Scholarlink

<p>Emoji-based sentiment analysis (SA) plays a major role in improving the accuracy of sentiment classification in text content. However, the performance of traditional approaches is affected by overfitting due to poor generalization on unseen data. The existing deep learning (DL) approaches with a large number of parameters make the SA task complicated. To address these issues, a novel hybrid approach combining modified term frequency (TF)- inverse document frequency (IDF) with arithmetic Gaussian kernel extreme learning machine (AGKELM) is developed. The proposed model incorporates pre-trained emoji embeddings and modified feature extraction for efficiently representing emoji-based characteristics with fewer parameters. These features improve the accuracy with less computational burden. Moreover, the hyperparameters of the model are adjusted using arithmetic Gaussian optimization (AGO), which minimizes the effect of overfitting. Initially, data are pre-processed, and data labeling is performed to interpret text data with sentiment labels. In order to convert text data into numeric feature vectors, various techniques of feature vectorization, such as count vectorization<i>,</i> TF-IDF and modified TF-IDF<i>,</i> Word2Vec models, including continuous Bag of Words (CBOW) and Skip-gram (SG), were utilized. Finally, to classify the text and emoji-based SA, several machine learning (ML) methods, include logistic regression (LR)<i>,</i> random forest (RF)<i>,</i> K-nearest neighbors (KNNs) classifier, support vector machine (SVM), and AGKELM, are employed. The proposed model’s performance is assessed by the Flipkart Product reviews dataset, Facebook dataset, Twitter dataset, and multi-lingual dataset. Moreover, the proposed model is compared with various ML and advanced DL approaches. In processing sparse emoji-based text data, the best performance is noted with modified TF-IDF and AGKELM. The experimental results proved that the proposed model achieved higher accuracy, precision, recall, and f1-score of 95.65%, 93.34%, 92.8%, and 93%, respectively. Therefore, the proposed approach improves the adaptability and classification accuracy with lower computational cost.</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 hybrid approach for sentiment analysis: combining modified feature extraction with an arithmetic Gaussian kernel extreme learning machine

  • 期刊论文

  • 发表日期：

    2026年2月24日

查看全文
  • Padigapati Anitha,
  • A. V. Praveen Krishna

摘要

Emoji-based sentiment analysis (SA) plays a major role in improving the accuracy of sentiment classification in text content. However, the performance of traditional approaches is affected by overfitting due to poor generalization on unseen data. The existing deep learning (DL) approaches with a large number of parameters make the SA task complicated. To address these issues, a novel hybrid approach combining modified term frequency (TF)- inverse document frequency (IDF) with arithmetic Gaussian kernel extreme learning machine (AGKELM) is developed. The proposed model incorporates pre-trained emoji embeddings and modified feature extraction for efficiently representing emoji-based characteristics with fewer parameters. These features improve the accuracy with less computational burden. Moreover, the hyperparameters of the model are adjusted using arithmetic Gaussian optimization (AGO), which minimizes the effect of overfitting. Initially, data are pre-processed, and data labeling is performed to interpret text data with sentiment labels. In order to convert text data into numeric feature vectors, various techniques of feature vectorization, such as count vectorization, TF-IDF and modified TF-IDF, Word2Vec models, including continuous Bag of Words (CBOW) and Skip-gram (SG), were utilized. Finally, to classify the text and emoji-based SA, several machine learning (ML) methods, include logistic regression (LR), random forest (RF), K-nearest neighbors (KNNs) classifier, support vector machine (SVM), and AGKELM, are employed. The proposed model’s performance is assessed by the Flipkart Product reviews dataset, Facebook dataset, Twitter dataset, and multi-lingual dataset. Moreover, the proposed model is compared with various ML and advanced DL approaches. In processing sparse emoji-based text data, the best performance is noted with modified TF-IDF and AGKELM. The experimental results proved that the proposed model achieved higher accuracy, precision, recall, and f1-score of 95.65%, 93.34%, 92.8%, and 93%, respectively. Therefore, the proposed approach improves the adaptability and classification accuracy with lower computational cost.