The simplest thing to do is to treat the response as a categorical variable with three possible outcomes. You might arguably say that "I don't know" is between "Yes" an "No", in which case you could treat the variable as ordinal, but this really isn't necessary since it is simple to deal with (non-ordered) categorical variables.
It isn't clear from your question whether you have longitudinal data (i.e., whether you can match the responses of individual respondents in the pre- and post- data) and whether the questions in the two runs of the surveys are the same. Assuming all this is true, and excluding cases of missing data, there are $3 \times 3=9$ possible outcomes for the pre- and post- responses for each individual question. Your data should allow you to see which of these nine outcomes occurs for each respondent-question, which should allow you to describe how the answers to questions changed between the pre- and post- surveys.