Deep Learning Based Weighted Feature Fusion Approach for Sentiment Analysis

Abstract

Deep learning algorithms have achieved remarkable results in the natural language processing(NLP) and computer vision. Hence, a trend still going on to use these algorithms, such as convolution and recurrent neural networks, for text analytic task to extract useful information. Features extraction is one of the important reasons behind the success of these networks. Moreover passing features from one layer to another layer within the network and one network to another network have done. However multilevel and multitype features fusion remains unexplored in sentiment analysis. So, in this paper, we use three datasets to display the advantages of extracting and fusing multilevel as well as multitype features from different neural networks. Multilevel features are from different layers of the same network, and multitype features are from different network architectures. Experiment results demonstrate that the proposed model based on multilevel and multitype weighted features fusion outperforms than many exiting works with an accuracy of 80.2%, 48.2%, and 87.0% on MR, SST1, and SST2 datasets respectively.INDEX TERMS Convolution neural network, recurrent neural network, feature fusion, feature learning, and sentiment analysis.

Sentiment Classification Using a Single-Layered BiLSTM Model

“…In summary, our single-layered BiLSTM based model offered significant results on MR dataset, and are competitive with recently published studies [25], [35]- [37], as depicted in Table 4.…”

“…It is important to mention that although [42] slightly outperformed our model but its underlying architecture is relatively more complex compared to our proposed system. In summary, it is inferred that our single-layered BiLSTM based system achieved satisfactory results on SST2 dataset but could not outperform a few recently proposed models with complex architectures [37], [42], as given in Table 6.…”

“…Opposite to BiGRU+CNN, our proposed structure depends on BiLSTM only and still provided 1.2 percentage points higher accuracy than [36]. Recently, Usama et al [37] investigated the parallel fusion of CNN and LSTM models along with the multitype selection of features and obtained an accuracy of 85.70% on SST2 dataset. In contrast, we employed a single BiLSTM layer with few nodes and achieved the same accuracy.…”

“…Contrary to BiGRU+CNN, our model carries solely a BiLSTM network and provided 1.2 percentage points higher accuracy than [36]. Finally, Usama et al [37] presented various models by encountering multilevel and multitype fusion of features along with the combination of LSTM, GRU, and CNN models. Amongst all, the CNN-GRU-multilevel & multitype fusion and CNN-GRUmultilevel & multitype fusion models showed the best results with the accuracies of 79.80% and 80.20%, respectively.…”

Hybrid CNN Classification for Sentiment Analysis under Deep Learning

“…In this, multilevel features were from different layers of the same network, and multitype features were from different network architectures. It was found from the research that the model based on multilevel and multitype weighted features fusion out performed many existing works with a greater accuracy [24]. In another research, a network architecture was introduced to analyze sentences meaning through character-level representations by using a combination of long short-term memory (LSTM), CNN and conditional random field (CRF) [25].…”