It's probably worth noting that in this context stemming is probably most appropriate. Lemmatizing is unlikely to give better results and in most implementations you suffer a significant performance hit.

Btw, in case you want a good lemmatisation implementation, there's one in NLTK that uses the WordNet database.