Rachel Hogue`s presentation on Big Data in Education

Big Data
in Education
Rachel Hogue
 Big Data and Education Communities
 Why Collect Educational Data?
 Learning Theories
 eLearning
 What Data Can We Collect?
 Examples of eLearning Companies and their Use of Big Data
 Data Analysis
 Existing implementations using educational data
 Methods that work well for educational data
 MOOCdb
 Privacy concerns
 International Educational Data Mining Society
 Founded July 2011
 EDM workshop in 2005 (at Association for Advancement of
Artificial Intelligence)
 EDM conference in 2008
 Journal of Educational Data Mining (JEDM) since 2009
 Society for Learning Analytics Research
 First conference: Learning Analytics and Knowledge (LAK)
 Journal of Learning Analytics, founded 2012
Why Collect Educational Data?
Why Collect Educational Data?
 Personalize education
 Better assessment of learners
 Multiple dimensions: social, cognitive, emotional, meta-
 Multiple levels: individual, group, institutional levels
 To promote new scientific discoveries and to advance
learning sciences
 Many theories; little hard data to support them
 Opportunity to discover new learning patterns
Why Collect Educational Data?
“Not only can you look at unique learning trajectories of individuals,
but the sophistication of the models of learning goes up enormously.”
Arthur Graesser, Editor,
Journal of Educational Psychology
A Look Backwards
 Collecting educational data was highly resource-intensive and
difficult to scale
 Much of the data that was easily collectible was purely
summative in nature
 Getting data on learning processes and learner behaviors, in
field settings, required methods like
 Quantitative field observations
 Video recordings
 Think-Aloud studies
 None of which scale easily
Learning Types
Learning Types
 Visual (spatial)
 Auditory
 Kinesthetic / haptic
Learning Theories
 Problem-Based Learning
 Anchored Instruction
 Cognitive Apprenticeship
 Situated Learning
 WBI –Web Based Instruction
 Learning technology
 Networking and computing technologies are used to improve
educational practices
 WBI –Web Based Instruction
 Learning technology
 Networking and computing technologies are used to improve
educational practices
Massive Online Open Course
What Data Can We Collect?
What Data Can We Collect?
 Administrative data - who are you?
 Address, name, birth date
 Content data – inferred properties about material
 Difficulty, subject
 Longitudinal data - data from a long period of time
 Grades
 Standardized testing results
 Time on task
 Attendance
 Click patterns
 How long a student holds a mouse pointer over a particular answer
What Data is Available Already?
 PSLC DataShop
 a central repository to secure and store research data
 a set of analysis and reporting tools
 >250,000 hours of students using educational software
 >30 million student actions, responses & annotations
 Actions: entering an equation, manipulating a vector, typing a
phrase, requesting help
 Responses: error feedback, strategic hints
 Annotations: correctness, time, skill/concept
 http://pslcdatashop.org/about/
Online Education Formats
 Video
 Online modules
 Written documents
 Audio files
 Instructions for activity or task
 Embeds technology directly into digital textbooks
 Provides an “engagement index score”, which measures how
much students are interacting with their eTextbooks (viewing
pages, highlighting, writing notes, etc.).
 Researchers have found that that the engagement index score
helps instructors to accurately predict student outcomes
more than traditional measurement methods, such as class
 Site and smartphone app to help people learn foreign
 Luis von Ahn
 Professor at Carnegie Mellon
 “twofer”
Data from duoLingo
 How long does it take someone to become proficient in a
certain aspect of a language?
 How much practice is optimal?
 What is the consequence of missing a few days?
 There are theories about learning languages, such as the idea
that adjectives should be taught before adverbs, but
previously, there was little hard data to support these
Conclusions from duoLingo Data
 The best way to teach a language depends on the students’
native tongue and the language they’re trying to acquire
 Example: Spanish -> English
 “it” tends to confuse and create anxiety for Spanish speakers,
since the word doesn’t easily translate into their language
 Women do better at sports terms
 Men do better at cooking and food terms
 In Italy, women as a group learn English better than men
Learning Analytics Implementations
 Still very few
 Knewton : https://www.youtube.com/watch?v=LldxxVRj4FU
 Signals project at Purdue University:
 Ellucian Degree Works, “a comprehensive academic advising,
transfer articulation, and degree audit solution that aligns
students, advisors, and institutions to a common goal: helping
students graduate on time.”
 Blackboard Analytics http://www.blackboard.com/Platforms/Analytics/Overview.asp
Analysis Methods
 Prediction
 Structure Discovery
 Relationship Mining
 Develop a model which can infer a single aspect of the data
(predicted variable) from some combination of other aspects
of the data (predictor variables)
 Which students are off-task?
 Which students will fail the class?
Structure Discovery
 Find structure and patterns in the data that emerge
 No specific target or predictor variable
Relationship Mining
 Discover relationships between variables in a data set with
many variables
 Correlation or causation
 Collaborative, online learning research
Different Formats of Data
SQL Dump
Student state information
XML files
Course information
EdX Platform
Emails and Surveys
JSON lines
Clickstream data
Multiple Platforms and Data Control
 EdX and Coursera
 Controlled by MIT and Stanford, separate entities
 Data model to organize raw data streams
 Unifies different platforms
 Each class:
 Student Information Tables
 Observations Tables
 Submissions Tables
 Collaboration Tables
 Feedback Tables
Benefits of MOOCdb
 Public, shared data model; avoid redundant work
 Foster analytic consistency
 Engage more people
Resource use compared by country
Privacy Concerns
 Hardcopy records were phased out in favor of district-based
hard drive storage some time ago, but the advent of cloud
computing has seen a trend toward the creation of thirdparty data silos (or clouds).
 Teachers and parents are concerned about privacy breaches
by hackers and marketers
 InBloom
 Gates-funded nonprofit that houses student data in the cloud
 Closed its doors after parental protest
Privacy Concerns
 This past May, the Obama administration released an 85-page
report on big data and its use in the US among consumers
and businesses
"Big data and other technological innovations, including new online
course platforms that provide students real time feedback, promise to
transform education by personalizing learning. At the same time, the
federal government must ensure educational data linked to
individual students gathered in school is used for educational
purposes, and protect students against their data being shared or used
History of Educational Big Data Policies
2011: FERPA
law is amended
once again,
of state
access to student
2005: New
granting money
to states that
Data Systems
1974: The Family
Rights and
Privacy Act of
1974 (FERPA)
2000: The
Online Privacy
Protection Act of
1998 (COPPA)
2008: FERPA
law is expanded:
vendors and
volunteers now
have access to
the data, with or
without parental
2011: The
Shared Learning
(SLC) — which
will later
become inBloom
— is created
Questions or Comments?
Email me at [email protected] with any questions.

similar documents