Scholarlink

<p>In real-world applications, particularly in platforms reliant on user-generated content like review websites and e-commerce platforms, evaluating the quality of webpage content is essential to ensure that users access reliable, relevant, and up-to-date information. Traditional classification methods face difficulties in determining the quality of a webpage, especially when dealing with new, unseen reviews that exhibit diverse patterns and contexts. To address these challenges, a novel Residual Convolutional Neural Networks and Drop Connect Long Short-Term Memory with Deep Deterministic Policy Gradient (RCNN-DCLSTM-DDPG) reinforcement learning is proposed. This model utilizes a Residual Convolutional Neural Networks (RCNN) for feature extraction, the Dilated CNN (D-CNN) captures long-range dependencies within review text by expanding the receptive field without increasing computational complexity. Additionally, the Residual Network (ResNet) architecture incorporates threshold-weighted mapping, enabling the network to focus on the most relevant features while eliminating unnecessary layers, which enhances classification accuracy. The Drop connect regularization is applied to the LSTM, randomly removing connections during training to reduce overfitting, enhancing the model's robustness. In the Deep Deterministic Policy Gradient (DDPG) framework, the actor network utilizes these feature representations from the RCNN-DCLSTM to predict the quality classifications of reviews, such as very high quality, high quality, moderate quality, low quality, or very low quality. The critic network incorporates the output from the RCNN-DCLSTM classifier to evaluate the accuracy of the actor's decisions, providing feedback in the form of a reward signal based on the accuracy of the predictions. This enables the system to adjust its parameters based on the feedback, improving the quality classification process. The proposed RCNN-DCLSTM-DDPG is tested on four datasets and compared to previous methods. Experimental results for dataset-4 show that the proposed technique achieves an accuracy of 99.3%, precision of 99.10%, recall of 98.20%, F-1 measure of 98%, latency of 28&#xa0;ms, and throughput of 128 ops/sec. These results highlight the model's ability to adapt to and accurately classify webpage quality, even for new and unseen reviews.</p>

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

Deep reinforcement Learning Based on Residual Convolutional Neural Networks and Drop Connect Long Short-Term Memory with Adaptive Feedback for Webpage Quality Classification

  • 期刊论文

  • 发表日期：

    2025年1月15日

查看全文
  • Atul Kumar Srivastava,
  • Dhiraj Pandey,
  • Alok Agarwal

摘要

In real-world applications, particularly in platforms reliant on user-generated content like review websites and e-commerce platforms, evaluating the quality of webpage content is essential to ensure that users access reliable, relevant, and up-to-date information. Traditional classification methods face difficulties in determining the quality of a webpage, especially when dealing with new, unseen reviews that exhibit diverse patterns and contexts. To address these challenges, a novel Residual Convolutional Neural Networks and Drop Connect Long Short-Term Memory with Deep Deterministic Policy Gradient (RCNN-DCLSTM-DDPG) reinforcement learning is proposed. This model utilizes a Residual Convolutional Neural Networks (RCNN) for feature extraction, the Dilated CNN (D-CNN) captures long-range dependencies within review text by expanding the receptive field without increasing computational complexity. Additionally, the Residual Network (ResNet) architecture incorporates threshold-weighted mapping, enabling the network to focus on the most relevant features while eliminating unnecessary layers, which enhances classification accuracy. The Drop connect regularization is applied to the LSTM, randomly removing connections during training to reduce overfitting, enhancing the model's robustness. In the Deep Deterministic Policy Gradient (DDPG) framework, the actor network utilizes these feature representations from the RCNN-DCLSTM to predict the quality classifications of reviews, such as very high quality, high quality, moderate quality, low quality, or very low quality. The critic network incorporates the output from the RCNN-DCLSTM classifier to evaluate the accuracy of the actor's decisions, providing feedback in the form of a reward signal based on the accuracy of the predictions. This enables the system to adjust its parameters based on the feedback, improving the quality classification process. The proposed RCNN-DCLSTM-DDPG is tested on four datasets and compared to previous methods. Experimental results for dataset-4 show that the proposed technique achieves an accuracy of 99.3%, precision of 99.10%, recall of 98.20%, F-1 measure of 98%, latency of 28 ms, and throughput of 128 ops/sec. These results highlight the model's ability to adapt to and accurately classify webpage quality, even for new and unseen reviews.