Conditional random field in the application of the product feature extraction

摘要

Extracted product features from user comments is the basis of fine-grained sentiment analysis, it is of great significance for manufacturers and users. Faced with product feature extraction accuracy is not high, firstly, the paper use the method of conditional random field(CRF) to identify the nominal information; Then, through the map of product feature make display semantic merge for product features, and FP-growth algorithm is used to extract the product feature candidate set; Finally, using TF-IDF and TextRank collaborative to filter non-product feature. Experiments show that the proposed method has good validity and applicability. The paper use real user reviews for study, the correct rate reached 85.7%, the recall rate reached 77.5%.