<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>This study develops a sentiment analysis model for English academic discourse based on word information to effectively understand and analyze the sentiment tendencies in English literary texts. The structure of the model includes word embedding layer, character-level feature extraction, word-level feature extraction and feature fusion and classification layer. The word embedding layer realizes the mapping between word vectors and word vectors by microblogging pre-trained word vectors. The character-level feature extraction session uses a multi-window convolutional layer to capture N-Gram information. In contrast, the word-level feature extraction obtains deeper semantic information through a Bi-LSTM layer and fuses it with character-level information to enhance robustness. The feature fusion and classification layer further combines these features and determines the fusion weights through a linear layer to achieve sentiment classification. In performance tests, the model achieves 92.5% sentiment classification accuracy on the standard dataset, an improvement of about 6% compared to traditional methods. In particular, the accuracy is improved by 5% when dealing with text with sentiment polarity transition, showing good adaptability. In addition, using 657 positive and 679 negative sentiment words as seed words effectively expands the sentiment lexicon and enhances the comprehensiveness and accuracy of sentiment analysis.</p>
</abstract>

This study develops a sentiment analysis model for English academic discourse based on word information to effectively understand and analyze the sentiment tendencies in English literary texts. The structure of the model includes word embedding layer, character-level feature extraction, word-level feature extraction and feature fusion and classification layer. The word embedding layer realizes the mapping between word vectors and word vectors by microblogging pre-trained word vectors. The character-level feature extraction session uses a multi-window convolutional layer to capture N-Gram information. In contrast, the word-level feature extraction obtains deeper semantic information through a Bi-LSTM layer and fuses it with character-level information to enhance robustness. The feature fusion and classification layer further combines these features and determines the fusion weights through a linear layer to achieve sentiment classification. In performance tests, the model achieves 92.5% sentiment classification accuracy on the standard dataset, an improvement of about 6% compared to traditional methods. In particular, the accuracy is improved by 5% when dealing with text with sentiment polarity transition, showing good adaptability. In addition, using 657 positive and 679 negative sentiment words as seed words effectively expands the sentiment lexicon and enhances the comprehensiveness and accuracy of sentiment analysis.

Exploring the Subjectivity of English Academic Discourse in the Context of Big Data

Harbin University of Science and Technology

Applied Mathematics and Nonlinear Sciences

This work is licensed under the Creative Commons Attribution 4.0 International License.

This study develops a sentiment analysis model for English academic discourse based on word information to effectively understand and analyze the sentiment...

Exploring the Subjectivity of English Academic Discourse in the Context of Big Data

Published Online: Feb 26, 2024

Page range: -

Received: Jan 10, 2024

Accepted: Jan 19, 2024

DOI: https://doi.org/10.2478/amns-2024-0489

Keywords
N-gram, Bi-LSTM, Sentiment lexicon, Word vectors, Sentiment levelness

© 2024 Ying Pan, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Exploring the Subjectivity of English Academic Discourse in the Context of Big Data

Published Online: Feb 26, 2024

Page range: -

Received: Jan 10, 2024

Accepted: Jan 19, 2024

DOI: https://doi.org/10.2478/amns-2024-0489

KeywordsN-gram, Bi-LSTM, Sentiment lexicon, Word vectors, Sentiment levelness

© 2024 Ying Pan, published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Keywords
N-gram, Bi-LSTM, Sentiment lexicon, Word vectors, Sentiment levelness