Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 573 https://internationalpubls.com Nonlinear Deep Learning Framework for Precise Sentiment Classification in Textual Data 1DR.Attili Venkata Ramana, 2Dr. Kalli Srinivasa Nageswara Prasad3, Annaluri Sreenivasa Rao 1Associate Professor, Department of CSE (AIML), Geethanjali College of Engineering and Technology, Cheeryala, Kesara, Hyderabad-501301. Email Id: avrrdg@gmail.com 2Department of Computer Science and Engineering, Sasi Institute of Technology and Engineering, Tadepalligudem. A.P. Orcid Id : 0000-0002-8083-3672. Email Id: kallisnprasad@gmail.com 3Department of Information Technology,VNR Vignana Jyothi Institute of Engineering and Technology, Hyderabad, Orchid id 0000-0003-1618-672X. Email: annaluri.rao@gmail.com Article History: Received: 01-09-2024 Revised: 16-10-2024 Accepted: 01-11-2024 Abstract: The study presents the Nonlinear Deep Learning Framework for Precise Sentiment Classification (NPSC), focusing on improving sentiment analysis in text data. The model addresses the challenge of capturing complex and hidden sentiment patterns using a combination of BERT embeddings, Bidirectional LSTM (BiLSTM), Convolutional Neural Networks (CNN), and an attention mechanism. This setup helps in identifying subtle relationships within text, enabling more precise sentiment predictions. NPSC tackles common issues faced by existing models that struggle with detailed text analysis. Using the IMDB movie review dataset, the approach relies on BERT to convert words into dense vectors, BiLSTM for capturing context from both directions, and CNN to extract key local features. An attention layer highlights important words, refining the sentiment detection process. The integration of these components in a nonlinear fusion layer allows for enhanced feature interaction, leading to accurate classification through a final softmax layer. The findings show that NPSC performs better across metrics like accuracy, precision, recall, and F1-score. Ablation studies reveal the impact of each component, demonstrating their role in enhancing the model’s effectiveness. Error analysis shows how NPSC manages mixed sentiments and challenging text patterns. The study highlights NPSC’s potential as a reliable method for sentiment analysis, contributing to better handling of complex textual data. Keywords: Nonlinear Framework, Sentiment Classification, Text Analysis, BERT, BiLSTM, CNN, Attention Mechanism, NPSC Introduction Sentiment analysis of text is important in understanding opinions and emotions in various fields like social media, customer feedback, and market research. Traditional methods using rule-based or machine learning models often fail to capture the deeper meaning of language, such as sarcasm, context changes, or hidden sentiments. This makes the sentiment analysis task complex and less accurate. Advanced deep learning models, such as LSTM, GRU, and BERT, have been used to address these issues, showing better performance in analyzing and classifying sentiments in text [1]. However, these models still face challenges, especially when dealing with domain-specific language, diverse data, and multi-emotion expressions in a single text [2], [3]. mailto:avrrdg@gmail.com mailto:kallisnprasad@gmail.com mailto:annaluri.rao@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 574 https://internationalpubls.com Many studies focus on neural networks, but they often overlook how integrating these models with traditional lexical approaches can enhance sentiment analysis. Combining neural and lexical methods can capture subtle expressions and improve context understanding, yet this combination is rarely explored deeply [4]. Dependency-based techniques that extract features from grammatical structures offer additional insights but need further refinement for better accuracy [5]. The lack of such integrated, comprehensive frameworks limits the effectiveness of current sentiment analysis models, especially when applied to complex real-world data. Current neural network models excel in processing sequential data, but they struggle with capturing deeper contextual relationships, especially when sentiment is subtle or hidden in complex sentence structures. While there are efforts to integrate neural models with traditional methods, they often lack the depth and flexibility needed to adapt to various domains. Emotion recognition and multi-label classification, which require models to identify multiple sentiments in a single text, are also challenging areas where existing methods fall short [6][7]. Addressing these gaps is essential for creating more effective sentiment analysis tools that work well across different applications. The motivation for this research lies in developing a nonlinear deep learning framework that integrates different neural models and lexical approaches to improve sentiment classification. The aim is to create a more adaptable and accurate framework that addresses the current limitations by enhancing how sentiment and emotions are captured and analyzed in text. This research introduces a framework that combines neural network models like Bi-LSTM and GRU with traditional lexical methods to improve sentiment analysis. The key objectives are: • To enhance sentiment analysis by integrating neural and lexical methods, capturing subtle and complex expressions in text [4]. • To improve emotion recognition and handle multi-label classification where multiple sentiments are expressed together [6][8]. • To create a flexible framework that can be adapted to various domains and datasets by incorporating domain-specific lexicons and embeddings [4]. This study presents a new approach to sentiment analysis, combining neural models and lexical techniques to overcome common challenges. The proposed framework improves accuracy in identifying complex and nuanced sentiments in text. This makes it useful for practical applications in fields where understanding public opinion and emotional expression is important. The research contributes a flexible tool that can be tailored to specific industries, enhancing the value of sentiment analysis in real-world scenarios [9][10]. The paper is organized as follows: Section 2 reviews related work and discusses key deep learning architectures used in sentiment analysis. Section 3 details the proposed nonlinear framework, explaining the integration of neural and lexical methods. Section 4 covers the experiments and evaluation of the framework's performance. Section 5 discusses the results, highlighting how the framework improves sentiment classification. Section 6 concludes with the main findings and suggests future directions for research. This layout aims to clearly present the research's goals, methods, and contributions. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 575 https://internationalpubls.com 1 Related Work Recent work in sentiment analysis focuses on deep learning models to better understand and classify text data. Different methods have been tried to improve accuracy, handle various text types, and merge several techniques. Research can be grouped into key areas based on the methods used and what each study brings to the field. Some studies mix neural network models to improve how sentiment is understood. Adilakshmi et al. [11] used a method called RMDL, combining several neural networks and transfer learning to improve classification in context-rich text. Rose et al. [4] explored hybrid models, mixing traditional sentiment analysis methods with deep learning to handle complex emotions in text. Sachin Sambhaji Patil et al. [12] used CNNs to capture the relationships between words and added advanced pooling techniques to boost accuracy. These approaches highlight how mixing different models helps understand text from multiple viewpoints. Multi-source and transfer learning approaches have been used to merge data from different places, improving how text is classified. Nguyen et al. [13], Suganya, V., et al., [14] introduced LIFA, a framework that uses transfer learning to pull together information from various pretrained models, boosting sentiment analysis across different fields. This technique shows the flexibility of transfer learning in refining models, though it also needs quality data to work well. Adilakshmi et al. [11] also showed how transfer learning could optimize neural networks, cutting down training time and adapting models to new data. Handling multiple emotions in a single text is challenging. Priyanka et al. [15] used neural networks to classify emotions and recognize complex expressions in text. Radha et al. [7] developed deep learning models to identify multiple emotions using advanced embeddings, showing promise in managing different emotional layers within the same text. Chen et al. [16] introduced contrastive learning to refine how models distinguish subtle emotional shifts, improving performance in scenarios where data is limited. Sentiment analysis often requires adapting models to specific language challenges. Jadon et al. [17] developed hybrid models to handle mixed languages like Hinglish, capturing the unique patterns of such text. Gamal et al. [18] focused on Arabic language processing, using neural networks to manage dialects and informal styles, demonstrating that tailored models can handle diverse language needs. Ensemble and active learning methods bring together different models to create more robust systems. Garg et al. [19] combined multiple neural networks in an ensemble approach, boosting classification by integrating various perspectives of the text. Raja et al. [20] used active learning strategies to selectively train models on the most informative data points, enhancing accuracy, especially in unbalanced datasets. These methods show how combining and refining learning techniques can lead to better sentiment classification results. Some studies explore ways to improve sentiment classification by focusing on the relationships between words and refining feature extraction. Liu et al. [5], and Kiran, K., Abhishek Appaji et al., [21] proposed a dependency-based model that combines grammar rules with neural networks to capture more detailed meanings in complex sentences. Chen et al. [16] used contrastive learning Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 576 https://internationalpubls.com techniques to sharpen how models differentiate subtle sentiment changes, which is particularly useful when working with less data. Social media presents unique challenges due to its fast-changing and noisy nature. Rangarjan et al. [22] developed frameworks integrating neural networks to capture real-time sentiment trends, providing useful insights for applications like tracking public opinion. Kathiravan et al. [6] and Hernández, Nayeli et al., [23] focused on deep learning methods like LSTM and GRU to analyze large volumes of social media text, capturing hidden emotional cues and improving classification efficiency. The research shows that combining various deep learning and traditional approaches improves sentiment analysis in different contexts. By using hybrid models, transfer learning, and techniques like contrastive learning, sentiment classification becomes more precise and adaptable. However, balancing accuracy with the need for simpler, faster models remains a challenge, highlighting the ongoing need to refine these methods for broader, real-world use. 2 Methods and Materials The proposed nonlinear sentiment analysis model architecture is designed to capture complex dependencies in textual data through a sequence of specialized layers. Initially, raw text undergoes preprocessing and embedding using BERT, transforming words into dense, context-rich vectors. These vectors feed into a Bidirectional LSTM, which captures sequential context from both past and future words, creating comprehensive hidden states for each word. These hidden states are further refined by a Convolutional Neural Network (CNN) that identifies crucial local patterns and phrases, emphasizing sentiment-carrying elements like adjectives and adverbs through convolution and pooling. The attention mechanism then dynamically adjusts focus on sentiment-critical words, enhancing their influence in the analysis. A dense nonlinear feature fusion layer integrates the outputs from BiLSTM, CNN, and attention, applying ReLU activation to capture intricate feature interdependencies while dropout regularization prevents overfitting. Finally, the sentiment prediction layer uses a fully connected network followed by a softmax or sigmoid activation to produce the sentiment classification with high precision, translating complex text patterns into clear, confident sentiment scores. This architecture excels in extracting and leveraging subtle, non-linear relationships within text for highly accurate sentiment predictions. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 577 https://internationalpubls.com Figure 1: Architecture of Nonlinear Sentiment Analysis Model The diagram in figure 1 illustrates the architecture of a nonlinear sentiment analysis model, showing the sequential processing through layers from input to prediction. Components within each layer, such as the Bidirectional LSTM, CNN, Attention, and Dense layers, are arranged left-to-right, highlighting key data transformations, conditions, and iterations that refine text inputs into accurate sentiment predictions. This architecture integrates advanced layers designed to capture complex, non-linear relationships between textual features and sentiments. Each component is optimized for its specific role in the overall architecture. 2.1 Input Layer: Text Preprocessing and Embedding The input layer transforms raw text into numerical vectors that represent semantic meanings, serving as the foundation for further analysis. The embedding layer converts words into dense vectors that capture semantic and syntactic nuances using pre-trained models such as BERT (Bidirectional Encoder Representations from Transformers). BERT embeddings provide contextual word representations, accounting for the word’s role within sentences. Let 𝑇 = [𝑤1, 𝑤2, . . . , 𝑤𝑛] be the sequence of words in the input text. The embedding layer maps each word 𝑤𝑖 to a high-dimensional vector 𝑒𝑖 ∈ ℝ𝑑, where 𝑑 is the embedding dimension: Eq 1 𝐸 = 𝑓(𝑇) = [𝑒1, 𝑒2, . . . , 𝑒𝑛] ....(Eq 1) where 𝑓 represents the embedding function provided by BERT. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 578 https://internationalpubls.com 2.2 Feature Extraction Layer: Bidirectional LSTM This layer captures long-term dependencies in both forward and backward directions, providing a deep understanding of the context surrounding each word. A Bidirectional LSTM (Long Short-Term Memory) network is used to model sequential data, effectively managing long-range dependencies in text. This approach captures context from both the past and future, enhancing sentiment prediction by considering the full sentence structure. Given input sequence 𝐸 = [𝑒1, 𝑒2, . . . , 𝑒𝑛], the BiLSTM generates hidden states 𝐻 = [ℎ1, ℎ2, . . . , ℎ𝑛] where each ℎ𝑖 is the concatenation of forward and backward LSTM outputs: Eq 2 ℎ𝑖 = 𝐿𝑆𝑇𝑀 ⃖ (𝑒1, . . . , 𝑒𝑖) ⊕ 𝐿𝑆𝑇𝑀 (𝑒𝑛, . . . , 𝑒𝑖) ....(Eq 2) 2.3 Feature Extraction Layer: Convolutional Neural Network (CNN) CNN layers identify key phrases and local features crucial for sentiment analysis by scanning through the input with filters. Multiple convolutional filters of varying sizes (e.g., 2, 3, and 4) slide over the input sequence, capturing local dependencies and significant patterns that may indicate sentiment. Max pooling follows to retain the most prominent features, compressing the representation. For a filter 𝑊 of size 𝑘, the convolution operation generates feature map 𝑐𝑖 as: Eq 3 𝑐𝑖 = 𝑅𝑒𝐿𝑈(𝑊 ∗ ℎ𝑖:𝑖+𝑘−1 + 𝑏) ....(Eq 3) where 𝑏 is a bias term, and ∗ denotes convolution. 1. Initialize multiple filters of different sizes. 2. Convolve each filter over the BiLSTM output. 3. Apply ReLU activation. 4. Perform max pooling over each feature map. 5. Concatenate pooled features to form the final output vector. 2.4 Attention Mechanism Layer The attention layer selectively focuses on significant words, enhancing the model’s interpretability and accuracy. Attention scores are calculated for each word, amplifying those contributing most to the sentiment. This dynamic weighting improves the model’s focus on sentiment-relevant text portions. Given hidden states 𝐻, attention weights 𝛼𝑖 are computed: Eq 4 𝛼𝑖 = 𝑒𝑥𝑝(ℎ𝑖 𝑇⋅𝑊𝑎) ∑𝑛 𝑗=1 𝑒𝑥𝑝(ℎ𝑗 𝑇⋅𝑊𝑎) ....(Eq 4) where 𝑊𝑎 is the attention weight matrix. The context vector 𝐶 is then: Eq 5 𝐶 = ∑𝑛 𝑖=1 𝛼𝑖ℎ𝑖 ....(Eq 5) Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 579 https://internationalpubls.com 2.5 Nonlinear Feature Fusion Layer This layer combines all extracted features using dense neural networks, applying nonlinear activation functions to learn complex sentiment patterns. Dense layers with ReLU activation merge the CNN and BiLSTM outputs, capturing nonlinear dependencies between extracted features. Dropout regularization ensures the model does not overfit. For dense layer input 𝑥, the output is: Eq 6 𝑦 = 𝑅𝑒𝐿𝑈(𝑊𝑑 ⋅ 𝑥 + 𝑏𝑑) ....(Eq 6) where 𝑊𝑑 and 𝑏𝑑 are the weight matrix and bias vector, respectively. 2.6 Sentiment Prediction Layer The final layer produces the sentiment classification, providing a probability distribution over sentiment classes. A fully connected layer aggregates the feature fusion layer’s outputs, feeding into a softmax activation for multi-class classification or sigmoid for binary classification. For multi-class classification, the softmax function outputs probabilities: Eq 7 𝑃(𝑦𝑖) = 𝑒𝑥𝑝(𝑧𝑖) ∑𝐾 𝑗=1 𝑒𝑥𝑝(𝑧𝑗) ....(Eq 7) where 𝑧 is the input to the softmax layer and 𝐾 is the number of classes. 3 Experimental Study The experimental study aims to evaluate the performance of the proposed NPSC (Nonlinear Precise Sentiment Classification) framework against contemporary sentiment classification models using the IMDB movie review dataset. This section details the experimental setup, including dataset descriptions, model configurations, evaluation metrics, and a comparative analysis of results. Dataset: The IMDB movie review dataset, sourced from Kaggle, was used for the experiments. This dataset contains 50,000 movie reviews split evenly between positive and negative sentiments, providing a balanced classification task. The reviews vary in length and complexity, making the dataset suitable for testing the model's ability to handle diverse and nuanced textual data. Each review underwent preprocessing, including text cleaning, tokenization, removal of stop words, and conversion into BERT embeddings, which served as inputs to the NPSC model. Model Configuration and Training: The NPSC model architecture utilized BERT embeddings to capture word semantics, followed by a Bidirectional LSTM (BiLSTM) to understand sequential dependencies from both directions. A Convolutional Neural Network (CNN) layer extracted local features, and an attention mechanism focused on sentiment-relevant words. The outputs were combined in a nonlinear fusion layer with ReLU activation, followed by a dense layer with dropout regularization to prevent overfitting. The final sentiment prediction was made using a fully connected layer with softmax activation. Training was conducted using the Adam optimizer with a learning rate of 0.001, a batch size of 64, and early stopping based on validation loss to avoid overfitting. The model was trained for up to 50 epochs, with dropout rates of 0.5 applied in the dense layers. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 580 https://internationalpubls.com Performance was assessed using several metrics: accuracy, precision, recall, and F1-score. These metrics provided a comprehensive evaluation of each model's classification capabilities, focusing on both the correct identification of sentiments and the minimization of false positives and negatives. Two contemporary models, RMDL [11] (Random Multimodel Deep Learning) and CLM [16] (Convolutional LSTM Model), were selected as baselines for comparison with NPSC. RMDL uses an ensemble of deep learning models, including LSTMs and CNNs, while CLM combines convolutional layers with LSTM networks to capture both local and sequential patterns in text data. These models represent advanced yet distinct approaches in sentiment classification, making them suitable comparators for evaluating NPSC. 3.1 Results and Analysis The results demonstrated that NPSC outperformed both RMDL [11] and CLM [16] across all evaluated metrics, as shown in Table 1. NPSC achieved the highest accuracy, precision, recall, and F1-score, highlighting its superior capability in precise sentiment classification. While CLM] showed better performance than RMDL, it still lagged behind NPSC, particularly in precision and recall, indicating that NPSC’s nonlinear architecture effectively captures and leverages complex relationships in textual data. Table 1: Performance Comparison of NPSC, RMDL, and CLM Model Accuracy (%) Precision (%) Recall (%) F1-Score (%) NPSC 94.2 93.8 94.5 94.1 CLM 91.6 90.9 91.2 91.0 RMDL 88.3 87.5 88.0 87.7 Ablation Study: The ablation study was conducted to assess the contribution of individual components within NPSC by systematically removing key elements and observing the impact on performance. Table 2 shows the results of removing the CNN layer, attention mechanism, and the nonlinear feature fusion layer. Table 2: Ablation Study Results of NPSC Model Configuration Accuracy (%) Precision (%) Recall (%) F1-Score (%) Full NPSC 94.2 93.8 94.5 94.1 NPSC without CNN 90.1 89.7 89.9 89.8 NPSC without Attention 88.9 88.2 88.6 88.4 NPSC without Nonlinear Fusion 86.5 85.8 86.1 85.9 The results highlight that each component plays a critical role in the overall performance of NPSC. Removing the CNN layer significantly reduced the accuracy and precision, demonstrating its importance in capturing local patterns. The absence of the attention mechanism led to a noticeable drop in recall, emphasizing its role in focusing on sentiment-critical elements. The nonlinear fusion layer was particularly impactful, as its removal caused the most significant decrease in performance metrics, underlining its essential function in integrating features to capture complex dependencies. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 581 https://internationalpubls.com Error Analysis: An error analysis was performed to examine the misclassifications made by NPSC, with a focus on understanding the nature of errors and identifying potential areas for improvement. The analysis revealed that most errors occurred in reviews with mixed or ambiguous sentiments, where both positive and negative elements were present within the same review. For instance, reviews that praised certain aspects of a movie while criticizing others posed a challenge for precise classification. Table 3: Error Analysis Results Error Type NPSC Misclassifications (%) CLM Misclassifications (%) RMDL Misclassifications (%) Mixed Sentiments 5.6 7.4 9.2 Neutral or Subtle Sentiments 3.1 4.6 6.3 Complex Language Usage 2.8 3.9 5.1 NPSC demonstrated a lower misclassification rate compared to CLM [16] and RMDL [11] across all error categories. The model’s advanced attention mechanism and feature fusion enabled it to better navigate the complexity of mixed sentiments and subtle language cues. However, further fine-tuning of the attention weights and incorporating external sentiment lexicons could potentially enhance NPSC’s ability to handle these challenging cases even more effectively. 3.2 Result Discussion The experimental results confirmed that NPSC significantly outperforms contemporary models in sentiment classification tasks. The nonlinear approach adopted by NPSC, integrating advanced components like BERT, BiLSTM, CNN, and attention, provides a clear advantage in accurately identifying sentiment nuances. CLM [16], although performing better than RMDL [11], lacked the precision and adaptability demonstrated by NPSC, highlighting the importance of a multi-layered and nonlinear feature extraction strategy in sentiment analysis. Figure 2: Performance Metrics of NPSC vs. Contemporary Models This bar graph presented in figure 2 compares the performance of the NPSC, CLM, and RMDL models across four key metrics: accuracy, precision, recall, and F1-score. The NPSC model demonstrates superior performance in all metrics, showcasing its advanced capability in precise sentiment Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 582 https://internationalpubls.com classification. CLM performs better than RMDL, but both are outperformed by NPSC, highlighting the effectiveness of the nonlinear approach and attention mechanisms in NPSC. Figure 3: Ablation Study Results of NPSC This line graph presents in figure 3 the results of the ablation study conducted on the NPSC model by removing critical components: CNN, attention mechanism, and nonlinear fusion layer. The study shows a significant decline in performance when these components are excluded, especially the nonlinear fusion layer, confirming their essential roles in achieving high accuracy and precision in sentiment classification. Figure 4: Error Analysis of NPSC and Baseline Models This bar graph shown in figure 4 illustrates the misclassification rates for NPSC, CLM [16], and RMDL [11] across different error types, including mixed sentiments, neutral or subtle sentiments, and complex language usage. NPSC shows the lowest error rates in all categories, demonstrating its superior ability to handle ambiguous and nuanced text, compared to the higher misclassification rates observed in CLM and RMDL. The experimental study validates the effectiveness of the NPSC framework, showcasing its ability to outperform established models like RMDL and CLM. The ablation studies and error analysis further confirmed the critical roles of each component within NPSC, emphasizing its superior capability in precise sentiment classification. The findings highlight the strengths of a nonlinear approach in sentiment classification, making NPSC a robust and precise tool for analyzing complex textual data. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 583 https://internationalpubls.com 4 Conclusion The study focused on building the Nonlinear Deep Learning Framework for Precise Sentiment Classification (NPSC) to improve how sentiment is analyzed in text data. The model combined BERT, BiLSTM, CNN, and attention layers to capture complex sentiment patterns, showing strong results in key areas like accuracy and recall. This combination allowed the model to pick up on subtle emotions and context within reviews, making the predictions sharper and more reliable. NPSC’s design moves beyond simple models by layering different techniques that work together, showing that blending multiple approaches captures more hidden sentiment cues. The findings point to a clear advantage of using nonlinear methods in sentiment analysis, especially in texts with mixed or unclear opinions. However, the model was tested on one dataset, and its performance could vary with other text types. Future work could look at testing NPSC across different datasets, tweaking each layer, or adding new features to deepen the analysis. This study shows that NPSC is not just another sentiment analysis tool; it’s a step towards better understanding complex emotions in text. The approach lays the groundwork for more precise and adaptable sentiment models, making it a valuable addition to fields where accurate emotion detection matters. References [1] Ain, Qurat Tul, Mubashir Ali, Amna Riaz, Amna Noureen, Muhammad Kamran, Babar Hayat, and A. Rehman. "Sentiment analysis using deep learning techniques: a review." International Journal of Advanced Computer Science and Applications 8, no. 6 (2017). [2] Wang, Wenliang. "Text Sentiment Classification Method Based on Bilstm." Highlights in Business, Economics and Management 21 (2023): 679-687. [3] Wu, Yichao, Zhengyu Jin, Chenxi Shi, Penghao Liang, and Tong Zhan. "Research on the Application of Deep Learning-based BERT Model in Sentiment Analysis." arXiv preprint arXiv:2403.08217 (2024). [4] Rose, J. Dafni, S. Bhuwaneshwaran, and S. Dhiyanesh. "Advanced Novel NLP Approach for Emotion Detection in Poetry." In 2024 International Conference on Advances in Computing, Communication and Applied Informatics (ACCAI), pp. 1-6. IEEE, 2024. [5] Liu, Jingyi, and Sheng Li. "A dependency-based hybrid deep learning framework for target-dependent sentiment classification." Pattern Recognition Letters 176 (2023): 160-166. [6] Kathiravan, M., S. Saravanan, M. Jagadeesh, I. Lakshmi, V. Sathya Durga, and N. Bharathi Raja. "Deep Learning- Driven Sentiment Analysis in Textual Data." In 2024 IEEE International Conference on Computing, Power and Communication Technologies (IC2PCT), vol. 5, pp. 1339-1342. IEEE, 2024. [7] Radha, Dinesh Raja, and L. Mary Gladence. "Text Emotion Multi-Label Classification with Vectoring and Deep Learning." In 2024 International Conference on Inventive Computation Technologies (ICICT), pp. 371-376. IEEE, 2024. [8] Bashynska, Iryna, Mykhailo Sarafanov, and Olga Manikaeva. "Research and Development of a Modern Deep Learning Model for Emotional Analysis Management of Text Data." Applied Sciences 14, no. 5 (2024): 1952. [9] Salman Al-Tameemi, Israa K., Mohammad-Reza Feizi-Derakhshi, Saeed Pashazadeh, and Mohammad Asadpour. "An Efficient Sentiment Classification Method with the Help of Neighbors and a Hybrid of RNN Models." Complexity 2023, no. 1 (2023): 1896556. [10] Zhao, Yuxia, Mahpirat Mamat, Alimjan Aysa, and Kurban Ubul. "A dynamic graph structural framework for implicit sentiment identification based on complementary semantic and structural information." Scientific Reports 14, no. 1 (2024): 16563. [11] Adilakshmi, Konda, Malladi Srinivas, Anuradha Kodali, and V. Srilakshmi. "Optimized RMDL with Transfer Learning for Sentiment Classification in the MapReduce Framework." Journal of Web Engineering 22, no. 8 (2023): 1101-1132. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 1s (2025) 584 https://internationalpubls.com [12] Sachin Sambhaji Patil, Anthon Rodrigues, Rahul Telangi, Vishwajeet Chavan. "Convolutional Neural Networks for Text Classification: A Comprehensive Analysis", International Journal for Research in Applied Science & Engineering Technology (IJRASET), ISSN: 2321-9653; pp. 6039-6047, Volume 11 Issue V May 2023 [13] Nguyen, Cuong V., Khiem H. Le, Anh M. Tran, Quang H. Pham, and Binh T. Nguyen. "Learning for amalgamation: A multi-source transfer learning framework for sentiment classification." Information Sciences 590 (2022): 1-14. [14] Suganya, V., and M. Rajeev Kumar. "Advancing Sentiment Analysis Precision through ConvNet-SVMBoVW Mode in a Hybrid Context-Enhanced Deep Learning Paradigm." In 2024 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC), pp. 836-843. IEEE, 2024. [15] Priyanka, Madineni, Pushadapu Karthik, and Prasanth Yalla. "Text Classification Into Emotional States using Deep Learning based BERT Technique." In 2023 8th International Conference on Communication and Electronics Systems (ICCES), pp. 1005-1014. IEEE, 2023. [16] Chen, Long, Zhilin Ren, Shaoshuai Lu, Xiaohua Huang, Wenjing Wang, Cai Xu, Wei Zhao, and Ziyu Guan. "A Simple Weakly-Supervised Contrastive Learning Framework for Few-shot Sentiment Classification." (2023). [17] Jadon, Adarsh Singh, Mahesh Parmar, and Rohit Agrawal. "Hinglish Sentiment Analysis: Deep Learning Models for Nuanced Sentiment Classification in Multilingual Digital Communication." In 2024 2nd International Conference on Device Intelligence, Computing and Communication Technologies (DICCT), pp. 318-323. IEEE, 2024. [18] Gamal, Donia, Marco Alfonse, Salud María Jiménez-Zafra, and Mostafa Aref. "Arabic Sentiment Classification on Twitter Using Deep Learning Techniques." In The International Symposium on Computer Science, Digital Economy and Intelligent Systems, pp. 236-251. Cham: Springer Nature Switzerland, 2022. [19] Garg, Sarita Bansal, and V. V. Subrahmanyam. "Super Deep Learning Ensemble Model for Sentiment Analysis." In 2023 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS), pp. 341-346. IEEE, 2023. [20] Raja, M. Ramesh, and J. Arunadevi. "Deep Active Learning Multiclass Classifier for the Sentimental Analysis in Imbalanced Unstructured Text Data." In 2023 International Conference on Data Science, Agents & Artificial Intelligence (ICDSAAI), pp. 1-6. IEEE, 2023. [21] Kiran, K., Abhishek Appaji, and P. Deepa Shenoy. "Sentiment Analysis of Textual Data using Word Embedding and Deep Learning Approaches." In 2023 IEEE North Karnataka Subsection Flagship International Conference (NKCon), pp. 1-6. IEEE, 2023. [22] Rangarajan, Prasanna Kumar, Bharathi Mohan Gurusamy, Gayathri Muthurasu, Rithani Mohan, Gundala Pallavi, Sulochana Vijayakumar, and Ali Altalbe. "The social media sentiment analysis framework: deep learning for sentiment analysis on social media." International Journal of Electrical and Computer Engineering (IJECE) 14, no. 3 (2024): 1-1x. [23] Hernández, Nayeli, Ildar Batyrshin, and Grigori Sidorov. "Evaluation of deep learning models for sentiment analysis." Journal of Intelligent & Fuzzy Systems 43, no. 6 (2022): 6953-6963.