My organization is implementing a tagging system to assign meta-data to incoming emails.
The goal is to extract a more diverse range of information which in turn allows us to analyze trends & issues. I'm having problems putting together a vocabulary that is both broad enough to cover each possible scenario and that the user knows when it should be used.
Does anybody have any tips or suggestions?
If you are putting together a tagbase yourself you run the risk of infecting the content with your own personal perspective of what should and should not be tagged, whereas the intention of tags is for them to act as the users' own interpretation of what is meaningful to them.
Depending on the role of a person within the company, different people will be interested in different sets of tags - whether for marketing, technical, hr, sales, management, etc. My bet is that if you asked people to assign tags to the same bulk of prior mail you will get different sets of tags from each role, let alone the differential between the actual incoming mail that is assigned to each person.
I think it will be very hard to correctly initialize a valid static tagbase, without allowing it to evolve naturally. You might want want to consider a data-collection phase whereby for some period of time, people at the company each determine the tags that they would use on their incoming mail and then feed this back into your process, thus getting real users input on real data to instantiate the tagbase for the implementation phase. You might learn something interesting in the process too about what users actually understand about tags and how they use (or expect to use) them.
Keeping a tag set consistent over time is difficult for one tagger (e.g. blogs) and even more difficult for groups of people. (Look at any SE site, for instance.) So whatever your initial tag set, you should plan for tags to be refactored, renamed, and removed as the group reaches consensus or needs change.
If you plan for retagging from the start, then you can begin with coarse tags and refine over time. Do you (and your taggers) understand what the major "buckets" are? If you all independently tag an existing corpus of email, do you mostly agree on the answers? If you have consensus on initial coarse tags, I recommend starting with just that and allowing your taggers to nominate additional, finer tags during the tagging process. Eventually you will reach equilibrium, but I don't think you can go straight to that state.
Edited to add: if your users currently sort their email into folders, you might learn something by examining those structures.
Tags introduce a layer of abstraction that is nice and clean but introduces two places where you can skew data:
Based on your desire to analyze trends & issues, ie perform statistical analysis, you might get better results by directly analyzing the text of the messages.
This probably-out-of-date SO answer discusses algorithmically pulling tags out of text sources.
