Reinventing IT | Viewpoint
Not Just Big Data--Fast Data
The recent ability of universities to process large amounts of data at superfast speed opens the door to automated, mobile, and deeply personalized education.
- By Vince Kellen, Ph.D.
A quick and not-so-quiet revolution has started in higher education. The front end of this revolution involves the likes of Stanford University (CA), MIT, and Harvard University (MA) all staking ground in the open-university land rush. The maneuvers by these premier brands are attracting a great deal of attention at the moment--and prompting some head-scratching as other universities ponder what it all means to them. But the revolution may be much bigger, more complex, and more helpful than what we see right now. I believe that the back end of this revolution, which involves so-called big-data technology, is where the greatest opportunity lies.
The term "big data" is something of a misnomer. As terms go, it is short and sweet and evokes a clear picture in the mind. But I prefer the term "fast data." The real promise of big-data systems doesn't lie in how much data they can handle, but in how fast these systems can retrieve and compute data. Some of the tools out there now make the promise of a superfast data warehouse--with data updates from various source systems occurring immediately--an accessible reality.
At the University of Kentucky, for example, it took us only a few days to set up our big-data analytic tool (SAP HANA) and start replicating data in real time from our enterprise system (SAP) into HANA. On the surface, the SAP HANA tool looks like a traditional database tool with direct access from different end-user tools. Underneath, HANA is built for speed, with parallel computing and an in-memory approach to high-speed analytics.
The platform lets us quickly create queries that are executed, on average, about 350 times faster than in the old system. Queries that took 15 minutes in the past now take 2.5 seconds. As a result, we plan on getting out of the old ETL business (extraction, transformation and loading, which is the practice of combining, filtering, and manipulating data in batches that run nightly). Instead, we are reallocating our IT staff to assist business users with modeling needs.
Improving Student Learning
What do we plan to do with all this horsepower? At the moment, our aim is simple: Improve the student experience quickly. But our ideas are bigger. Here are some of the concepts we intend to pursue:
- Create an intelligent reminder service that learns students' preferences and needs. While some students are fastidious, others require more reminding. We will comb through data from our LMS and student systems to build a personalized reminder service that doesn't overload students with spurious messages.
- Provide students with tailored advice based on their demographic data, academic background, and performance. Through conventional data analysis and statistics, many universities know the likelihood that a student will graduate. I always ask, "Does the student know this about himself?" After all, it is the student who must learn to adjust to the demands that college life brings. We plan to provide students with their own "academic health record" via their mobile devices. This academic health record will have actionable indicators for students, and easier digital access to key services, personalized information, and interaction with peers.
- Through an analytically driven alert system, we will automatically notify staff advisers when individual students need help. We currently have an alert system in place that works well, but it can handle only manual entries. Now we want to create alerts automatically, and fine-tune them based on analytical models. We will use our enterprise system workflow tool with our in-memory analytics engine, so that we can pinpoint exactly when students need intervention, send alerts to specific faculty and advisers, and make escalation and follow-up easier.
- Longer-term, we believe big-data analytics can power automated and personalized learning objects that complement--not replace--face-to-face instruction. Rather than "flip" a classroom by requiring students to consume canned lectures and static materials outside class, we would like to "flip" the classroom by pointing students at adaptive and intelligent learning objects that can adjust to individual learner attributes. These attributes will be a combination of what we already know about individual students (academic strengths and weaknesses based on high school or prior college coursework and test scores) and what we can learn about students while they interact with the learning object. Imagine learning matrix algebra from a smart learning object over the web that adjusts the display and method of interaction--in real time--as it learns how you learn best, using data owned and protected by the institution.
Our inspiration for these ideas comes from the social-networking world. Sites such as LinkedIn, Facebook, and Twitter use real-time analytics to recommend to users which groups to join, which members to connect with, and, of course, which ads to click. If you think about it, over the past decade, the web and e-commerce world has invested billions of dollars in high-speed analytics with literally one goal in mind: Get users to click on that darned ad. With big data systems now in the early stages of more widespread adoption, universities can do something similar: Use high-speed analytics to help students learn better and more quickly.
This back-end revolution is not without its challenges. Universities are very good at analyzing, planning, and then doing, sometimes with a very lengthy phase of analysis. High-speed analytics will require universities to move instead toward small, fast "do-learn" cycles, where the goal is to build an intervention with a small audience and then learn quickly how to improve it. This fast-fail approach, long a hallmark of entrepreneurial startups, will need to take root in university administration.
Recently, I was explaining the promise of big-data technology to a colleague, a computer science faculty member who certainly understands the mechanics and the capability of the technology. But when we started talking about adaptive learning objects and automated learning, he asked an important question: "But what will I do?"
I quipped, "Stop giving boring lectures and instead focus more deeply on individual students." We laughed together. Over time, by using more automated forms of self-study through smart learning objects, perhaps universities can improve the ratio of students to teachers in face-to-face settings. Flipping a class, whereby the lecture is delivered online and smart learning objects let students learn material outside normal class time, can provide significant benefits inside the classroom. The long-term savings from increased automation and self-study will help universities afford smaller classroom sessions with more student-teacher interaction. This will give students what they crave: rich, meaningful, and, in many cases, profound interactions with their teachers. It's not an either-or proposition. The online and face-to-face learning experiences need to be more personal and more compelling, not less. They need to work together.
A few public universities--stressed by our mission to educate more students with varying abilities even as budgets continue to shrink--have already started to discuss how to use these big-data technologies to help our students. Our conversations with the University of Nebraska and Central Michigan University, for instance, have been thought provoking.
The real revolution in higher education has started, and it's not the biggest or the richest institutions that will necessarily emerge on top. With IT, the barriers to entry are lower than before and getting lower. Universities are now only limited by their imaginations and their organizational will. Ultimately, the institutions that finish ahead will be those that master the organizational and technical challenges to provide a richer student experience.