Social media–driven machine learning approach to identify emotional support needs in cancer care

Abstract

Abstract Early detection of emotional imbalance in cancer patients is essential for delivering timely and appropriate supportive care for such people. Cancer presents not only a severe physical challenge but also a profound emotional burden for patients and caregivers. However, existing research has inadequately explored these emotional experiences, often relying on small, limited, or less accurate datasets that fail to capture the full complexity of mental health challenges. As an assistive way for such direction, this study addresses this problem by collecting and using a novel dataset that includes 10,087 online posts collected from various social media platforms such as Reddit, Daily Strength, and Health Boards. To our knowledge, this is the first study to apply advanced NLP techniques to this specific dataset. As the previous research suffers from small size datasets, which may lead to inaccurate and inconsistent results, we implement advanced preprocessing and data augmentation techniques, including Simple Augmenter, Word2Vec, GloVe, and back translation. These techniques greatly improve the dataset’s quality and diversity, leading to a better consistent dataset. Several machine learning models, including Logistic Regression, Linear Support Vector Classifier, and artificial neural networks, are utilized to detect the mental health condition of the patients through this dataset. Logistic Regression achieved the highest accuracy of 97.9%, which is a significant improvement compared to earlier methods. By leveraging a richer dataset and refined analytical methods, this work provides more accurate insights into the emotional and psychological needs of cancer patients and caregivers, filling a critical gap in the literature in this direction.