This study takes an intersectional and community-informed approach to characterizing and mining the Twitter accounts of thousands of users that self-identify openly in their Twitter bios as members of the Bisexual+ community. Twitter data was chosen due to the documented hesitations many Bi+ community members have about reporting their identities when completing official health documents. Accounts were assessed by multiple coders for inclusion in the dataset, in order to best represent the great diversity of the community while removing fake, automated and inactive accounts. The presenters share both the nuanced classification guidelines used for this project, and the process of developing guidelines that are responsive to community rhetoric and the intricacies of mining social media -- real, messy "data." The presenters also share on topic modelling methods and challenges, the nuances of natural language processing with tweets and the results found in the health topics Bi+ users share about online.