Understanding Social Support Needs in Health Question: An Answer-Enhanced Semi-Supervised Deep Learning Approach

Increasingly, individuals are seeking social support through online, which can significantly improve both mental and physical well-being. However, inadequate or misaligned support can be ineffective or even be harmful. This underscores the need for classifying the social support needs in health-related questions. Developing such a model requires substantial labeled data, which is often expensive to manually annotate. To address this issue, we develop a novel Answer-Enhanced Semi-Supervised Deep Learning (AESSDL) Approach. The AESSDL approach incorporates a dynamic 2D interaction kernel designed to capture complex interaction patterns between questions and answers, along with a quality-aware attention layer that assigns varying weights to multiple answers for each question. Extensive empirical analyses demonstrate that our method outperforms existing question classification techniques and alternative semi-supervised approaches. Practically, our approach enables online platforms to better understand user needs, facilitating timely and personalized interventions, such as recommending appropriate responders and relevant digital resources.