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

Authors: Sumanth Manduru, Carlotta Domeniconi

Subject: Computation and Language (cs.CL)

arXiv: 2611.12900 [cs.CL]

Submitted: 14 October 2026

DOI: https://doi.org/10.48550/arXiv.2611.12900

Abstract

Longitudinal social-media histories offer an opportunity to assess depression and can potentially support the inference of standardized depression questionnaire responses. However, this task requires evidence that is semantically relevant to a symptom, attributable to the user, and sufficiently specific to support an individual questionnaire item. Existing approaches rely on semantic similarity to retrieve posts, which can introduce misleading or non-attributable evidence—particularly in multi-symptom posts and discussions of others' experiences.

We propose SREE (Symptom Retrieval and Evidence-Grounded InfErence), an evidence-grounded pipeline that:

1. Extracts first-person and symptom-specific text spans,
2. Selects a representative subset using a Facility Location-inspired objective, and
3. Performs Evidence-Gated Inference over the selected evidence.



We evaluate evidence retrieval on BDI-Sen and evidence-conditioned questionnaire inference on eRisk 2019–2021. Compared with state-of-the-art retrieval approaches, SREE achieves relative improvements of 70% in post-level precision and 59% in F1, while retaining at least 30% fewer posts. On multi-symptom sentences, it also outperforms dense retrieval in precision and F1 even with twice the retrieval budget. Across all three eRisk collections, SREE achieves the highest exact item-level agreement while remaining competitive on ordinal and aggregate metrics.




These results highlight the importance of localizing attributable evidence and controlling its composition before inferring depression questionnaire responses.




Key Contributions



• Evidence localization: Moves beyond post-level semantic similarity toward explicit, first-person, symptom-specific text spans.
• Evidence-grounded pipeline (SREE): Combines span extraction, Facility Location-inspired subset selection, and Evidence-Gated Inference.
• Strong empirical gains: 70% relative improvement in post-level precision and 59% in F1 versus state-of-the-art retrieval, with fewer retained posts.
• Robustness on multi-symptom text: Outperforms dense retrieval even under a doubled retrieval budget.
• Consistent questionnaire inference performance: Highest exact item-level agreement across eRisk 2019–2021, competitive on ordinal and aggregate metrics.



Significance in 2026 Context




As LLM-based mental-health screening and digital phenotyping mature, the field increasingly emphasizes attributability, specificity, and auditability of evidence rather than broad semantic relevance. SREE's focus on localized, first-person symptom spans aligns with this shift—supporting more transparent, accountable questionnaire inference from social-media histories while reducing the risk of misleading evidence drawn from multi-symptom posts or third-party discussions.

via ArXiv CL+LG