Utilizing Machine Learning for Language Prediction and Text Emotion Analysis

Neelamadhab Padhy; Jyotiram Manik; Jogabrata Manik

Abstract:

Context:

In this study, we developed a robust model for the classification of user input text data into language and emotion classes. Language and emotion classification are pivotal tasks in various domains.

Objective:

Our objective was to precisely categorize text data into specific language and emotion classes using supervised learning algorithms.

Methods:

For language classification, we employed the Multinomial Naive Bayes (NB) algorithm due to its effectiveness in multi-class classification. This algorithm outperformed other classifiers such as Support Vector Machine (SVM) and Random Forest Classifier, particularly in handling large and sparse feature spaces inherent in text data.

For emotion classification, we utilized a Neural Network Model, specifically a Sequential Model, known for its efficiency in capturing complex patterns in textual data, thereby enabling granular emotion classification. By leveraging layers of interconnected neurons, Neural Networks can effectively capture the latent structure of emotional expressions embedded within text.

Additionally, feature selection techniques including Count Vectorizer, label encoding, and performance metrics evaluation were employed to assess model accuracy, precision, F1 score, and recall.

Results:

Our results demonstrated an accuracy of 95.32% in Language Classification and 83% in Emotion Classification, which represents a significant improvement compared to other classification techniques. The choice of using a Neural Network for emotion classification was justified by its ability to discern detailed patterns in text, facilitating precise emotion categorization. Count Vectorizer, a widely used feature extraction technique in natural language processing tasks, was employed for its simplicity, efficiency, and effectiveness in converting text documents into a matrix representation of token counts.

Conclusion:

In conclusion, our developed model provides a reliable tool for language and emotion classification tasks. By leveraging supervised learning algorithms such as Multinomial NB and the Neural Network Model, the model achieves high accuracy. Furthermore, feature selection techniques and performance metrics evaluation ensure robustness and generalization across diverse datasets.