Our entity-level sentiment analysis works via a sophisticated combination of a (very) large dictionary of sentiment-bearing phrases and lexical techniques for associating these sentiment phrases with the entities.

You can have as many different sentiment dictionaries as you have space on your hard drive.  These are simple tab-separated text files that have the phrase and then the sentiment weight for that phrase.

When you call the Salience API, you specify which data directories you want to use for that Salience session.  Part of that is specifying which sentiment dictionary you want to use.

The Sentiment Toolkit is an efficient way to parse a large number of documents for potential sentiment phrases, weight those phrases, and put them into a "hand-scored" sentiment file for use by the Salience Engine.

The Sentiment Toolkit is distributed as part of the standard Salience Engine software package.