Derrick Discussion

This page provides a summary of several discussion points from Colorado's discussion with Derrick Coetzee on 17 October 2013 (see the agfk group email for additional context):

• Since metacademy is largely based on enabling more efficient targeted learning, it would be excellent if we could orient the provided resources around the goal concept. Derrick provided the example of a user that wants to learn a specific concept in 3d graphics, where this concept has a large number of linear algebra dependencies. In a perfect system, the resources and learning path would be oriented towards 3d graphics, discussing the relevance of the prereq concepts to 3d graphics, and providing relevant examples throughout the learning experience. On a similar note, a user's particular background should ideally impact the learning experience: a chemist approaching physics topics should be presented with a different learning experience than a high-schooler learning physics concepts for the first time. These ideas both address the issue that our content is independent of the user and his goals. To ideally address this problem, we would have to take into account a model of the user's current knowledge, their goals, etc, as well as probably specify O(N^2) relationships between the concepts, and design a learning system for this representation. To approximate this system, Derrick brought up the idea of using "conditional resources" to take into account the user's current goals or general background. This idea could be applied at the individual concept level. For instance, we could specify 10 resources for a concept and have fields that specify the conditional use of the different resources

if (leaf concept is X): present resources 3, 4, 5 if (leaf concept is Y): present resources 3, 7, 9 if (user knows tangential field Z): present resources 1-5 else: present resource 3, 6, 8

• ^ This type of system would be backwards compatible with our current approach but also allow users to specify different resources for users with different goals or background. It might be interested to test the idea of "conditional resources" with content related to deep neural networks, e.g. resources with a general machine learning orientation and resources with an acoustic modeling orientation. Also, I've started working on a physics map, and I've found that a number of concepts would use a different set of resources if the user had a chemistry background or goal concept.

• Derrick pointed out that the content within the list of resources is a bit ambiguous: e.g. how should a user decide which resource to select if they want a decent overview of the topic, or if they want an in-depth analysis, or if they want to thoroughly learn why concept X depends on concept Y? Granted, our current implementation provides some details, but it's not clear why a user should choose one resource over another. Derrick suggested adding an "overview" section to our resources (overview, core, and supplementary) and also "starring" one resource in each section to help users choose a resource (note: the star could be determined by the users, e.g. by using an up/down vote system, where the most up-voted resource receives a star). Finally, it would be nice to know how/why a given resource is "supplementary," e.g. what aspect/extension/etc of the concept is discussed in the resource.

• Users don't like navigating to external pages, and they certainly don't like navigating to external resources that present them with a login screen. We should consider embedding content. Side note: a lot of coursera videos are hosted on youtube, perhaps we could link to these or simply download and host the content ourselves (this would probably violate their terms of service); we could also provide global metacademy login credentials for coursera/edx/udacity/etc, but of course users could easily change these (it's probably against their terms of service); is linking to pirated versions of textbooks considered copyright infringement (quick google answer: yes)

• Allowing user contributions: Derrick advocated using a wikipedia-style bold-revert-discuss system for content-based user contributions. That is, allowing [non-logged-in] users to contribute/change the displayed content for a given concept without requiring approval from a moderator or the original author. He raised the point that a large number of potential users will not contribute if we have any type of entry barrier---e.g. you must log-in to edit, you must have editor privileges, etc---and for a small project, spam, and non-useful contributions will probably not be much of an issue, while encouraging users to contribute will be an issue. However, since structural changes to the graph can propagate throughout metacademy, he mentioned that we would probably not want to allow a bold-revert-discuss system for dependency changes. A pending changes system may be more appropriate for structural changes. Also, allowing users to privately "fork" a graph and develop large changes before submitting their results would be ideal.

• From his wikipedia experience, Derrick mentioned that a number of (long term) users contribute to wikipedia in order to establish a reputation in the community (sites like stackoverflow also do this, but they also model their system so that you use your reputation outside stackoverflow, e.g. to get a job). Finding ways to let users establish a reputation within metacademy seems like a good initial goal for encouraging user submissions: e.g. perhaps we could do something akin to wikipedia's "good page" type of badges, or give users a score, or let users see/display the number of pageviews their page has garnered.

• Derrick suggested investigating the following related ideas: dynamic textbooks (I think this is an example: http://physwiki.ucdavis.edu), wikipedia books, and other projects that use dependency structures of knowledge (here's a wiki I started on similar projects that I'll try to improve soon: https://github.com/metacademy/metacademy-application/wiki/Related-Projects-Ideas)