OK - this type of solution can be prototyped with text mining via Python or R. It is interesting to mix text analytics with graph / network analysis (also can be done in R). For enterprise scale, likely want to use supported and scalable tools such as SAS Text Miner working with a graph database like Neo4J (if wanting to mix text analytics with graph analysis, i.e. mixing parsing text for meaning and categorizing with examining a user's positional importance in a social network and/or discussion thread). For very large volumes, might consider a Hadoop architecture, with attention to ETL / import-output engineering (this can have a high overhead in Hadoop implementations). Hortonworks looking good as emerging commercial Hadoop árchitecture'... Good luck! I am currently with Deloitte, so not sure if you have appetite for commercial consulting, but always happy to discuss if so!