Educational Data Mining Success Stories

Slides:



Advertisements
Similar presentations
Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon.
Advertisements

The Q-matrix method: A new artificial intelligence tool for data mining Dr. Tiffany Barnes Kennedy 213, PhD - North Carolina State University.
Learning decomposition WARNING. Goals Understand what learning decomposition is –And basic intuition See how it was applied to a variety of problems Think.
Knowledge Inference: Advanced BKT Week 4 Video 5.
Ai in game programming it university of copenhagen Reinforcement Learning [Outro] Marco Loog.
Ryan S.J.d. Baker Adam B. Goldstein Neil T. Heffernan Detecting the Moment of Learning.
Mining Data from Randomized Within-Subject Experiments in an Automated Reading Tutor Joseph E. Beck and Jack Mostow Project LISTEN (
Stat 112: Lecture 9 Notes Homework 3: Due next Thursday
Chapter 7 Correlational Research Gay, Mills, and Airasian
Algebra 1 R. Jenkins, M.S., M.A..
RUNNING RECORDS GUIDED READING &. © STEPS Professional Development3 THE MULTIDIMENSIONAL MODEL OF READING connecting comparing Reading Strategies synthesising.
Adolescent Literacy Peggy McCardle, Ph.D., MPH National Institute of Child Health and Human Development, NIH Archived Information.
Using Technology to Increase Engagement in Large(r) Courses Mark A. Laumakis, Ph.D. Faculty in Residence Instructional Technology Services San Diego State.
Understanding Statistics
Psych 230 Psychological Measurement and Statistics Pedro Wolf September 30, 2009.
Bivariate Distributions Overview. I. Exploring Data Describing patterns and departures from patterns (20%-30%) Exploring analysis of data makes use of.
Scientific Method A blueprint for experiment success.
1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon.
College of Science and Engineering Evaluation of the Learning and Teaching Strategy: The Way Forward? Velda McCune Centre for Teaching, Learning and Assessment.
Advantages of Using Children’s Literature provides a motivating introduction to complex curriculum topics mathematical vocabulary can be reinforced and.
Improving the Help Selection Policy in a Reading Tutor that Listens Cecily Heiner, Joseph E. Beck, Jack Mostow Project LISTEN
1 Psych 5510/6510 Chapter 10. Interactions and Polynomial Regression: Models with Products of Continuous Predictors Spring, 2009.
Curiosity-Driven Exploration with Planning Trajectories Tyler Streeter PhD Student, Human Computer Interaction Iowa State University
Stat 112 Notes 9 Today: –Multicollinearity (Chapter 4.6) –Multiple regression and causal inference.
Carnegie Mellon Mostow 12/7/2015, p. 1 The Sounds of Silence: Towards Automated Evaluation of Student Learning in a Reading Tutor that Listens Jack Mostow.
Assessment embedded in step- based tutors (SBTs) CPI 494 Feb 12, 2009 Kurt VanLehn ASU.
Carnegie Mellon How does the amount of context in which words are practiced affect fluency growth? Experimental results Jack Mostow, Jessica Nelson, Martin.
Jump to first page Inferring Sample Findings to the Population and Testing for Differences.
Back to School Night Accelerated Math 7 Thank you for coming this evening! Colleen Mitchell Accelerated Math 7 Thank you for coming this evening! Colleen.
Kenya Evidence Forum - June 14, 2016 Using Evidence to Improve Policy and Program Designs How do we interpret “evidence”? Aidan Coville, Economist, World.
Data-Driven Education
Quantitative Methods in the Behavioral Sciences PSY 302
Response to Intervention & Positive Behavioral Intervention & Support
I never dreamed about success,
Machine Learning Inductive Learning and Decision Trees
Computational Reasoning in High School Science and Math
Science 8--Nature of Science—Scientific Problem Solving
Unit 2 Exploring Data: Comparisons and Relationships
General principles in building a predictive model
Classroom Assessment Validity And Bias in Assessment.
26134 Business Statistics Week 6 Tutorial
Micro-analysis of Fluency Gains in a Reading Tutor that Listens:
Detecting Prosody Improvement in Oral Rereading
AP English Language and Composition
Mingyu Feng Neil Heffernan Joseph Beck
Big Data, Education, and Society
CHAPTER 26: Inference for Regression
Jonathan Supovitz Abigail Gray
Science vocabulary (12) 8/22/18 quiz
AP Statistics Introduction to Elementary Statistical Methods Mr. Kent
An Embedded Experiment to Evaluate the Effectiveness of Vocabulary Previews in an Automated Reading Tutor Jack Mostow, Joe Beck, Juliet Bey, Andrew Cuneo,
Causal Data Mining Richard Scheines
Interim Assessment Training NEISD Testing Services
Neil T. Heffernan, Joseph E. Beck & Kenneth R. Koedinger
Sampling Distributions
Student evaluations of teaching
Jack Mostow* and Joseph Beck Project LISTEN (
CHAPTER 9 Testing a Claim
Experimenter-defined measures in a Reading Tutor that Listens
IERI educational data mining panel
Positive Behavior Support
Searching for Graphical Causal Models of Education Data
Year 2 SATs Information meeting
EDUC 2130 Quiz #10 W. Huitt.
Intelligent Systems (AI-2) Computer Science cpsc422, Lecture 7
Standard Normal Table Area Under the Curve
AP Statistics Introduction to Elementary Statistical Methods Mr. Kent
Teaching a receptive lesson
Intelligent Systems (AI-2) Computer Science cpsc422, Lecture 7
Standard Normal Table Area Under the Curve
Presentation transcript:

Educational Data Mining Success Stories Jack Mostow Project LISTEN (www.cs.cmu.edu/~listen) “Home run”: demonstrable increase in learning “Base hit”: likely to improve learning by informing: Educational researchers Teachers Students Tutor developers Automated tutors IERI PI Meeting panel: Data Mining and Analysis Funding: National Science Foundation 1

Predicted a student’s speedup on a word 1. Informing (Reading) Researchers: Microanalysis of repeated vs. wide reading Mostow & Beck (SSSR2005) Predicted a student’s speedup on a word Reduction in word reading time From one (“practice”) encounter of a word To the next (“test”) encounter in a new context By training a linear model based on: What was the student’s reading level? How many letters long was the word? How often had the student seen the word before? Had the student seen the practice sentence before? (More predictor variables …) 2

Results based on Reading Tutor data N = 243,172 speedup opportunities for 352 gr 1-6 students. Speedup averaged 18 ms per encounter (for the first seven). Higher readers sped up less: 3 ms less per grade level. Longer words sped up more: 2 ms more per letter. A new practice sentence helped 27 ms more than an old one. Wide reading beat rereading! 3

2. Informing Teachers: What influences student outcomes in an online course? Scheines et al. (JECR 2005): TETRAD related variables logged in “Causal and Statistical Reasoning” (N = 47 students): pre: pre-test % quiz: average % on quizzes final: % on final exam print: % of modules printed Instrument to estimate effects of voluntary questions so as to infer causality from observation voluntary questions: % attempted Positive effect on performance Online-only, inhibited by printing Telling next-year students helped! (Standardized regression coefficients of variables regressed on parents) 4

3. Informing students: Proactive help to prevent likely mistakes Merceron & Yacef (AIED 2005): Induced prediction rules “if missed X, likely to miss Y” for web-based logic tutor (N = 860 students) Warn students before predicted mistakes occur Warning phrased by teacher = tutor designer 5

Aist (2001 PhD): Explain some new words; later, test all. 4. Informing tutor designers: Does explaining new vocabulary help more than just reading in context? Aist (2001 PhD): Explain some new words; later, test all. Did kids do better on explained vs. unexplained words? Overall: NO; 38%  36%, N = 3,171 trials Rare, 1-sense words tested 1-2 days later: YES! 44% >> 26%, N = 189. 6

5a. Informing tutors: Infer knowledge from behavior Corbett et al. (1995): Knowledge tracing updates at each step the probability that the student has learned the relevant rule. Helps tutor decide what to teach next Predicts error rate with r = 0.85 (0.90 after refining the rules) Average learning curve for 21 cognitive rules 7

5b. Informing tutors: learn what to do Beck et al. (AAAI 2000): learned a teaching policy for a given goal, e.g. “problems average 30 sec” Learned policy cut time per problem by 30% (p<.001) N = 58 students using middle school math tutor in classroom Simulated student (predicts effects of tutor actions) Tutorial agent Data from prior users of tutor Teaching policy Tutor action “try again” Result “correct answer, took 15 sec.” PA uses reinforcement learning “environment” is the PSM (I.e. the student) reward function is based on the teaching goal 8

Summary Educational data mining can inform: Educational researchers Wide reading apparently beat rereading (Mostow & Beck, SSSR2005) Teachers Infer causal effects of observed student choices (Scheines et al., JECR 2005) Students Warn of likely mistakes to prevent them (Merceron & Yacef, AIED 2005) Tutor developers Discover not just whether an intervention works, but when (Aist PhD 2001) Automated tutors Infer student knowledge from student behavior (Corbett et al., 1995) Learn teaching policies that improve outcomes (Beck et al., AAAI 2000) 9