大数据科学与人才培养的互利关系 Education for Big Data and Big Data for Education: Towards Integration of Big Data and Education ChengXiang Zhai (翟成祥) Department of Computer Science University of Illinois at Urbana-Champaign USA BDSE2016, May 25, 2016, Guiyang, China
The Big Data revolution: “DataScope” enhances human perception (数据镜) Microscope Telescope
of Real World Variables Joint Mining of Non-Text DataScope enables prediction & optimal decision making Predicted Values of Real World Variables Change the World Multiple Predictors (Features) … Predictive Model Teacher Student Joint Mining of Non-Text and Text Real World Sensor 1 … Non-Text Data Sensor k … Text Data
Big Data creates both challenges and opportunities for education Challenges for education: Educate many data scientists & engineers quickly and affordably Opportunities for education: Leverage Big Data technology to scale up and improve education Big Data and education are mutually beneficial Integration! Education supplies workforce for developing innovative Big Data technology and applications Big Data supplies technology for scaling up and improving quality of education Education for Big Data Big Data for Education
Rest of the talk Education for Big Data Big Data for Education Integration of Big Data and Education
Part 1: Education for Big Data “….(in the next few years) we project a need for 1.5 million additional analysts in the United States who can analyze data effectively…“, -- McKinsey Big Data Study, 2012 The need is global …
Educating workforce for Big Data Question 1: What to teach in Big Data? Question 2: How to teach Big Data effectively at large scale with low cost? PhD, MS, BS in Data Science Massive Open Online Courses (MOOCs)
What to teach? New degrees in Data Science? Application Cloud computing Artificial intelligence Operations research Human-computer interactions … + Health, Medicine, Finance, Smart City, Education, … Analysis Highly interdisciplinary! Data mining Machine learning Statistical modeling Scalable systems … Acquisition Aggregation Databases Information retrieval NLP, Computer vision … Sensor network Internet of things Statistical sampling …
How to teach? Emergency of Massive Open Online Courses (MOOCs) Many platforms: Coursera, Edx, Udacity, 清华大学慕课平台,… Characteristics Free/affordable education at large scale on all kinds of topics Limited assessment support, but strong online community support Partnership with universities Early stage of “education revolution” enabled by IT & Big Data (more later)
My experience with MOOCs Taught 2 MOOCs in 2015 = CS410 Text Info Systems at UIUC Text Retrieval and Search Engines Text Mining and Analytics Coordinated Data Mining Specialization: 5 courses + Capstone Pattern Discovery Cluster Analysis Text Retrieval Text Mining Visualization Capstone Project
Text Retrieval & Text Mining MOOCs Each lasted 4 weeks Modularized video lectures Weekly quizzes Programming assignment (open challenge with a leaderboard) with auto grading Enrollment ~50,000 signed up > 10,000 seriously watched lecture videos 1,000~1,500 completed the course 700~900 did programming assignments
Students are from all over the world! 64,651 Learners 181 Countries
The majority of learners are 25~44 years old
US, India, and China have most of the learners United States India China
Most learners have full-time job and {BS, MS} degree
Challenges in teaching “big data” at large scale General challenges in MOOCs Variable student background Variable student needs Reliability of assessment Special challenges to “big data” Programming assignments are essential: variable student resources & background Availability of interesting real-world data sets Automated grading of programming assignments
Programming Assignments for Text Retrieval & Text Mining MOOCs Coursera provides no computing resource Students must work on programming assignments on their own computers Students download assignments to their own computers Auto grading is necessary Grading the output of a program, not the code Help students learn complex algorithms with minimum effort We provide a sophisticated toolkit to every student (through a virtual box image) Students only touch the key algorithm components in the toolkit Students can experiment with existing algorithms and explore new algorithms Leverage students to create data sets Crowdsource annotation of data sets to the students in the course Open competition with public leaderboard to encourage creative exploration
Self-Sustaining Data Set Annotations & Open Challenge Test Collection Open Challenge Competition Assignment ... Annotation Assignment ... Auto Grader Annotations ... Leaderboard #1 Team1 0.81 #2 Team 2 0.75 … Raw Data Set
Example of a new data set (for online course retrieval) High grades More reliable annotations
Search Engine Contest: Leaderboard
Overall lessons from the MOOCs Learners of MOOCs are a different crowd than the on-campus students Practical mindset, self-motivated, but less background and less time Pre-quiz is necessary for such technical courses (set realistic expectation) Learners form self-supporting online communities Short modularized lecture videos are preferred Programming assignments are very much appreciated Crowdsourcing annotations and open competition worked well MOOC goes beyond education to support research! Limitations of current MOOCs Lack of “individual care” (students don’t all get the needed help) Solely rely on peer grading of sophisticated assignments (unreliable grading & ineffective feedback to students)
Current Trend: Integration of MOOCs and Traditional Education Flipped/Blended classroom + Traditional Classrooms Quality LOW cost Online Degree High Engagement component + HIGH cost Campus Degree MOOC MINUM cost Specialization Certificate MINUM cost Course Certificate FREE No Certificate Scalability
A new online MOOC-based program: MCS-DS at UIUC MCS-DS = Master of Computer Science in Data Science Tuition = $20,000 Courses =MOOCs + High Engagement Components Interdisciplinary Courses mostly offered by Computer Science Department Data Mining Specialization Cloud Computing Specialization Machine Learning Other units include School of Information Science & Statistics Department
Part 2: Big Data for Education Quality Scalable Intelligent MOOC Small Classrooms Towards Intelligent MOOC “Big Data Technology” MOOC Automate grading with machine learning Automate question answering on forums Scalability
Traditional Manual Grading Submitted Assignments Graded Assignments Grade: 93 85 …. Proposed Automated Grading Graded Assignments Submitted Assignments Grade Verification Multi-dimensional Grade Predictor Detailed Grading Results Batch grading Clustering Improvement Performance & Behavior Analysis
Preliminary results on grading medical case assignments are promising [Geigle et al. 2016] Chase Geigle, ChengXiang Zhai, Duncan Ferguson, An Exploration of Automated Grading of Complex Assignments, ACM Learning at Scale 2016.
Towards Intelligent MOOC: Limitations of Current MOOC Instruction materials limited to those pre-defined by an instructor can’t take advantage of useful materials on the Web Limited search capability inside a course can’t easily find the most relevant video clip or discussion posts about a topic No understanding of students can’t personalize the instruction and learning experience Limited support for collaborative learning can’t leverage massive student behavior data to recommend materials for individual students Limited support for interactions with students can’t engage students in a natural dialogue
Novel Features of an Intelligent MOOC Seamless integration of MOOC and Web search enable students to learn from the Web Concept/Topic search, navigation, and summarization enable students to quickly find all materials about a concept or topic Dynamic and adaptive student modeling enable deep understanding of student state of knowledge Lifetime learning from student behavior data enable effective support of collaborative learning Interactive personalized teaching enable personalized natural conversations between students and the system
Traditional MOOC Platform Current MOOC Student Record Traditional MOOC Platform MOOC Course Content … MOOC Activity Log
… An Intelligent MOOC Open Web MOOC Course Content Student Model Modeler Open Web Concept Recommender Personalized Search agent Topic/Concept Graph generator Interactive Teaching Interface … Concept/Topic Search agent Concept Navigator MOOC Course Content MOOC Activity Log
Many existing technologies can be applied Algorithms for intelligent information retrieval Interactive personalized search technologies Algorithms for dynamic topic map generation Algorithms for topic discovery, summarization, and analysis Algorithms for search log analysis Algorithms for opinion integration and summarization Algorithms for collaborative filtering and recommender systems …
Part 3: Integration of Big Data and Education Educate Intelligent MOOC Platform ? Scalability & Quality Improve Research & Develop Applied to MOOC Log Education Big Data Big Data Technology
Toward a Cloud-based Big Data Virtual Lab … Leaderboard #1 Team1 0.81 #2 Team 2 0.75 … Log Data Leaderboard #1 Team1 0.5 #2 Team 2 0.3 … App Data 1 App Data N … Big Data Tool 1 Big Data Tool 2 Big Data Tool 1 Big Data Education System …
Unification of education, research, and applications! 4. Industry data sets not released to students & researchers Privacy-preserving Big Data education & research 3. Well-archived interaction history Reproducibility of research 2. Encourage open exploration (research) Remove gap between education & research 1. Directly work on industry data sets and problems Remove gap between education & applications
Final Thoughts: Education Revolution & Automation Big Data and IT enable education revolution and automation toward more affordable high-quality education IT enables one teacher to teach many more students than before (efficiency) Big Data technology would enable “automated” TA/instructor (scalability) Intelligent MOOC would improve quality of education at low cost Implications: Many traditional boundaries will likely disappear! No strict distinction between a teacher and a student (everyone learns from each other) No strict distinction between grade levels or age groups (learn at your own pace) No inherent boundaries between different courses (due to high modularization) No boundaries of subject areas (due to high modularization) No boundaries of institutions (MOOCs unify all institutions!)
Thank You! Questions/Comments?