I am running the Open Data Service at the University of Southampton. http://data.southampton.ac.uk/
For the love of everything that is sacred, do not invent a new dictionary - there is already a lot of solid work.
This site is designed to become a model for other people to raise ideas. We used simple solutions that are easy to use for users, and all tools are open source. We mainly use simple spreadsheets and then XSLT to create RDF data. All this is also open source. All the tools used to create each dataset are published using the dataset to make it very simple to either directly copy our approach or use it as a starting point.
Choosing a good, simple, robust URI is key. We use the domain for this, and only this: for example. http://id.southampton.ac.uk/building/59
People - start with FOAF.
Personality roles - I suggest you take a look at the “organogram” released by the UK Civil Service .. this refers to a role as a type of group that contains zero or more people. (several people can have the same role, and roles can exist where people are empty)
Publications - we use bibo and dcterms for this. The hardest part is to carefully connect it with people and parts of the organization, etc.
Meetings - we do not currently model this, but you can simply use the standard "event" vocab if there is no reason: http://motools.sourceforge.net/event/event.html
Projects are what we are just starting to consider, and my current plan is to use http://ontoware.org/swrc/
Field of study; I have no good ideas yet.
We are trying to document reasonable start-up templates for this kind of tasks on openorg.ecs.soton.ac.uk - it will be an ongoing job. The goal is to create data with sufficient commonality of the structure that you can create with applications that work with several organizations, and more easily combine information from other sites.
I wrote a blog post demonstrating one tool that works with (places) RDF data created by Southampton, Oxford and Lincoln (although the Lincoln data server is currently down) http://blogs.ecs.soton.ac.uk/data/2011 / 11/25 / roi /
We have an RDF namespace called "openorg" for an additional dictionary that we found that we needed to complete our tasks. We try to minimize this, and most things are described using existing dictionaries.
As for creating a site on top of data, we use a PHP library called "Graphite" (which I am a leading developer and I am a Free Open Source). It is relatively easy to get started. http://graphite.ecs.soton.ac.uk/
This is all very new, so we will update things when we learn good and bad methods.