Full Spectrum Opinion Mining: Integrating Domain, Syntactic and Lexical Knowledge

If NLP systems could better simulate how people would evaluate various states of the world in contexts of interest, this would make it easier to accurately extract embedded sentiments and avoid being led astray by solely linguistic cues. If this knowledge could then be combined with 'fullsemantics' linguistic processing capable of modeling the interplay between lexical and syntactic semantics and then interweaving these with domain knowledge, this would allow the use of important semantic information (including argument and, especially, valence structure) implicit in phrases such as 'Critics say' and 'Despite this.' The present paper seeks to implement these insights, employing domain models grounded in the INTELNET/COGVIEW 'energy-based' knowledge representation formalism and the Radical Construction Grammar-based COGPARSE parser, bringing together concepts, knowledge, language processing, and opinion mining.