From Relevant Posts to Attributable Spans: Explicit Symptom Evidence for Depression Questionnaire Inference

The authors report in this preprint: Inferring standardized depression questionnaire responses from longitudinal social media histories often suffers from retrieval errors, such as picking up discussions about other people or multi-symptom posts that lack precise attribution. To address this, the authors propose SREE (Symptom Retrieval and Evidence-Grounded InfErence). SREE extracts first-person and symptom-specific text spans, selects a representative subset using a Facility Location-inspired objective, and performs Evidence-Gated Inference. Evaluated on BDI-Sen and eRisk 2019-2021 datasets, SREE achieves significant improvements in post-level precision and F1 compared to standard dense retrieval, while obtaining the highest exact item-level agreement across eRisk collections. This is a preprint paper, and while it improves evidence localization and attribution, the source does not detail independent replication or code release status.