Product Idea: A Community Compiled, Searchable and Traversable Skills Graph with links to resources to attain those skills
April 05, 2019
Where this came from
I worked at Coursera from 2014 to 2018. We used to have twice yearly hack-a-thons, (known more fondly as Make-a-thons) that my friend Jon started back when Coursera was still in its infancy. During these make-a-thons, bright graduates from the world’s best education system were scratching their own itches – coming up with ideas and MVPs for future education technology.
One of the most interesting and exciting ideas that was suggested back in 2015 was unfortunately never built: “A traversable skills graph that unbundled learning into the most granular concepts.”
I starkly remember the excitement that I had when I was chatting about it with Nikhil, Jon, and a few other early Courserians back in 2015. We couldn’t stop talking about it. It would be amazing have a historical record of your education laid out against all the concepts in the world. It would be empowering for students that are looking to expand their horizons.
Four years later, this still doesn’t exist in any form so I’m resolving to put together a MVP and validate whether or not this should exist in the world.
The rest of this post intends to explain why I’ve concluded that it should.
The bigger picture
Education needs to be viewed in a structured manner. People need to get a sense of what their learning pathways look like — whether or not it is prescribed to them or if they can piece it together themselves.
The First Ed-tech Boom
This wasn’t very viable back in 2015 because it wasn’t the biggest market problem. Successful Ed Tech companies that came up in the late 2000s and early 2010s focused around content creation/delivery, learning management systems, and alternative instruction.
Professors were starting to upload their lectures online, and some more enterprising entrepreneurs realized that their was a huge market of underserved people who would consume these videos like crazy. There was clearly a need for this content to move from physical classrooms to being online. Why not?
As the market started to get competitive and content strategy needed to evolve, it became pretty clear that pathways within education were one of the most important factors towards providing a useful tool for people.
It was easier to sell a single specialization (bundle of courses) that promised mastery of a skill because that was a concrete outcome from a concrete plan. People resonated well with this sort of product offering — paying a couple hundred dollars for 3 - 9 month programs. Companies in this space spent the next couple of years experimenting with different content bundles: micro-degrees, nano-degrees, full online degrees. These were all exciting products that were structured and pushed students towards a life changing outcome.
Increasing top of funnel
All the education technology companies I’ve seen since then have followed the same model, but education was still expensive — and the promise of a dramatically improved life was still too costly for some potential students. Ed-tech companies dove further into optimizing the payment model to drive the cost down — leading to more students. Notably, Lambda School offers a bootcamp program that only asks for payment after you land a high paying job.
A long way to go
One of the key phrases that was being touted back in the early days at Coursera was that education was a multi-trillion dollar industry that needed disruption. We had a lot of room to grow. Fast forward to today, and despite the existence of multiple ed-tech unicorns, technology companies have only claimed a tiny fraction of that pie.
Still, a considerable amount of the population are not re-skilling or re-training as some economists predicated.
I’m willing to bet that a part of the answer is in the lack of personalized learning.
Standardized learning pathways (curriculums), are great for pushing out students — but standardized pathways also exclude personalized needs. Schools usually make up for this by creating special needs programs or programs for the gifted and talented. Other deficiencies in standard curriculums are rectified by specialized curriculums designed for students in private school systems or through private instruction. Not all communities in the world have access to these resources, however … at least not yet.
Google Maps for Learnable Concepts
Imagine if all the learnable formalized concepts were thrown into a graph that was searchable and traversable exactly how the Earth is made searchable and traversable through Google Maps. A student in the 3rd grade might be able to just compute the shortest path to becoming an astrophysicist and get a birds eye view of what they need to get done to be eligible for that job.
While students are attempting to make hops along the graph, the product could make recommendations for local tutors, educational institutions, or online courses that could help accelerate that students learning.
Additionally, this could be a central hub for students and parents to discover a more personalized learning pathway that leverages the wealth of online content and the market of local tutors.
Looking at prior art
There has been some movement in the space of skills graphs or knowledge graphs. These have generally fallen into two buckets:
- skills graphs with high level concepts to give learners and enterprises a shallow understanding of what courses to enroll in
- knowledge graphs that are incredibly useful to answer questions about the world but are more interested in being engines to answer questions about the world or to find research papers (Open Knowledge Graph)
- How our Skills Graph is helping learners find the right content to reach their goals (Emily Glassberg Sands, Coursera)
- LinkedIn’s Economic Graph
- DiffBot’s Knowledge Graph
- Salesforce Trailhead Skill’s Graph
- Khan Academy Knowledge Map
For all of the companies above, the graph is positioned as a small product to serve the core product or other corporate interests. Understandably, there is no core business metric that they could associate to building a generalized skill graph that people could use to chart out their knowledge progression.
A product vision
This Wikipedia-like crowd-sourced concept graph would allow for easy visual traversal and search. As a traveler, the graph can be used as a tool to understand the depth and breadth of different topic areas, and also to plan one’s learning.
A suggestion / moderator-commit model can help ensure that the graph grows over time through intrinsically motivated community contributions.
After building the graph as a platform, a graph layer can show personal achievement graphs on top of this concept graph. Concept nodes can reveal teachers in the area that can jump on a 1 hour video chat to teach a concept. Pathway memberships can exist to seamlessly provide content from Lynda, Coursera, Udacity, and local tutoring programs, all at once.
A graph-based representation
In order to build a graph that is flexible enough, the designs for nodes and edges need to be simple but powerful.
Here are some rules for the nodes in the graph.
- The graph will be represented as some combination of force directed tree layout, multi-foci force layout, etc. This allows the graph to automatically shape itself when nodes are added. Exact implementation still TBD.
- Nodes will have implicit edges scoring how related they are to each other by default. These implicit edges are intended to be invisible during visualization, but will determine the repulsive forces within the layout. How this implicit score is calculated is still TBD but will probably be a factor of it’s explicit relationships.
- Nodes are foldable/expandable. This will be the primary method by which to prevent visual overload. Eg. Expanding an “Algebra” node will reveal Algebraic Expressions, Formulas, Functions, etc. Each of those nodes can also be expanded or folded.
- The graph is intended to be alive and evolving. Contributions and modifications can be made by anyone but must be vetted by an assigned expert/moderator before being committed.
- Nodes must be learnable and/or achievable concepts. No exceptions.
- Nodes must be local to a domain. Same or related concepts in different domains are separate nodes and can have edges authored between them depicting their relationship.
- Concepts must have standardized names and/or legitimate names. Examples: Math, Data Science, Quadratic Equations.
- There are two types of unidirectional edges: “part of” and “prerequisite for”. This is subject to change.
- There are two types of bidirectional edge: “same as” and “related to”. This is subject to change.
- In order to keep the core graph simple, extensibility of the graph will exist through a layers system.
- Layers can define behaviors on both nodes and edges. An example layer might allow someone to track their own achievements, mastery, and / or add notes.
- In order to support many use cases, despite trying to construct a “main graph”, the underlying technology should allow the creation of federated graphs that are unattached to the main graph.
- Attaching these graphs should be as easy creating edge connections. Federated graphs can stay separate forever.
Still a WIP.
This post is still a work in progress but will be done soon. Please check back at a later time, or send me an email @ lewis [dot] f [dot] chung [at] gmail [dot] com and I’ll push an update to you.
Written by Lewis Chung. Founder @ ShopWith. Previously: Coursera, Amazon. Writes about technology, products, and life.