Lunch at 12:30pm, talk at 1pm, in 148 Fitzpatrick

Title: Toward Emotional Support on Social Media Forums

Abstract: Emotional support plays an essential role in individuals’ mental well-being. Providing emotional support for online help-seekers who suffer from negative life events and/or mental illnesses with sincere and targeted responses would help them deal with these challenges. Existing work on emotional support focused on dialog systems, which aimed to generate multi-turn, short and brief responses to keep the conversation going smoothly. Unlike the nature of dialogs, however, posts on social forums are usually long self-disclosures, with a great amount of details and an explicit need for targeted solutions. An effective response to these posts with rich contents would require a thorough understanding of the context and a proper combination of emotional support strategies. To study this problem, we collected a dataset from Reddit mental health channels, containing posts from help-seekers and corresponding responses from other users. We then propose both NLU and NLG tasks for emotional support on social media forums. For NLU, we propose short-text classification for emotional support strategies. Identifying which strategy each sentence belongs to can help us gain a better understanding of the “strategic flow” of effective responses. We observed that some semantic patterns are closely related to certain strategies, which could be useful for leveraging the classification task. For NLG, we aim to generate long responses that can directly provide emotional support for online help-seekers using language models.

Bio: Lingbo is a second year student in the joint Ph.D program of computer science and engineering department and psychology department, where she is advised by Prof. Zhiyong Johnny Zhang and Prof. Meng Jiang. Her research interest lies in natural language processing, social media analysis, and emotion computing.