A novel framework for sentiment classification employing Bi-GRU optimized by enhanced human evolutionary optimization algorithm

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 16, 2025

Sentiment analysis of content is highly essential for myriad natural language processing tasks. Particularly, as the movies are often created on basis public opinions, reviews people have gained much attention, and analyzing sentiments has also become a crucial demanding task. The unique characteristics this data, such length text, spelling mistakes, abbreviations, necessitate non-conventional method additional stages sentiment in an environment. To do so, paper conducted two different word embedding models, namely GloVe Word2Vec, vectorization. In study, Bidirectional Gated Recurrent Unit was employed, since there were polarities, including positive negative. Then, it optimized by Enhanced Human Evolutionary Optimization (EHEO) algorithm, hence improving hyperparameters. findings showed that using GloVe, Bi-GRU/EHEO model achieved 97.26% precision, 96.37% recall, 97.42% accuracy, 96.30% F1-score. With suggested attained 98.54% 97.75% 97.54% 97.63% These compared with other models like GRU accomplished F1-score values 89.24, 90.14, 89.57, 89.68 Glove well 89.67, 90.18, 90.75, 89.41 Word2Vec; Bi-GRU 90.13, 90.47, 90.71, 90.30 Glove, 90.31, 90.76, 90.67, 90.53 Word2Vec. approaches demonstrated potential to be used real-world applications, customer feedback evaluation, political opinion analysis, social media analysis. By these models' high efficiency could offered some practical solutions diverse industries forecast trends, enhance decision-making procedures, examine textual data.

Language: Английский

---

EFFECTS OF STRATIFIED CROSS-VALIDATION AND HYPERPARAMETER TUNING ON SENTIMENT CLASSIFICATION WITH THE CHI2-RFE HYBRID FEATURE SELECTION TECHNIQUE IN THE IMDB DATASET

ShodhKosh Journal of Visual and Performing Arts, Journal Year: 2024, Volume and Issue: 5(5)

Published: May 31, 2024

Data analysis from social networking sites provides government entities, businesses, and event planners with insights into public sentiments perceptions. Sentiment (SA) resolves this need by classifying the sentiment of network users multiple classes. Despite their usefulness, data platforms frequently exhibits challenges, including unstructured formats, high volume, redundant or irrelevant information, which can cause issues like overfitting, underfitting, curse dimensionality. In response to these study proposes using term frequency-inverse document frequency (TF-IDF) for feature extraction along a hybrid selection method that combines Chi2 recursive elimination (RFE), called Chi2-RFE. This approach seeks identify optimal subset filtering out features. The proposed is tested several classifiers, KNN, LR, SVC, GNB, DT, RFC, employing stratified K-fold cross-validation hyperparameter tuning on an IMDb dataset obtained Kaggle. By effectively addressing overfitting underfitting issues, shows before StratefiedKfold tuning, LR gives 0.81975 training accuracy test 0.815 data. After mentioned above, removed enhancing 0.864833 KNN also enhanced its 0.891667 0.857333. SVC 0.846666 0.883667, GNB 0.809666 0.829583. Precision improved 0.826 0.853 0.848 0.897 0.852 0.868 0.799 GNB. Recall improvement 0.600 0.857 0.894 0.847 0.873 0.810 F1-score increased 0.764 0.843 0.883 0.819 0.862 0.790

Language: Английский

---

On the Utilization of Emoji Encoding and Data Preprocessing with a Combined CNN-LSTM Framework for Arabic Sentiment Analysis

Modelling—International Open Access Journal of Modelling in Engineering Science, Journal Year: 2024, Volume and Issue: 5(4), P. 1469 - 1489

Published: Oct. 7, 2024

Social media users often express their emotions through text in posts and tweets, these can be used for sentiment analysis, identifying as positive or negative. Sentiment analysis is critical different fields such politics, tourism, e-commerce, education, health. However, approaches that perform well on English encounter challenges with Arabic due to its morphological complexity. Effective data preprocessing machine learning techniques are essential overcome provide insightful predictions text. This paper evaluates a combined CNN-LSTM framework emoji encoding Analysis, using the Twitter Corpus (ASTC) dataset. Three experiments were conducted eight-parameter fusion evaluate effect of preprocessing, namely real emotional meaning. Emoji meanings collected from four websites specialized finding meaning emojis social media. Furthermore, Keras tuner optimized parameters during 5-fold cross-validation process. The highest accuracy rate (91.85%) was achieved by keeping non-Arabic words removing punctuation, Snowball stemmer after into text, applying embedding. approach competitive other state-of-the-art approaches, showing enriches accurately reflecting emotions, enabling investigation allowing hybrid model achieve comparable results study same ASTC dataset, thereby improving accuracy.

Language: Английский

---

Deep Learning in Written Arabic Linguistic Studies: A Comprehensive Survey

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 172196 - 172233

Published: Jan. 1, 2024

Language: Английский

---

Optimized Global Aware Siamese Network based Monkeypox disease classification using skin images

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 101, P. 107125 - 107125

Published: Nov. 18, 2024

Language: Английский

---