You need to extract or tag the sentiment words that are associated with the features. You can use the Dictionary Tagger to find the Sentiwords within your text.
You can do the same with your feature set. Then extract the features and sentiwords and group on the document id or features and summarize the Pos/Neg Scores. You will get a tendency if the feature is rated more towards positive or towards negative.
But be aware that this is a very basic approach. Other maybe important factors like negations, irony/sarcasm etc. won’t be covered. I experienced that word lists for sentiment analysis often perform bad in other domains than those they have been extracted from originally (e.g. News vs. Social Media data).