Multi-label Topic Classification of Patient Generated Content in a Breast-cancer Community Forum

The research community has been noticing the importance of the online forums in healthcare in understanding the nuances of health-related problems. Breast Cancer is the most common malignance among women worldwide and its survival rate is steadily increasing thanks to early diagnosis and timely and effective treatment. Still how they manage the disease and maintain the quality of daily life are worth understanding. And it is in this context that the online posts of patients with breast cancer, discussing several topics, become patient generated content. The multi-label nature with these posts arising out of the combined contents of these posts can bring out a multitude of divulging conclusions. The resulting classification issues of these online posts under various categories of topics is examined in the present work. A semi-supervised multi-label classification, followed by refinement of multiple assignments of labels based on fuzzy logic and a neighborhood technique is proposed in the paper. Multiplicity of labels, occurs during the assignment of labels to clusters based on proximity. While extending the refined label set to a greater number of unlabeled posts clustered on the basis of proximity, it is observed that the proposed method could bring out more information on the description of posts. The results thus convey that the most discussed topic in the posts is about diagnosis, along with tangential reference to adverse drug effect, presumably to offer support in terms of information or viewpoint. The results show how the diverse nature of multiple labels can be effectively harnessed to draw conclusion from the potential of social media posts of patients' experience in critical health problems.