I am a MS candidate student who is doing research on tag recommendations. My research topic is tagging recommendations for Stack Overflow based on developers' experiences.
My hypothesis is:
If the system can recommend tag terms according to some various criteria, but not only with frequecy or relevance, an unbiased and more diverse tag set can be annotated.
The criteria can be defined in the perspective of developers' experience: namely, question and answer sets within the development domain. Some main criteria in question and answer set can be defined as tentatively:
- the problematic situation, - e.g. error, error-handling, memoryleaks, debugging
- development environment the article-writer is working on - e.g. c#, c++, visual-studio-2005..
- the recommended solution in the answers provided
In other words, my observation is that we usually (when I go over the tags) annotate articles with tags immediately and intuitively. So it is highly probable to be biased, escpecially in the direction of development environment such as language name, framework name, service model, etc. (shown as above)
My concerns is that these kinds of criteria should be justifed prior to conducting the experiment.
So, I want to ask: what do you think of those criteria? Does it make sense to you, and is that how you tag?
If not, or you have any other suggestions, I could use some feedback.