Learning opinion-related patterns for contextual and domain-dependent opinion detection

摘要

A method for extracting opinion-related patterns includes receiving (S102) a corpus of reviews, the reviews each including an explicit rating of a topic. The reviews are partitioned among a predefined plurality of classes, based on the ranking. Syntactic relations are identified in each review (S108). The syntactic relations may each include an adjective and a noun. A set of patterns is generated (S112), each of the patterns having at least one of the identified syntactic relations as an instance and the patterns clustered (S116) into a set of clusters based on a set of features. At least one of the features is based on occurrences, in the predefined classes, of the instances of the patterns. A polarity is assigned to ones of the clusters and propagated to patterns in the respective clusters. The polarity-labeled patterns can each be instantiated as a contextual rule for opinion mining.