Scholarlink

<p>Sentiment classification is a discipline of artificial intelligence that classifies evaluations as positive, neutral, or negative based on the emotional opinion detected through natural language processing. A key challenge in sentiment classification is high dimensionality, as it often leads to reduced model performance due to redundant or irrelevant features. Removing such features, a common practice in machine learning and data processing, helps create an optimal feature space by reducing the number of input variables. This study proposes a novel multi-view ensemble learning model that integrates ElasticNetCV-based feature selection across multiple text representations—namely, Term Frequency-Inverse Document Frequency (TF-IDF), DistillBERT, and Global Vectors for Word Representation (GloVe). To our knowledge, this is the first systematic evaluation of ElasticNetCV applied to both traditional and contextual embeddings within a unified model. The selected features are used to train a multi-view soft voting ensemble model incorporating Random Forest (RF), Logistic Regression (LR), and Artificial Neural Networks (ANN) as base classifiers. Experiments on Yelp, Amazon, and IMDB datasets demonstrate notable F1-scores of 97.1%, 91.2%, and 95.4%, respectively, highlighting the efficiency of the proposed approach.</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

An Optimized Feature Selection Approach for Multi-View Ensemble Learning in Sentiment Analysis of User Reviews

  • 期刊论文

  • 发表日期：

    2025年9月5日

查看全文
  • Fatih Demirci,
  • Zeynep Garip,
  • Ekin Ekinci

摘要

Sentiment classification is a discipline of artificial intelligence that classifies evaluations as positive, neutral, or negative based on the emotional opinion detected through natural language processing. A key challenge in sentiment classification is high dimensionality, as it often leads to reduced model performance due to redundant or irrelevant features. Removing such features, a common practice in machine learning and data processing, helps create an optimal feature space by reducing the number of input variables. This study proposes a novel multi-view ensemble learning model that integrates ElasticNetCV-based feature selection across multiple text representations—namely, Term Frequency-Inverse Document Frequency (TF-IDF), DistillBERT, and Global Vectors for Word Representation (GloVe). To our knowledge, this is the first systematic evaluation of ElasticNetCV applied to both traditional and contextual embeddings within a unified model. The selected features are used to train a multi-view soft voting ensemble model incorporating Random Forest (RF), Logistic Regression (LR), and Artificial Neural Networks (ANN) as base classifiers. Experiments on Yelp, Amazon, and IMDB datasets demonstrate notable F1-scores of 97.1%, 91.2%, and 95.4%, respectively, highlighting the efficiency of the proposed approach.