.png){kind=icon}
Return
A mixed unsupervised method for aspect extraction using BERT
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-022-13023-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
With the increase of unstructured text on social media platforms from user opinions, deep neural network techniques have significantly contributed to the aspect extraction subtask of Aspect-Based Sentiment Analysis (ABSA). In a multi-sentence review, sentences are contextually interdependent, and static word embedding generates similar representations for the same word in different domains. Hence, existing techniques cannot capture inter-sentence dependencies for valid multi-word aspect extraction. Further, incorporating conceptual information to associate the context and aspect terms is still a challenging task. Therefore, this paper aims to remove inadequate information and capture aspect co-referencing by adding a sentence coreference resolution step before performing ABSA in an unsupervised rule-based method. Next, domain irrelevant aspects are pruned out using contextual embedding. Furthermore, aspects extracted using unsupervised way are given as labeled in training the hierarchical attention-based network using pre-trained language model BERT, Bidirectional Encoder Representations from Transformers. The experimental results on the SemEval-16 dataset show that F-score results are between 2.5% and 5% better than recent supervised deep learning approaches for laptop and restaurant domains, respectively.
Keywords:
Aspect extraction
BERT
Deep neural network
Sentiment analysis
Unsupervised learning
Journal
IF:
3
Papers:
2.0W
Citations:
3.2W
Organization
Cited Papers
.png){kind=icon}
ShareSaveA new topic modeling based approach for aspect extraction in aspect based sentiment analysis: SS-LDA
PREAI
ShareSaveShareSave
ShareSave
View more