SA-ASBA: a hybrid model for aspect-based sentiment analysis using synthetic attention in pre-trained language BERT model with extreme gradient boosting

Aspect-based sentiment analysis (ABSA) is a granular-level sentiment analysis task that aims to detect the sentiment polarities of a specified aspect in the text. This research shows excessive curiosity in modelling target and context through attention networks to attain effective feature representa...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:The Journal of supercomputing Jg. 79; H. 5; S. 5516 - 5551
Hauptverfasser: Mewada, Arvind, Dewang, Rupesh Kumar
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.03.2023
Springer Nature B.V
Schlagworte:
Accuracy
Algorithms
Classification
Compilers
Computer Science
Context
Datasets
Deep learning
Interpreters
Language
Machine learning
Natural language processing
Neural networks
Processor Architectures
Programming Languages
Representations
Semantics
Sentiment analysis
Words (language)
XGBoost classifier
Attention networks
BERT encoder
Aspect-based sentiment analysis
Synthetic attention
Pre-training BERT
ISSN:0920-8542, 1573-0484
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Aspect-based sentiment analysis (ABSA) is a granular-level sentiment analysis task that aims to detect the sentiment polarities of a specified aspect in the text. This research shows excessive curiosity in modelling target and context through attention networks to attain effective feature representations for sentiment detection works. We have proposed a synthetic attention in bidirectional encoder representations from transformers (SA-BERT) with an extreme gradient boosting (XGBoost) classifier to classify sentiment polarity in the review dataset. The proposed model generates dynamic word vector encoding of the aspect and corresponding context of the reviews. Then, the aspect and context of the reviews are meaningfully represented by a transformer that can input the vector word in parallel. After that, the model uses the synthetic attention mechanism to learn essential parts of context and aspects in reviews. Finally, the model places overall representation in the sentiment classification layer to predict sentiment polarity. Both proposed SA-BERT and SA-BERT-XGBoost models achieved the highest accuracy (92.02 and 93.71%) on the restaurant16 and highest F-1 scores (81.19 and 81.64%) on the restaurant14 dataset, respectively. The average accuracy and F1 scores are approximately 2 and 3.04% higher than the baseline models (DLCF-DCA-CDM, R-GAT+BERT, ASGCN-DG, AEN-BERT and BERT-PT). Therefore, proposed models outperform in comparison with baseline models.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0920-8542
1573-0484
DOI:10.1007/s11227-022-04881-x