Separate the semantic ordering from the physical ordering. It keeps your organization flexible, since you can assign each paper to several topics and projects, and change the assignments as you see fit.
This can be achieved by using a filing cabinet or folders on the one hand, and a bibliographic database (be it electronic or in paper) on the other.
Label each paper with some identifier. I recommend the date or month, but also a running number will do. The label doesn't have to carry any meaning. Collect the papers with a certain range of labels in one folder. For example all papers with labels January to March go in one folder called "Jan-March 2018".
In your database, each entry should also carry the corresponding label under which the paper can be retrieved, in addition to the bibliographic data, and any semantic information such as projects and topic keywords. Use the database, not the physical location, for organizing the papers by substance matter.
How the database itself should be organized is another question. I use folders for projects and keywords for topics. Creating a systematic ontology or adopting an existing ontology (e.g. the Dewey system) seems overkill. Instead, I recommend creating keywords on the fly ("tagging") and combing through the keyword list once in a while. This will be more difficult with a paper-based organization than with an electronic database, of course.