It’s evident that everyone in the tech world is obsessed with data: it is the fuel behind the ad you just saw and the email you just received. It’s how Facebook knows who your friends might be or how Amazon knows the shampoo you like. It can be creepy, but data is powerful.
Here at BenchPrep, we think great power comes with great responsibility (thank you, Spiderman). Each day our data scientists collect millions of data points over several dimensions: individual learning habits, discrete properties of course material and test content, test scores (broken into categories), and the time users spend on various parts of the platform—just to name a few. We know, for example, that students who play our custom-built learning games spend, on average, 18 minutes more per study session than students who do not. Or that students who have difficulty on Questions A, B, or C should be automatically redirected to relevant Topics X, Y, and Z.
We use this data to build and fuel an intelligent tool which will helps our customers learn faster and score higher than ever. Our adaptive elearning engine generates personalized study plans based on a variety of different data points, creating the most efficient learning experience possible.
In Part 1 of this two-part series on adaptive learning technology, I want to explain how we use data to organize our content before it makes it into your personal study plan. Part 2 explains how we take that content and make it into a personal study plan for our users.
Search Engines: Optimizing Data from the Web
In 2014, the internet had about 1 billion web pages (depending who you ask). That’s a lot of information to keep organized and make sense of. Enter: the search engine. Search engines like Google have developed mathematical algorithms which sort and rank web pages based on certain criteria in order to give the internet some structure and ability for discovery of useful information.
The process by which webpages are assigned relevance or importance by a search engine is called Search Engine Optimization, or SEO. Websites and pages are given their rank based on some combination of the following website data:
- keywords that users search for;
- how many other webpages link to the site;
- titles of webpages;
- what other users clicked on;
- properties of website text, image, or video content;
- whether the search term is a news item;
- reputation of the domain name; and
- social engagement and sharing.
The importance of each of the above factors is determined by the search engine’s algorithm. The exact combination is kept a secret, so that individual webpages can’t hijack the system for their own benefit.
BenchPrep: Optimizing Data from Learning
At BenchPrep, we have somewhere between 250,000-400,000 pieces of content on our system. This number doesn’t factor in the multiplication of these numbers that arise out of our users interaction with these pieces of content. While not in the billions, that’s also a lot of content to keep organized and make sense of. Enter: our own custom-built learning engine.
The mission: take old book-form content and digitize it to make it personalized and conversational.
We’ve developed an algorithm to sort and rank content files based on certain criteria to give BenchPrep material personalized meaning. The process by which we assign relevance or importance to each piece of content is called Learning Engine Optimization, or LEO. We give our content their rank based on some combination of the following data:
- semantic tags can mark the type and relative importance of lessons, e.g. Formulas, Key Takeaways, Rules, and Definitions.
- difficulty of questions
- topics across course materials: A student has difficulty with lines and angles while preparing for the math on the GRE. That student should be able to review lines and angles lessons from the ACT Math Prep Course.
- categories and subcategories are specific to each test: In the ACT Course, the test-makers might categories a question from “English” to “Usage and Mechanics” to “Punctuation”.
- connections where students who have difficulty with Questions A, B, or C are automatically redirected to relevant Topics X, Y, and Z.
- user feedback: users can flag questions, mark a high-medium-low confidence level, make notes, ask other users questions, or start discussion groups.
The importance of each of the above factors is determined by our learning engine’s algorithm. The exact combination is also kept a secret. Let’s take a closer look into what exactly happens in LEO:
- We start with static content from our education publishers. There is essentially no usable, fungible, manipulable data around it . . . let’s make it dynamic!
- First, we break the content down into three content types: Lessons, Flashcards and Questions.
- Then we add tags and categories to the content:
- - semantic tags to Lessons (Objectives, Notes, Equations, etc)
- - categories and subcategories to Flashcards and Questions (Math, Geometry etc)
- Next, we create connections between Lessons, Flashcards and Questions based on tags, categories and subcategories. These connections serve to link together the interactive parts of the learning platform with the more passive text and video lessons.
- We associate topics with each piece of content based on the related/connected files
- Voila! Now we have dynamic content we can use.
While this is happening, we are also collecting even more data from our users on each piece of content. We track how many notes or questions they’ve added to a piece of content, how difficult the question is, how long users spend on each piece of content, etc.
Finally, once we’ve determined as much data as possible around a piece of content, we assign it a rank based on difficulty and importance.
Did you catch all of that? It’s a little complicated, but it’s the backbone behind our adaptive learning engine. Content must be intelligently ranked before it can be distributed into adaptive plans, and none of this would be possible without creating dynamic and usable data.
But the fun’s not over yet! To learn how the adaptive learning engine creates personalized study plans from this content, keep reading more in Part 2.
Want to learn how to generate more revenue for your organization by utilizing a personalized learning platform? Download our case study, Taking Digital Learning To The Next Level, to learn more.
