PSG-Pair: A Psychological Snapshot Guided Model for Improved Query-Response Pairing in Psychological Wellness Communities for University Students

Psychological wellness, especially among university students, has become an increasingly significant global health issue, with millions affected worldwide. In response to the growing demand for accessible psychological support, we propose a novel model, Psychological Snapshot Guided Pairing (PSG-Pair), designed to enhance query-response pairing in Community Question Answering for Psychological Wellness (CQA-PW). Unlike traditional methods that focus solely on conceptual-level matching, PSG-Pair integrates role-based psychological snapshots derived from the historical posts of help-seekers and supporters. The model operates in two phases: the initial screening phase, which utilizes a BERT-based retrieval model to filter relevant supportive posts, and the pairing phase, which incorporates psychological snapshots using a stacked attention mechanism to refine conceptual pairings based on the psychological characteristics of users. Extensive experiments conducted on the CLPsych 2022 Shared Task dataset demonstrate that PSG-Pair significantly outperforms traditional single-phase models, enhancing both precision and recall in pairing processes. The inclusion of psychological snapshots allows the model to better handle the complexities of psychological wellness scenarios, especially for university students, thereby improving the overall effectiveness of automated psychological support systems. However, this study has several limitations. Firstly, the dataset used for experiments, although rich, still suffers from data imbalance and noise due to the high proportion of irrelevant negative samples, which could potentially impact the model’s performance. Secondly, while the approach demonstrates promising results in the context of psychological wellness, the generalizability of the model to other domains or applications remains uncertain. Further exploration into the adaptability of PSG-Pair to diverse scenarios is required. Additionally, while the current evaluation metrics adequately reflect the retrieval and pairing capabilities, there is a need for the development of more tailored evaluation systems to assess models within the unique context of psychological wellness support for university students. Future work should also investigate how to mitigate biases in user-generated content, as the quality and authenticity of answers in non-factual Q&A platforms can vary significantly, potentially affecting the accuracy of the pairing.