EmotionSence-AI : Online Sentiment Analysis Platform for Detecting Emotions

Abstract

Sentiment analysis is animportant part of natural language processing. Initially, the task was addressed using rule-based and statistical tools such as Naïve Bayes and Support Vector Machines (SVM). With the ever-increasing amount of available data, these methods became too simplistic. Deep learning has revolutionized not only the task of sentiment analysis and other tasks in natural language processing. Currently, the best models are context-aware models like BERT. Despite their advent and the advances they have made. The main issues holding the field back is that it\'s and recognize unstructured data effectively, especially when that unstructured data is concatenated with sarcasm and ambiguity. This study investigated the performance of basic traditional models (Naïve Bayes, Lexicon-based methods, Linear SVM) and modern deep leaning model (BERT) when applied to emotional analysis on unresolved big data issues (customer review data).

Introduction

Opinion mining, also known as sentiment analysis, is a critical area of Natural Language Processing (NLP), especially with the rise of user-generated content on platforms like social media, e-commerce, and review sites. Traditional methods (rule-based, Naïve Bayes, SVM) were limited in handling context, sarcasm, slang, and ambiguity in text.

Recent advancements like BERT (Bidirectional Encoder Representations from Transformers) have made it possible to better interpret contextual and emotional nuances in unstructured data. This study evaluates several sentiment analysis methods for use in recommendation systems using Amazon product reviews as the dataset.

2. Objectives

• Evaluate traditional and modern sentiment analysis techniques.

• Compare models like Naïve Bayes, SVM, VADER, and BERT.

• Measure performance in terms of accuracy, precision, recall, and F1-score.

• Integrate the best model into real-time applications like recommender systems.

3. Literature Review Highlights

• Pang & Lee: Early work using Naïve Bayes and SVM for movie reviews.

• Liu: Emphasized importance of topic-specific words.

• Cambria & White: Proposed concept-based sentiment analysis.

• Hutto & Gilbert (VADER): Rule-based sentiment tool for social media.

• Zhang et al.: Used character-level CNNs to handle slang and misspellings.

• Socher et al.: Applied Recursive Neural Tensor Networks for sentence-level sentiment.

• BERT (Devlin et al.): Revolutionized NLP with context-aware embeddings.

• ELMo (Peters et al.): Produced sentence-dependent word representations.

4. Proposed Framework

A complete end-to-end sentiment analysis system is built using the following steps:

A. Dataset Collection

• Source: Amazon product reviews.

• Reviews mapped into 3 classes:

  • 1–2 stars → Negative

  • 3 stars → Neutral

  • 4–5 stars → Positive

B. Preprocessing

• Clean and normalize text: lowercase, remove punctuation, stopwords, URLs.

• Tokenization and lemmatization/stemming for uniformity.

C. Feature Extraction

• Traditional: TF-IDF, Bag-of-Words.

• Deep Learning: Word embeddings, BERT tokenization.

D. Model Training

• Models trained on 80% of data and tested on 20%.

• Algorithms used:

  • Naïve Bayes (fast, simple)

  • SVM (high-dimensional data)

  • VADER (rule-based)

  • BERT (context-aware deep learning)

E. Prediction & Visualization

• Tools used: Confusion matrices, ROC curves, Precision-Recall graphs, Word clouds.

• Helps visualize how the model makes predictions and which words influence decisions.

F. Deployment

• A REST API built using Flask/FastAPI.

• Ready for integration into:

  • Recommendation engines

  • Customer feedback platforms

  • CRM tools

• Supports cloud deployment (e.g., AWS, GCP) and model retraining.

5. Results & Evaluation

A. Performance Metrics

• Metrics used: Accuracy, Precision, Recall, F1-score.

• These give a well-rounded view of model effectiveness, especially on unbalanced data.

B. Model Comparison

• BERT outperformed all others, showcasing its strength in understanding contextual and emotional complexity.

• Lexicon-based models (like VADER) are simple but less adaptable to modern text patterns.

Conclusion

This study presents a complete sentiment analysis system that addresses the challenge of identifying and categorizing emotions in large volumes of disorganized text. It begins by gathering real-world data from Amazon product reviews. The method follows a clear process that includes cleaning the text, extracting key features, training models, and presenting the results. The research used and tested both traditional models and deep learning models, applying standard methods to measure performance. The outcome highlights the traditional models like Naïve Bayes and SVM are effective and cheaper to operate, transformer-based models like BERT achieve significantly higher F1-score. This demonstrates that the system can manage complex language, context, and subtle emotions. The results show that the system is effective at solving the main problem it was designed for, improving the accuracy and trustworthiness of sentiment analysis for real-world uses. The system is designed to be flexible and simple to use in various areas, such as recommendation tools, customer feedback systems, and market trend trackers. Although deep learning models require greater time and resources to train, their strong performance is worth the effort, especially when context matters. The system also includes tools to visualize and deploy the results. This makes sentiment analysis both clear and easy to use, providing real value for businesses and researchers. Looking ahead, the system can be improved in many ways. Adding multilingual capabilities will make it better for specific areas through fine-tuning. Using tools like Kafka or Spark for real-time sentiment analysis will increase its usefulness across different groups and industries. Explainability tools like SHAP and LIME will clarify the model’s decisions, which will help build trust. Including aspect-based sentiment analysis will provide more detailed feedback, allowing businesses to understand feelings about specific parts of a product or service. These future improvements will enhance the system, leading to smarter and more user-focused sentiment analysis tools.

References

[1] Sharma, H., & Bansal, D. (2023). Opinion Mining Using Double Channel CNN for Recommender System. arXiv preprint arXiv:2307.07798. [2] Singh, A., & Kaur, N. (2023). Improving Collaborative Filtering-Based Recommender System Using Sentiment Analysis. International Journal of Innovative Technology and Exploring Engineering. [3] Kumar, A., & Tyagi, S. (2024). Efficient Sentiment Analysis: A Resource-Aware Evaluation of Feature Extraction Techniques, Ensembling, and Deep Learning Models. Journal of King Saud University – Computer and Information Sciences. [4] Muhammad, N., Javed, A. R., & Ahmed, S. H. (2024). Deep Learning-Based Sentiment Classification: A Comparative Survey. IEEE Access. [5] Reddy, A., Kumar, R., & Goyal, D. (2023). A Smart Recommender Model Based on Learning Method for Sentiment Classification. Journal of Intelligent & Fuzzy Systems. [6] Zhang, S., Zhang, Y., He, X., & Chua, T.-S. (2024). Recommender Systems in the Era of Large Language Models (LLMs). arXiv preprint arXiv:2403.01871. [7] Chopra, R., & Sharma, S. (2024). Innovative Sentiment Analysis and Prediction of Stock Price Using FinBERT, GPT-4, and Logistic Regression. International Journal of Advanced Computer. [8] Kumari, P., & Kumar, R. (2023). Comparative Evaluation of Deep Learning and Machine Learning Techniques for Sentiment Analysis of Electronic Product Review Data. Procedia Computer Science. [9] Sharma, M., & Sharma, A. (2024). Challenges and Future in Deep Learning for Sentiment Analysis: A Review and a Proposed Novel Hybrid Approach. Journal of Big Data.

Copyright

Copyright © 2025 Ms. Umme Salmah N, Mrs. Subhashree D C, Dr. Girish Kumar D, Mrs. Sharvani V, Ms. M M Harshitha. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.