Scholarlink

<p>Sentiment analysis has become more well known in the last few years since it is crucial for regulating and examining the data that is shared online. Natural language processing is beginning to focus on sentiment analysis in regional and low-resource languages. Studies on sentiment analysis in Indian languages, including Telugu, Tamil, Malayalam, Kannada, etc., have received increasing attention from researchers. To the best of our knowledge, the absence of an annotated dataset has prevented any reports of microscopic study on Indian languages up until this point. However, the work of sentiment analysis becomes difficult for low-resource languages because annotated datasets, which are the foundation for developing sentiment classifiers, are few for texts written in languages other than English. This research proposes a new hybrid deep learning-based technique to classify the sentiments. Initially, we clean the text by removing punctuation, numbers, and special characters, converting all text to lower case, and removing stop words which are the common words that do not carry significant meaning, like “and”, “the”, etc. and tokenizing the text as splitting the text into individual words or tokens. Then, perform stemming and lemmatization to reduce words to their base or root form and spelling variations. Then, perform word embedding using the robustly optimized BERT approach (RoBERTa) method to word embedding to represent words in a continuous vector space. After that, extract the relevant features using the self-attention-based Transformer (SAT-Net) technique. Finally, employs the novel hybrid deep learning-based techniques of deep bidirectional long short-term memory (Deep BiLSTM) and inception recurrent residual convolutional neural network (IRRCNN) to classify the sentiments into their respective category. Using improved dung beetle optimization algorithm (IDBOA) to enhance model performance by fine-tuning hyper-parameters. We simulate the effectiveness of the proposed model on a multilingual dataset that combines several languages such as Tamil, Telugu, Kannada, and Malayalam with recall, accuracy, precision, and F1-score evaluation metrics.</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

Enhancing sentiment analysis performance: a multilingual approach with advanced text processing and hybrid deep learning techniques with improved dung beetle optimization algorithm

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Durga Satish Matta,
  • Saruladha Krishnamurthy

摘要

Sentiment analysis has become more well known in the last few years since it is crucial for regulating and examining the data that is shared online. Natural language processing is beginning to focus on sentiment analysis in regional and low-resource languages. Studies on sentiment analysis in Indian languages, including Telugu, Tamil, Malayalam, Kannada, etc., have received increasing attention from researchers. To the best of our knowledge, the absence of an annotated dataset has prevented any reports of microscopic study on Indian languages up until this point. However, the work of sentiment analysis becomes difficult for low-resource languages because annotated datasets, which are the foundation for developing sentiment classifiers, are few for texts written in languages other than English. This research proposes a new hybrid deep learning-based technique to classify the sentiments. Initially, we clean the text by removing punctuation, numbers, and special characters, converting all text to lower case, and removing stop words which are the common words that do not carry significant meaning, like “and”, “the”, etc. and tokenizing the text as splitting the text into individual words or tokens. Then, perform stemming and lemmatization to reduce words to their base or root form and spelling variations. Then, perform word embedding using the robustly optimized BERT approach (RoBERTa) method to word embedding to represent words in a continuous vector space. After that, extract the relevant features using the self-attention-based Transformer (SAT-Net) technique. Finally, employs the novel hybrid deep learning-based techniques of deep bidirectional long short-term memory (Deep BiLSTM) and inception recurrent residual convolutional neural network (IRRCNN) to classify the sentiments into their respective category. Using improved dung beetle optimization algorithm (IDBOA) to enhance model performance by fine-tuning hyper-parameters. We simulate the effectiveness of the proposed model on a multilingual dataset that combines several languages such as Tamil, Telugu, Kannada, and Malayalam with recall, accuracy, precision, and F1-score evaluation metrics.