Personal knowledge organization a is a very intimate choice. Something that works for one, may not work for another. Instead of providing a system, I'm going meta - a system to think about knowledge organization. Outlining the key philosophical principles allows clarity of thought. This document should prove helpful to anyone who's scouring the internet for panacea.
To understand how to organize, we need to identify the key system elements. I stress a lot on interoperability - our ability to avoid vendor lock-in and stay flexible to adapt new systems to the content. This is in contrast to choices where one would adapt content to systems and make migration hell. Along the way, we address any crossroads that may pop up.
A realization of this philosophy is outlined in Elements of the Knowledge Base.
Everything starts with just raw letters and words and a single unit of content would be a text file. This is the source of truth for all content in the knowledge base. Value is added to this content by applying appropriate transforms to this raw piece of text - e.g. converting to HTML and styling with CSS for interactive web presentation, converting to a PDF for static document exchange, categorizing in a navigation menu etc. Starting with simple files allows extreme interoperability. In case a better system becomes available in the future, files will be the least painful to migrate. Of course, under this philosophy, the new system must operate with files or else it wouldn't be better in the first place. This page itself started as a simple text file.
Each unit of content needs associated metadata - a simple key-value store associated with the file. Metadata allows flexible configuration for the transforms mentioned before. For instance, a system that allows date-based sorting can be facilitated via metadata fields. Metadata should be stored within the text file for maximum portability and atomicity of the content - everything that one wants to know about the file is within the file. Further, allowing only primitive values (text, integers and lists) in metadata maximizes portability. This page has its own metadata for creation date, modified date, descriptions etc.
We need an interface to interact with the text files and metadata. Naturally, text editors come to mind. A key principle to remember is that the text editor shouldn't dictate how the interaction happens. Its only purpose is to facilitate the interaction and then move out of the way. Enforcing this principle maximizes interoperability. In this spirit, stay away from WYSIWYG editors and their enticing custom configurations when possible. It is more distraction and less writing.
A collection of such units of content need a storage medium. The storage medium
should have no bearing on the presentation of content to maximize interoperabilty.
Storage medium should be strategically picked to optimize for ease of access and
flexibility of read/write operations. The good news is that the filesystem
on our computers is the best to work with text files. However, this is a good
place to start thinking about organization. A filesystem augmented with
allows further goodies like tracking changes.
Arguably, the most popular format is Markdown which provides a reasonable hierarchy for organizing text within the unit of content, our text file. This page was itself written in markdown and subsequent transforms were applied for layouts and styling.
Navigation connects all units of content in a knowledge base. We realize the greatest benefit from a knowledge base, when we are able to discover connections between disparate items. This is the purpose of this exercise.
Deliberately thinking about navigation while writing content also largely reduces the need to have advanced search. Once we've semantically categorized knowledge, random search queries play a less significant role. Search effectively becomes a guided navigation than a haphazard spray of letters.
Before anything else, the system needs a way to identify uniquely what to navigate to. Each unit of content must be assigned a unique identifier. These allow us to stay robust to pretty much all externalities. For instance, when using URLs, aliases can be utilized in the event of change of URLs.
Clearly, the storage medium should not dictate the generation of identifiers. Most static site generators rely heavily on this. Even the Apache Web Server relied on this anti-pattern. I think this is reckless. Of course, I am guilty of doing this multiple times in the past. Changing the storage medium or organization within that shouldn't change identifiers.
Every unit of content should be assigned a primary classification. Indeed, content may not be exclusive to a single class and can be easily extended in metadata. To exponentiate capacity of any classification, it is easy to introduce sub-classification. For instance, using 10 classifiers with another 10 sub-classifiers within each already allows us a 100 classification schemes. Going beyond these two levels may defeat the purpose and may not be really needed. Remember everything is fluid. No unit of content is ever complete and no organization will ever be enough.
Classifications should largely stay static. Sub-classifications however can be semi-fluid. I say semi-fluid to convey that we don't want an outpour of sub-classes with every new unit of content. This forces us to rethink the semantics of the knowledge we gather and stay organized.
Once we've written, stored and organized content, we need a systematic way to surface what we know. It is easy for information to get stale.
Every classification should be introduced via an overview page. Think of this as a meta-page contextualizing information within the class. The precise content is of course subjective and can be presented in varied style. In addition, we want to make sure everything under the classification is surfaced in a list page. This list could also be a part of the overview page and also contain any sub classification.
Overview pages are a way to discover intra-class connections. Inter-class connections should be surfaced in a portal page. This is the place where we realize connections between previously disparate pieces of information.
The concept of overviews and portals obviates the need to have an advanced search system. Sophisticated linking between units of content and visually-appealing graphs is largely a good to have feature. Learning happens when one deliberately contextualizes information and draws connections. Links are an effect of learning and not the cause.
It is easy to fall into the trap of alternative systems. For instance, relying on custom relational databases with all normalized column definitions and a content management system on top. These cut away a lot of the writing and thinking time. However, without doubt there is broader utility of such systems. Wikipedia couldn't possibly be run using text files. We aren't trying to build Wikipedia though, are we?