Table 1 Summary of related works in text classification using language modeling
From: Sentiment analysis of Indonesian datasets based on a hybrid deep-learning strategy
Publication | Technique applied | Claimed outcome | Limitation |
|---|---|---|---|
[6] | BERT and a hybrid model of CNN and attention-based Bi-GRU | The proposed technique attained the best scores of metrics in the experiment | Trained in a single dataset and English language |
[9] | A combination of feature extraction and classifier models in various dataset | Using SVM as a classifier and CNN & LSTM as feature extraction improves the performance result | Longer computation time and trained in English language |
[16] | Proposing various pre-trained models for word embedding and classifiers | RoBERTa and funnel transformers with CNNs as the classifier had excellent performance | Get a low accuracy on the LIAR dataset |
A fusion model based on BERT and LSTM-CNN for prediction | The proposed technique obtained the best result of metrics scores | Trained in a single dataset and English language | |
[19] | Glove vectors and a hybrid model of LSTM-CNN | A hybrid scheme to determine the emotional polarity of tweets with high accuracy | Trained in a single dataset and English language |
[20] | Using RoBERTa as text encoder and BiGRU & attention as feature extractor | A pre-training language model is effective for word vector extraction and the addition of attention can improve the accuracy | Get a low accuracy on the SST dataset |
[21] | Keras embedding and convolutional neural network | Achieving accuracy levels above 94% | Focused on short sentences with maximum words of 20 |