Comment Target Extraction Based on Conditional Random Field & Domain Ontology

Comment target extraction is the first task for sentiment analysis. Conditional random fields (CRF) method has achieved good performance on the Named Entity Recognition. In this paper, CRF model is used for comment target extraction in Chinese in combination with domain ontology of products with word and part of speech as features. Domain ontology of products is considered as a semantic feature, such as the name of the product, the component of the product (model appearance, function etc), synonyms or upper and lower relationships, which are used to extract implicit comment target. Experiments on the third task corpus of COAE2008 showed remarkable performance of the proposed approach. Combination of ontology and CRF method increased the precision and recall of comment target extraction, achieved 86.56% and 77.23% on cell phone data, 83.16% and 73.97% on notebook data.