Mathematical modelling of attention-guided deep learning for real-time text emotion recognition in assistive...

This study introduces an Attention-Guided Deep Learning Framework for Real-Time Text Emotion Recognition for Individuals with Disabilities (AGDL-TERID) model. The AGDL-TERID model aims to enhance assistive human-computer interaction by accurately identifying emotions from textual inputs generated by users with disabilities. In the initial stage, text pre-processing includes symbol removal and the removal of punctuation, stop word removal, tokenisation, stemming, and removal of HTML Tags or URLs, which are applied to standardise and clean the data. Following text pre-processing, features are extracted using term frequency-inverse document frequency (TF-IDF) word embeddings, which capture the importance of relevant terms. Finally, classification is performed utilising an attention-based convolutional recurrent neural network (A-CRNN) approach to capture both local patterns and long-range dependencies from text. The experimental result investigation of the AGDL-TERID model are investigated using a benchmark dataset. The experimental analysis of the AGDL-TERID approach portrayed a superior accuracy values of 97.47% and 98.76% over existing models under the dual datasets. The comparative results confirmed the improved performance of the proposed model compared to existing approaches in terms of diverse metrics, confirming its reliability and suitability for real-time assistive emotion recognition applications. This model can be used in real-world systems like social media, mental-health observation systems and chatbots for identifying user emotions from text. It provides good and highly personalised replies that are based on the emotions detected. It also supports assistive systems to understand their emotions efficiently. In general, it enhances human computer communications using different emotions of users.Implications for RehabilitationThe proposed AGDL-TERID framework can help assistive systems recognise emotional states from users' textual inputs, supporting more responsive and personalised human-computer interaction.Emotion recognition can enable assistive applications and chatbots to provide responses that are better aligned with the user's detected emotional state, potentially improving communication experiences for people with disabilities.The reported classification performance suggests potential for integration into emotion-aware assistive systems, including communication and monitoring applications.Future work should develop user-specific assistive datasets and improve interpretability through attention visualisation and personalisation, which may strengthen the adaptability of real-time assistive communication systems.