CS Paper Radar

Leakage-aware calibrated decision-level fusion for mental-health text classification

Wadii Boulila, Giuseppe Varone, Bilel Benjdira, Maha Driss

cs.CL ·submitted 2026-11-14 11:58 UTC · v1
published

Abstract

Decision-level fusion benefits from complementary errors, calibrated probabilities, and leakage-aware evaluation. We evaluated a heterogeneous transformer fusion framework on 262,276 proxy-labeled Reddit records covering anxiety, depression, schizophrenia, and suicidal ideation. In a stratified random split, MinHash-LSH identified 29.1% of test records as lexical near-duplicates of training data. Augmentation-lineage auditing found 50.2% lineage overlap with training, including 30.0% of records retained after lexical filtering. We therefore used lineage-disjoint partitions. Twelve encoders, comprising three seeds each of Bio-ClinicalBERT, MentalBERT, MentalRoBERTa, and RoBERTa-base, underwent cross-fitted validation for calibration and ensemble selection. A class-specific Complementarity Index (CI), defined as the Jaccard distance between members' correctly classified validation sets, was assessed against independent fusion gain. CI correlated with gain for depression (r=0.50) and suicidal ideation (r=0.68), but not anxiety; results for schizophrenia were mixed. The selected MentalBERT, MentalRoBERTa, and RoBERTa-base triad achieved macro-F1 0.844 [0.823, 0.863], MCC 0.779, and ECE 0.0033 after post-fusion recalibration, improving macro-F1 by 0.021 [0.007, 0.037] over the validation-selected best single encoder. Integrated-Gradients attribution diversity was 2.3 times greater in fusion-rescued cases. Removing divergence-driving words destroyed 58% of rescues, versus 12% under random removal. These findings show that lineage auditing detects dependencies missed by lexical filtering, while class-specific complementarity and post-fusion calibration reveal ensemble behavior obscured by aggregate accuracy.

Closest papers

Nearest neighbours by meaning across the whole index.

From Retrieval to Typed Decisions: Calibrated System One Models from Biomedical Sentence Encoders

cs.CL cs.AI ·2026-10-01 ·Pritam Deka
highly cited similarity 0.61
Abstract

Typed decision models answer schema-constrained questions about a text in one forward pass and return probabilities meant to be thresholded. We ask whether biomedical sentence encoders trained for retrieval are good starting points for such models. We present SBERT2S1, which converts Sentence-Transformers encoders into bi-encoder, cross-head (C) and prior-fused residual (PFR) decision models, together with BIODECIDE, a biomedical typed-decision suite, and MEDLINE-S1, 243k training decisions derived from NLM indexing. Across six parent-retriever pairs, retrieval training improves zero-shot matching of content-bearing options. After fine-tuning, its effect depends on the head: across five pairs and three training-set sizes, retrieval training significantly helps PFR, which keeps the retrieval prior, in 10 of 15 comparisons, but helps C in one and hurts it in five. A matched grid of two heads and five training objectives shows that C outperforms PFR under every objective, and that the released RLCD recipe of open System One models trails cross-entropy by 2.5-3.0 points. The deficit stems mainly from its reward normalisation, which inflates the noisy score-function term 3.6-15-fold; an unbiased leave-one-out estimator recovers most of the gap. After temperature scaling, no objective is clearly better calibrated than cross-entropy. We release the code, the MEDLINE-S1 labels and a model.

0.44

A corpus-only AUC diagnostic for pooled speech-based depression detection

eess.AS ·2026-10-22 ·Cezary Polak, Milosz Dudek, Daria Hemmerling
similarity 0.60
Abstract

When clinical speech corpora are pooled for evaluation, corpus membership can itself become predictive if class prevalences differ across sources. We derive the AUC of an oracle predictor that uses only corpus identity and empirical prevalence, a diagnostic computable directly from corpus composition. Across five depression corpora in three languages (590 speakers), the corpus-only AUC is 0.697, and corpus identity is predicted from six acoustic and self-supervised representations with 97.5-99.8% accuracy. Permuting depression labels within each corpus removes label structure while preserving prevalences; the residual pooled-minus-LOCO discrepancy then increases across three nested configurations in the same order as the diagnostic. Pooled 5-fold cross-validation also outscores corpus-held-out evaluation across all six representations. Prevalence equalisation substantially reduces the discrepancy for eGeMAPS and WavLM-Base, whereas a sample-size-matched control does not. Corpus-specific base rates therefore create a measurable shortcut opportunity in pooled clinical-speech evaluation, motivating corpus-aware diagnostics alongside pooled performance.

0.18

Candidate Generation and Definition-Guided Verification for Sentence-Level Depression Symptom Recognition

cs.CL ·2026-09-01 ·Weiming Li, Catarina Barata, Miguel Constante, Joao Sanches
publishedhot direction similarity 0.59
Abstract

Sentence-level recognition of depression symptoms is challenging because similar expressions can differ in symptom relevance, and language-model inference is insufficiently grounded in diagnostic definitions. This study proposes a two-stage framework separating symptom-candidate generation from definition-grounded verification. A contrastively fine-tuned sentence encoder generates a symptom candidate per sentence, and a fine-tuned language model verifies whether the candidate is present or absent using the sentence, its context, and a candidate-specific diagnostic definition, checking its judgment against that definition before answering. Evaluated against encoder, inference-based, medical, and general LLM baselines and a matched single-stage supervised classifier, the proposed pipeline attains the best accuracy and F1 scores of all methods, with rationales matching expert-authored annotations. A preliminary clinical audit indicates moderate alignment with diagnostic definitions, with explanation quality strongly dependent on prediction correctness. The results support decomposing symptom recognition into candidate generation and definition-grounded verification, though performance remains limited for rare categories.

0.45

When Layer Selection Misleads Speech Depression Detection

eess.AS ·2026-10-05 ·Paula A. Perez-Toro, David Gimeno-Gómez, Daniel Rückert, Andreas Maier
position similarity 0.59
Abstract · Submitted to ICASSP 2027

Pretrained speech representations are increasingly used for depression detection, but selecting the best encoder layer on the same data used for evaluation biases reported performance. On DAIC-WOZ, with repeated nested cross-validation across five deep encoder families, naive best-of-25-layer probing inflates AUC by up to $0.09$. On shuffled labels over the \emph{real} latent representations, naive best-of-25 still reaches AUC $0.59$ versus $0.50$ under nested selection, and this bias grows with the number of probed layers and with smaller samples, isolating selection, not signal, as the cause. The effect replicates on a second clinical corpus and language (Androids, Italian). Under leakage-free selection, a fair comparison across representation families identifies a compact, task-aligned affect--prosody model as competitive with substantially more complex SSL, deep-learning, audio-LLM, and codec-based approaches, while using $\sim\!10^{3}\times$ fewer trainable downstream parameters. Extending the analysis to individual PHQ-8 symptoms shows that the same selection bias persists at this finer-grained level. We argue nested model-selection protocols should be standard when probing encoder layer representations on small clinical speech datasets.

0.17

Bag of Tricks or Bag of Myths? Reducing Modeling Complexity with Task Knowledge in Explainable Suicide Risk Assessment

cs.CL cs.AI ·2026-09-07 ·Shlok Shelat, Shrey Salvi, Souvik Roy, Manas Gaur, Amit Sheth
similarity 0.58
Abstract · This work was shaped by the IEEE Big Data Cup, where our team, NIDHI, ranked 3rd overall. NIDHI (Neurosymbolic Intelligent Diagnostics for H…

Assessing suicide risk from social media text is a small-data, high-stakes setting requiring not only severity prediction but also supporting evidence and clinically relevant risk and protective factors. Yet common NLP techniques, including model scaling, synthetic data, loss reweighting, ensembling, and threshold tuning, are often applied without testing whether their gains hold up under severe class imbalance, coupled outputs, and limited author-level data. We study 1,635 clinician-annotated posts and audit 31 pre-specified techniques from 7 methodological families through roughly 300 controlled experiments on author-disjoint partitions. We found no prior audit of this playbook in this regime. The findings guide a task-grounded system for three outputs: 4-level suicide risk, evidence spans, and 24 clinical risk and protective factors. Only 5 of 31 comparisons produced reliable gains. We reformulate factor prediction as entailment between each post and its codebook definitions, using an architecturally diverse ensemble with class-balanced training and score rescaling. Risk predictions condition a 7-model evidence tagger ensemble; evidence restricts symbolic risk rules; and a difficult risk class is routed separately. The factor predictor remains independent because risk evidence provides no additional factor signal. We also correct a mismatch between validation scores used for threshold fitting and test-time ensemble scores through deployment-consistent calibration, yielding the largest improvement to the factor system. The final system achieves 0.8203 for risk, 0.7953 for evidence, and 0.7045 macro-F1 for factors, with a 0.7781 composite, ranking third among 53 teams. We call the underlying principle task-conditioned technique selection: retain techniques only when task-specific knowledge, structure, or empirical evidence justifies them.

0.22

Beyond Final-Token Classification: Heterogeneous Readouts for Evidence-Grounded Suicide Risk Detection

cs.CL cs.LG ·2026-09-19 ·Zirui Li, Yanling Li, Kaolanglang Gao
benchmark similarity 0.57
Abstract · Competition paper for the IEEE Big Data Cup 2026. Ranked 2nd among 31 participating teams and invited for submission to IEEE Big Data 2026

The IEEE BigData Cup benchmark combines three prediction problems with different output structures: ordinal suicide-risk classification, multi-label psychosocial factor detection, and extraction of supporting phrases. We introduce heterogeneous readout decomposition (HRD), which separates semantic verification from output realization. A locally deployed Qwen3.8-27B model, adapted with task-specific QLoRA adapters, produces both answer-token margins and layer-63 answer states for card-conditioned queries. HRD compares four latent scores for ordinal risk, retains token margins for most factors while routing seven labels through one shared latent probe, and constructs evidence sets from verbatim span candidates with calibrated, risk-conditional constraints. On two held-out user-grouped confirmation folds, the latent risk readout improves weighted F1 from 0.8237 to 0.8372 and macro F1 from 0.7965 to 0.8185. Selective factor routing improves macro F1 by 0.0105 and tail-label macro F1 by 0.0189; in contrast, global latent replacement and independent label-specific probes fail. The constrained evidence decoder raises pooled phrase F1 from 0.7488 to 0.7609 in row-level out-of-fold evaluation. Lenormand's best public result is 0.8052 on Subtask 1 and 0.6636 on Subtask 2, giving a composite score of 0.7627. These results identify the answer-token boundary, rather than semantic representation alone, as a measurable source of error in this benchmark.

0.11

The Ceiling Is in the Channel: Auditing Learner Gaps and Measurement Frontiers in Clinical Prediction

cs.AI ·2026-09-01 ·Sayeed Shafayet Chowdhury, Nusrat Jahan, Snehasis Mukhopadhyay, Shiaofen Fang, Vijay R. Ramakrishnan
similarity 0.57
Abstract

Clinical prediction can saturate for two different reasons: a fitted learner may fail to extract available information, or the recorded variables may impose a population frontier. We separate these quantities through the \emph{learner gap} and the \emph{measurement-channel ceiling}. Optimal balanced accuracy is characterized by total-variation separation, yielding architecture invariance, a sharp partial-identification result under replacement contamination, a cross-fitted ceiling estimator, and exact conditions for multimodal decision improvement. We add two finite-sample diagnostics, namely a label-permutation optimism floor and an underfit curve, and validate the audit on three real cohorts: UCI readmission ($n=99{,}343$), BRFSS diabetes ($n=253{,}680$), and NHANES HbA1c ($n=10{,}219$). Well-tuned gradient boosting nearly reaches the estimated frontier in UCI and BRFSS, whereas deliberately or practically deficient learners retain large gaps. NHANES yields a null difference between questionnaire and measured marginal frontiers but a significant joint complementarity gain, refining the simplistic claim that an objective modality must dominate. Across all cohorts, modest AUROC gains coexist with substantially larger Bayes decision-flip rates, and several architectures estimate similar frontiers while their achieved balanced accuracy differs sharply. A PRISMA-guided synthesis of 104 clinical tasks then shows that the same channel-level regularities recur across more than 18 disease categories: a broad but non-universal structured-clinical region, diminishing same-channel gains across model families, and higher performance when measurement channels change. The framework converts saturation from an empirical observation into an auditable decision: improve the learner when headroom remains; improve measurement when it does not.

0.18

MedWER: A Reproducible, Model-Free Evaluation Protocol for Medical Speech Recognition

cs.CL ·2026-09-04 ·Justin Behling
code/data similarity 0.57
Abstract · 10 pages, 3 tables. Code, term list, and evaluation scripts: https://github.com/Nordis-Tech/medwer. Package: pip install medwer. Source arch…

Overall word error rate hides clinically critical errors: a transcript can be 95% correct and still swap one drug for another. The usual fix weights errors on medical entities, and almost always depends on an evaluation-time named-entity recognition (NER) model or cloud API, which makes the metric's denominator a versioned black box. We present MedWER, an evaluation protocol and open-source tool for medical ASR whose denominator is a fixed, license-clean term list: 19,373 drug, diagnosis, symptom, and injury-mechanism entries projected from public sources. The protocol couples a pinned text normalizer with a phrase-aware term-restricted WER, the MedWER, so the only versioned component is a normalizer dependency held at an exact release and checked against committed golden fixtures. Coverage is validated against an independent provincial drug-benefit file the list was not built from; the matching heuristic is calibrated against ground-truth entity spans. Baselines for Moonshine~base, Whisper~base.en, and MedASR on two open benchmarks are scored with the released tool and reported with 95% confidence intervals from resampled per-utterance scores.

0.33