Leveraging Large Language Model Embeddings and Machine Learning for Predictive Analytics in Mental Health Based Sentiment Analysis on a Social Media Data

Nowadays, most people use social media sites to explore their opinions. People use it to communicate their opinions and sentiments on a wide range of topics, such as politics, social events, meteorological conditions, and other people. A lot of new strategies that people can utilize in their daily lives have also been made possible by Natural Language Processing (NLP) and new Artificial Intelligence (AI) techniques. Also, a major worldwide health concern, mental health crises have a reflective effect on people, families, and societies. Consequently, traditional identification algorithms frequently fail to deliver prompt assistance, even though early recognition and intervention are essential in reducing the intensity and duration of severe crises. Social media's widespread use in recent years has opened up a new avenue for tracking and examining behavioural patterns that might point to new mental health issues. Therefore, this paper develops a novel Hybrid Elephant and Krill Herd optimization-based Large Language Model (HEHHO-LLM) to predict and classify the social networking large text data about mental health like anxiety, depression, emotional stress, etc with that sentiment polarity like positive, negative and neutral. Also, this model enhances the efficiency and accuracy of mental health prediction from social media text data. Initially, the process starts with data collection from a standard web source. After that, the pre-processing phase enables cleaning and filtering the vague noise from the data so that can be easily moved to the next process. Consequently, the pre-processing stage utilizes a Gated Recurrent Unit (GRU) for managing the sequential text data. Here, text embeddings, text normalization, reduced dimensionality, and the tokenization process are performed in each layer. Then, the BERT model was enabled in feature extraction. Hereafter, the proposed HEHHO-LLM is to refine the features and improve the LLM performance. Finally, early signs of mental health issues are effectively detected and classified the sentiment polarity. This model was implemented on the Python platform, and results were compared with traditional ML models. This HEHHO-LLM provides better performance for enhancing predictive analytics in mental health through improved accuracy and robust analysis of large social media text data.