VaderSharp – Sentiment analysis using Vader with C#
github.com
github.com
When I've seen teams do manual classification to test out the accuracy of their tools, it has turned out that the accuracy is generally all around horrible, most tools heavily bias towards "neutral" which really means "not sure". Not too surprising, tweets doesn't provide much context.
Also all the capitalization logic has meaning only in languages that use a bicameral script.