100% Rule 2. Confidence (Q1 → Q3) = 100% Rule 3. Confidence (Q2 → Q1) = 75% Rule 4. Confidence (Q2 → Q3) = 100% Rule 5. Confidence (Q3 → Q2) = 80% Rule 6. Confidence (Q4 → Q3) = 100% Rule 7. Confidence (Q5 → Q4) = 100% 17 Relationship Between Concept and Concept Conversion from “test question association rules” to the effect of “relation between concept and concept” Q𝑖: 𝑖th test question C𝑥: 𝑥th concept RQ𝑖C𝑥 : relavance degree between Q𝑖 and C𝑥 WC𝑥C𝑦: relevance degree between C𝑥 and X𝑦 18 Relationship Between Concept and Concept Question Concept Question 1) 2) 3) 4) 5) Concept C1 C2 C3 C4 C1 C2 C3 C4 Q1 → Q2 → 𝑊𝐶1𝐶2 = C1 →C2 = Confidence 1 = 1 10 1 = 1 0 0 (Q1 → Q2) Q1RQ1C1 RQ2C2 Q1 1 0 0 0 1 0 0 Q1 → Q3Q2 → Q2 𝑊𝐶1𝐶3 = 0 C1 0→C3 =1Confidence 0 0 (Q1 → Q3) Q3RQ1C1 RQ3C3 0.5 = 1 10 0.5 = 0.5 0.5 0 Q3 0.5 0 0.5 0 Q2 → Q1 → Q4 𝑊𝐶2𝐶1 = C20.3 →C1 = Confidence 0.4 0 0.3 Q4 0.3 0.4 0 0.3 Question Concept (Q2 → Q1) RQ2C2 RQ1C1 = 0.75 1 1 = 0.75 Q5 Rule 2. 0 Confidence 0 0 Q1) = 100% 0 (Q3 → Q5 0 C1 0 C2 0 C3 0 C4 C5 0 0 0 0 𝑊𝐶3𝐶2 = C3 →C2 = Confidence 0.4 0 (Q3 → Q2) Q4 RQ3C3 R0.3 Q2C2 = 0.8 0.5 1 = 0.4 Q5 0 0 0 19 0 0 0 0 1 1 C5 Q2 → Q3 → 𝑊𝐶2𝐶1 = C2 →C1 = Confidence (Q2 → Q3) Q1 RQ2C2 RQ3C1 =10 1 0.5 = 0.5 1 0 0 Question Concept (0.5 < 0.75(3)) Q2 0 →C3 1 0 0 100% 𝑊𝐶2𝐶3 Rule = C2 = Confidence Confidence (Q2 = C14. C2 C3 → Q3) C4 (Q2 → Q3) Q3 RQ2C2 R0.5 1 0.5 = 0.5 0.5 Q3C3 = 1 0 0 Q1 1 0 0 Q3 → Q2 →Q4 𝑊𝐶1𝐶2 = C1 0.3→C2 = Confidence 0.4 0 0 (Q3 → Q2) Q2 RQ3C1 RQ2C2 = 0.8 1 0.5 1 =0 0.4 Q5 0 0 0 (0.4 < 1(1)) Q3 0.5 0 0.5 C5 0 0 0.3 0 0 0 0 C5 0 0 0 0 1 0 0.3 0 0 1 Relationship Between Concept and Concept 6) Q4 → Q3 → 𝑊𝐶2𝐶3 = C2 →C3 = Confidence (Q4 → Q3) RQ4C2 RQ3C3 = 1 0.4 0.5 = 0.2 (0.2 < 0.5(4)) 𝑊𝐶4𝐶3 = C4 →C3 = Confidence (Q4 → Q3) RQ4C4 RQ3C3 = 1 0.3 0.5 = 0.15 7) Q5 → Q4 → 𝑊𝐶5𝐶1 = C5 →C1 = Confidence (Q5 → Q4) RQ5C5 RQ4C1 = 1 1 0.3 = 0.3 𝑊𝐶5𝐶2 = C5 →C2 = Confidence (Q5 → Q4) RQ5C5 RQ4C2 = 1 1 0.4 = 0.4 𝑊𝐶5𝐶4 = C5 →C1 = Confidence (Q5 → Q4) RQ5C5 RQ4C4 = 1 1 0.3 = 0.3 20 Preliminary Concept Maps(Stage 1) C1 0.3 1 0.75 0.4 C2 C5 0.5 0.5 0.4 C3 0.3 0.2 C4 21 Preliminary Concept Maps (Cont.) 1) Q1 → Q2 → 𝑾𝑪𝟏𝑪𝟐 = C1 →C2 = Confidence (Q1 → Q2) RQ1C1 RQ2C2 =111=1 2) Q1 → Q3 → 𝑾𝑪𝟏𝑪𝟑 = C1 →C3 = Confidence (Q1 → Q3) RQ1C1 RQ3C3 = 1 1 0.5 = 0.5 3) Q2 → Q1 → 𝑾𝑪𝟐𝑪𝟏 = C2 →C1 = Confidence (Q2 → Q1) RQ2C2 RQ1C1 = 0.75 1 1 = 0.75 4) Q2 → Q3 → 𝑾𝑪𝟐𝑪𝟏 = C2 →C1 = Confidence (Q2 → Q3) RQ2C2 RQ3C1 = 1 1 0.5 = 0.5 (0.5 < 0.75(3)) Q2 → Q3 → 𝑾𝑪𝟐𝑪𝟑 = C2 →C3 = Confidence(Q2 → Q3) RQ2C2 RQ3C3 = 1 1 0.5 = 0.5 5) Q3 → Q2 → 𝑾𝑪𝟏𝑪𝟐 = C1 →C2 = Confidence (Q3 → Q2) RQ3C1 RQ2C2 = 0.8 0.5 1 = 0.4 (0.4 < 1(1)) Q3 → Q2 → 𝑾𝑪𝟑𝑪𝟐 = C3 →C2 = Confidence (Q3 → Q2) RQ3C3 RQ2C2 = 0.8 0.5 1 = 0.4 6) Q4 → Q3 → 𝑾𝑪𝟐𝑪𝟑 = C2 →C3 = Confidence (Q4 → Q3) RQ4C2 RQ3C3 = 1 0.4 0.5 = 0.2 (0.2 < 0.5(4)) Q4 → Q3 → 𝑾𝑪𝟒𝑪𝟑 = C4 →C3 = Confidence (Q4 → Q3) RQ4C4 RQ3C3 = 1 0.3 0.5 = 0.15 7) Q5 → Q4 → 𝑾𝑪𝟓𝑪𝟏 = C5 →C1 = Confidence (Q5 → Q4) RQ5C5 RQ4C1 = 1 1 0.3 = 0.3 Q5 → Q4 → 𝑾𝑪𝟓𝑪𝟐 = C5 →C2 = Confidence (Q5 → Q4) RQ5C5 RQ4C2 = 1 1 0.4 = 0.4 Q5 → Q4 → 𝑾𝑪𝟓𝑪𝟒 = C5 →C1 = Confidence (Q5 → Q4) RQ5C5 RQ4C4 = 1 1 0.3 = 0.3 22 Preliminary Concept Maps(Cont.) C1 0.3 1 0.75 0.4 C2 C5 0.5 0.5 0.4 C3 0.3 0.2 C4 23 Adjusting Concept Map of Learning(Stage 2) Child Concept Parent concept NP Parent C𝑖 Child NC C1 C2 C3 C4 C5 C1 ― 0 0 0 0.3 1 C2 1 ― 0 0 0.4 2 C3 0.5 0.5 ― 0.2 0 3 C4 0 0 0 ― 0.3 1 C5 0 0 0 0 ― 0 2 1 0 1 3 NP: Number of father concepts contained in the son concept NC: Number of son concepts contained in the father concept 24 Complete Concept Map C5 WC5C1 = 0.3 C1 WC1C2 = 1 WC1C3 = 0.5 WC5C4 = 0.3 WC5C2 = 0.4 C4 C2 WC2C3 = 0.5 WC4C3 = 0.2 C3 25 Determination of Learning barrier Calculate the ratio of wrong answers given in the test portfolio: ER(C𝑗) = 𝑘 𝑒𝑘𝑗 𝑖 𝑒𝑖𝑗 𝑒𝑘𝑗 : weight of the 𝑗th concept of the 𝑘th test question which was wrongly answer 𝑒𝑖𝑗 : weight of the 𝑗th concept in the whole test paper 26 Table of Ratio of Wrong Answer (Failratio) Question Concept C1 C2 C3 C4 C5 Q1 1 0 0 0 0 Q2 0 1 0 0 0 Q3 0.5 Q4 0.3 0.4 Q5 0 0 0 ER(C1) (0.5+0.3) 0.5 0 = 0 1.8 0 0.3 (0+0.4)0 ER(C2) = = 0.44 1.4 1 0 0 = 0.29 𝑘 𝑖 𝑒𝑘𝑗 0.8 0.4 0.5 0.3 0 𝑒𝑖𝑗 1.8 1.4 0.5 0.3 1 0.44 0.29 1 1 0 ER(C𝑗) 27 Algorithm of Remedial-Instruction Path 010 020 030 040 050 060 070 080 090 100 110 120 130 140 150 160 170 180 190 Void main () Call Find_Remedial-Instruction_Path(k, Fault-Concept) End //Cj denotes the FaulConcept, and k denoted the index of a father concept on Cj Sub Find_Remedial-Instruction_Path(k,Cj) //judge whether the failratio of Concept Cj is greater than the tolerance for the ratio of the giving wrong answers. If ER(Cj) failratio then Push Cj W = Max{WCiCjj1 5 i 5 n} While (Cihi RootConcept)do //Not Find to Root-Concept push Ci base on W Wend While Stack is not empty //Find to Root-Concept //RIP: Remedial-Instruction_Path RIP = Find_Remedial-Instruction_Path(i,Pop()) Wend End if End Sub 28 Intelligent concept diagnostic system(ICDS) 29 Intelligent concept diagnostic system(ICDS) 30 Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 31 Design of Experiment and Data Analysis Target of Study 245 Grade 1 students of a senior high school Pre-test of “Visual Basic Programming Language” Table of discrimination index of Questions Discrimination Question Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 Q10 PH 0.93 0.93 0.93 0.93 0.93 0.71 0.86 0.64 0.14 0.86 PL 0.71 0.57 0.36 0.21 0.21 0.14 0.21 0.21 0 0.07 Discrimination Index 0.21 0.36 0.57 0.71 0.71 0.57 0.64 0.43 0.14 0.79 <0.2 32 Flow Chart of Data Analysis 33 Cluster In order to understand the difference of concept maps produced from the test portfolio of students at different standards Optimal ratio is 27% for the high-score and low-score clusters 34 Sub-Cluster 1. Experimental group: The RIP in concept map served as the learning guide 2. Control group: Traditional non-guided network learning way was adopted Group Cluster Cluster 1 (High-score cluster) Cluster 2 (Mediumscore cluster) Cluster 3 (Low-score cluster) Experimental group 33 56 33 Control group 33 51 33 Number of students 66 113 66 35 Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 36 𝑇- test & Analysis : significant standard H0 : There is a significant difference between the mean of experimental group and the mean of control group If P-value < , then H0 is rejected. If P-value ≥ , then H0 is not rejected 37 𝑇-test of Independent Samples of Experimental Group and Control Group of Three Cluster Cluster and group High-score cluster Mediumscore cluster Low-score cluster Item Mean Standard deviation Experimental group 72.57 15.83 Control group 67.86 12.97 Experimental group 52.25 11.52 Control group 40.13 10.23 Experimental group 37.29 9.59 Control group 19.29 8.62 *𝑝 < 0.1 *𝑝 < 0.01 38 𝑡-value Significance .610 .554 2409 .030* 3.695 .003** Outline Introduction Purpose in This Study Research Approach Experiment and Data Analysis 𝑇- test & Analysis Conclusion and Discussion 39 Conclusion and Discussion 1. Discrimination index of test questions If the test question is too simple or difficult? 2. Attribute of test questions Which type of test question? 3. Learning performance Which cluster(s) has better performance? 40 Thank You for Your Participating LOG O