Scholarlink

<p>The integration of pre-trained embedding methods with deep learning models for sentiment analysis often achieves high accuracy but at the cost of extended training times, which can hinder decision-making in real-time applications. To address this challenge, we propose ELECTRA-Fed-CNN, a novel sentiment analysis framework combining federated learning and distributed embeddings to achieve both high accuracy and efficiency. The pipeline comprises pre-processing review data, employing the ELECTRA embedding model on distributed GPUs, and training a CNN model using federated learning to promote better generalization by leveraging the diversity of data across clients. Hyperparameter optimization is employed to fine-tune the model overall performance. Our approach was validated on a French insurance reviews dataset, demonstrating its ability to balance performance and computational efficiency, while outperforming state-of-the-art sentiment analysis methods.</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

Optimizing sentiment analysis of insurance services’ reviews on social media: Navigating the balance between training time and accuracy using deep learning and federated learning

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Sabrine Ben Abdrabbah,
  • Priya Asse

摘要

The integration of pre-trained embedding methods with deep learning models for sentiment analysis often achieves high accuracy but at the cost of extended training times, which can hinder decision-making in real-time applications. To address this challenge, we propose ELECTRA-Fed-CNN, a novel sentiment analysis framework combining federated learning and distributed embeddings to achieve both high accuracy and efficiency. The pipeline comprises pre-processing review data, employing the ELECTRA embedding model on distributed GPUs, and training a CNN model using federated learning to promote better generalization by leveraging the diversity of data across clients. Hyperparameter optimization is employed to fine-tune the model overall performance. Our approach was validated on a French insurance reviews dataset, demonstrating its ability to balance performance and computational efficiency, while outperforming state-of-the-art sentiment analysis methods.