Scholarlink

In the realm of sentiment classification using deep learning models like Convolutional Neural Network (CNN), current research is striving to enhance the models’ ability to capture contextual relationships and long-range dependencies in textual data. While CNNs excel at extracting local features, their simplicity in network structure may hinder their capacity to incorporate nuanced contextual relationships necessary for sentiment analysis. This limitation underscores the importance of models that can comprehend and utilize contextual dependencies for more accurate sentiment analysis, highlighting the need for solutions beyond traditional CNN architectures. To address these limitations, specialized variants of Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) have been introduced to mitigate issues like gradient vanishing and explosion. The vanishing gradient problems can severely impact the training and performance of RNNs by hindering the network's ability to capture long-term dependencies and stabilize the training process. In response, LSTM and its bidirectional counterpart, BiLSTM, offer enhanced capabilities in capturing contextual information bidirectionally along the input sequence, aiding in more effective sentiment analysis in NLP tasks. To handle these problems, a hybrid CNN-BiLSTM model with an attention mechanism is proposed to enhance sentiment classification by combining the strengths of CNN and BiLSTM architectures with an added attention layer. The model leverages CNN for high-level feature extraction and local pattern capture, complemented by BiLSTM for contextual feature representation processing. The integration of an attention mechanism further enhances the model's ability to prioritize and interpret different parts of the input sequence, refining its sentiment analysis capabilities. By incorporating these advanced layers, the hybrid CNN-BiLSTM model aims to improve contextual features and sentiment classification performance in handling the intricacies of textual data, paving the way for more accurate and nuanced sentiment analysis outcomes. We evaluate the model on four benchmark datasets, and it outperformed previous models with state-of-the-art results.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Aspect-Based Sentiment Classification Using Enhanced CNN-BiLSTM Networks and Attention

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ousmane Sangary,
  • Umar Muhammad Mustapha kumshe

摘要

In the realm of sentiment classification using deep learning models like Convolutional Neural Network (CNN), current research is striving to enhance the models’ ability to capture contextual relationships and long-range dependencies in textual data. While CNNs excel at extracting local features, their simplicity in network structure may hinder their capacity to incorporate nuanced contextual relationships necessary for sentiment analysis. This limitation underscores the importance of models that can comprehend and utilize contextual dependencies for more accurate sentiment analysis, highlighting the need for solutions beyond traditional CNN architectures. To address these limitations, specialized variants of Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) have been introduced to mitigate issues like gradient vanishing and explosion. The vanishing gradient problems can severely impact the training and performance of RNNs by hindering the network's ability to capture long-term dependencies and stabilize the training process. In response, LSTM and its bidirectional counterpart, BiLSTM, offer enhanced capabilities in capturing contextual information bidirectionally along the input sequence, aiding in more effective sentiment analysis in NLP tasks. To handle these problems, a hybrid CNN-BiLSTM model with an attention mechanism is proposed to enhance sentiment classification by combining the strengths of CNN and BiLSTM architectures with an added attention layer. The model leverages CNN for high-level feature extraction and local pattern capture, complemented by BiLSTM for contextual feature representation processing. The integration of an attention mechanism further enhances the model's ability to prioritize and interpret different parts of the input sequence, refining its sentiment analysis capabilities. By incorporating these advanced layers, the hybrid CNN-BiLSTM model aims to improve contextual features and sentiment classification performance in handling the intricacies of textual data, paving the way for more accurate and nuanced sentiment analysis outcomes. We evaluate the model on four benchmark datasets, and it outperformed previous models with state-of-the-art results.