Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence

NAACL 2019  Β·  Chi Sun, Luyao Huang, Xipeng Qiu Β·

Aspect-based sentiment analysis (ABSA), which aims to identify fine-grained opinion polarity towards a specific aspect, is a challenging subtask of sentiment analysis (SA). In this paper, we construct an auxiliary sentence from the aspect and convert ABSA to a sentence-pair classification task, such as question answering (QA) and natural language inference (NLI). We fine-tune the pre-trained model from BERT and achieve new state-of-the-art results on SentiHood and SemEval-2014 Task 4 datasets.

PDF Abstract NAACL 2019 PDF NAACL 2019 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Aspect Category Detection SemEval 2014 Task 4 Subtask 3 BERT-pair-NLI-B F1 score 92.18 # 1
Precision 93.57 # 1
Recall 90.83 # 1
Aspect-Based Sentiment Analysis (ABSA) SemEval 2014 Task 4 Subtask 4 BERT-pair-QA-B Accuracy (3-way) 89.9 # 1
Accuracy (4-way) 85.9 # 1
Binary Accuracy 95.6 # 1
Aspect-Based Sentiment Analysis (ABSA) Sentihood BERT-pair-QA-B Aspect 87.9 # 1
Sentiment 93.3 # 2
Aspect-Based Sentiment Analysis (ABSA) Sentihood BERT-pair-QA-M Aspect 86.4 # 2
Sentiment 93.6 # 1

Methods