
The file contains JSON formatted results of annotation of opinions and their targets. 

Each item in the list is a sentence which contains following fields:
'file_id': ID of review (used internally, same sentence may appear multiple times with different sentiments and opinin target pairs)
'domain': review type (perfume or clothes)
'dist':  dependency path distance between S and T, filled only for the 1st batch of annotations
'isSentIncorrect': human annotation - is sentiment word S incorrect?
'isAttrIncorrect': human annotation - is opinion target word T incorrect?
'parsedSent': CONLL-formatted parsed sentence; last column contains pointers to :
	S = sentiment word
	T = opinion target word
'isStrError': human annotation - is dependency structure erroneous between S and T
'isAtrRelToSent': human annotation - is S related to T 
'rule_id': ID of extraction rule that pointed to T (see "extraction rules")



