Scholarlink

<p>The paper proposes an aspect extraction model based on deep learning for opinion mining, an important task for sentiment analysis. Aspect extraction is concerned with determining specific product or service features that are praised or critiqued in opinionated texts. In our method, each word within a sentence is labeled as aspect-based or non-aspect-based by employing a seven-layer convolutional neural network. Performance is improved by creating linguistic patterns (LP) and combining them with the neural network. The designed model vastly surpasses current approaches by combining both linguistic heuristics and deep learning into an ensemble classification method. Experiments on four datasets that serve as gold standards, MP3, DVD, Canon, and Nikon showed significant improvements in accuracy, reaching 89.91%, 88.74%, 89.44%, and 83.99% with linguistic patterns, as opposed to 86.21%, 85.14%, 86.71%, and 79.91% in its absence. The results validate that our hybrid method is effective in boosting aspect-level sentiment analysis.</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

A Novel Model Utilizing Deep Neural Networks for Opinion Mining

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月1日

查看全文
  • Jagendra Singh,
  • Syam Machinathu Parambil Gangadharan,
  • Prabhishek Singh,
  • Manoj Diwakar,
  • Shachi Mall,
  • Anchit Bijalwan,
  • Sandeep Kumar

摘要

The paper proposes an aspect extraction model based on deep learning for opinion mining, an important task for sentiment analysis. Aspect extraction is concerned with determining specific product or service features that are praised or critiqued in opinionated texts. In our method, each word within a sentence is labeled as aspect-based or non-aspect-based by employing a seven-layer convolutional neural network. Performance is improved by creating linguistic patterns (LP) and combining them with the neural network. The designed model vastly surpasses current approaches by combining both linguistic heuristics and deep learning into an ensemble classification method. Experiments on four datasets that serve as gold standards, MP3, DVD, Canon, and Nikon showed significant improvements in accuracy, reaching 89.91%, 88.74%, 89.44%, and 83.99% with linguistic patterns, as opposed to 86.21%, 85.14%, 86.71%, and 79.91% in its absence. The results validate that our hybrid method is effective in boosting aspect-level sentiment analysis.