A HYBRID MODEL FOR CONTEXT-AWARE EMOTION DETECTION USING NATURAL LANGUAGE PROCESSING (NLP) TECHNIQUES

Abstract

Emotion detection using text has been considered a critical activity in natural language processing (NLP) due to the fact that it is highly diverse in application in areas of social media analysis, customer services, and mental health tracking. The text data in the informal and noisy form can contain delicacy and context sensitive emotional reactions that are incapable of being represented and faithfully mirrored by the customary techniques, such as the lexicon-based and rule-based ones. The hybrid model proposed in this paper is a blend of deep contextual and deep learned representations of transformer-based models such as RoBERTa, with classical lexicon-based features and sequence modelling using Long Short-Term Memory (LSTM). The hybrid model is founded on the benefits of the two approaches to amplify emotion perception in social media texts that are usually featured by the use of informal language, slang, emojis, and ambivalent moods. The model was evaluated based on two publicly available datasets, the Emotion Corpus of Hugging Face (Twitter) and Go Emotions of Kaggle (Reddit) using the assistance of evaluation metrics like accuracy, precision, recall, and F1-score. The result indicated that hybrid model is more effective compared to the single models with the accuracy of 87.3 on the Twitter data and 85.6 on the Reddit data. The findings reveal the practicality of deep learning when collaborated with lexicon based methods in identifying the contextual sensitive emotions, and the solution is sure to apply in the real world situation in dynamism and noise factors.