Combining BERT semantic fusion and keyword feature extraction for aspect-level sentiment classification
Aspect-level sentiment classification aims to determine the sentiment polarity of a given aspect term in a sentence.Pre-vious methods for this task fail to extract semantically rich initial contextual representation vectors and cannot precisely capture the range of local key features.Therefore,this paper proposes KFE-BERTSF,an aspect-level sentiment classification model that com-bines BERT semantic fusion(BERTSF)and keyword feature extraction(KFE).BERTSF integrates high-level semantic informa-tion from the BERT encoder using a gating fusion function to extract semantically richer initial contextual representation vectors.KFE divides the sentence into local and non-local contexts using dynamic thresholds,and employs syntax distance mask(SD-Mask)and distance-aware attention(ADA)to extract local key features from both regions.Experimental results on three datasets show that KFE-BERTSF outperforms benchmark models.
万方数据
维普
