arXiv subjects

Dragan Stoll

Publications and source records attributed to Dragan Stoll.

3 recordsLinked to original sources

arXiv2026

Validation of a Small Language Model for DSM-5 Substance Category Classification in Child Welfare Records

Background: Recent studies have demonstrated that large language models (LLMs) can perform binary classification tasks on child welfare narratives, detecting the presence or absence of constructs such as substance-related problems, domestic violence, and firearms involvement. Whether smaller, locally deployable models can move beyond binary detection to classify specific substance types from these narratives remains untested. Objective: To validate a locally hosted LLM classifier for identifying specific substance types aligned with DSM-5 categories in child welfare investigation narratives. Methods: A locally hosted 20-billion-parameter LLM classified child maltreatment investigation narratives from a Midwestern U.S. state. Records previously identified as containing substance-related problems were passed to a second classification stage targeting seven DSM-5 substance categories. Expert human review of 900 stratified cases assessed classification precision, recall, and inter-method reliability (Cohen's kappa). Test-retest stability was evaluated using approximately 15,000 independently classified records. Results: Five substance categories achieved almost perfect inter-method agreement (kappa = 0.94-1.00): alcohol, cannabis, opioid, stimulant, and sedative/hypnotic/anxiolytic. Classification precision ranged from 92% to 100% for these categories. Two low-prevalence categories (hallucinogen, inhalant) performed poorly. Test-retest agreement ranged from 92.1% to 99.1% across the seven categories. Conclusions: A small, locally hosted LLM can reliably classify substance types from child welfare administrative text, extending prior work on binary classification to multi-label substance identification.

cs.CL↗

arXiv2026

Reasoning Language Models for complex assessments tasks: Evaluating parental cooperation from child protection case reports

Purpose: Reasoning language models (RLMs) have demonstrated significant advances in solving complex reasoning tasks. We examined their potential to assess parental cooperation during CPS interventions using case reports, a case factor characterized by ambiguous and conflicting information. Methods: A four stage workflow comprising (1) case reports collection, (2) reasoning-based assessment of parental cooperation, (3) automated category extraction, and (4) case labeling was developed. The performance of RLMs with different parameter sizes (255B, 32B, 4B) was compared against human validated data. Two expert human reviewers (EHRs) independently classified a weighted random sample of reports. Results: The largest RLM achieved the highest accuracy (89%), outperforming the initial approach (80%). Classification accuracy was higher for mothers (93%) than for fathers (85%), and EHRs exhibited similar differences. Conclusions: RLMs' reasoning can effectively assess complex case factors such as parental cooperation. Lower accuracy in assessing fathers' cooperation supports the argument of a stronger professional focus on mothers in CPS interventions.

cs.CY↗

arXiv2026

Small Models Achieve Large Language Model Performance: Evaluating Reasoning-Enabled AI for Secure Child Welfare Research

Objective: This study develops a systematic benchmarking framework for testing whether language models can accurately identify constructs of interest in child welfare records. The objective is to assess how different model sizes and architectures perform on four validated benchmarks for classifying critical risk factors among child welfare-involved families: domestic violence, firearms, substance-related problems generally, and opioids specifically. Method: We constructed four benchmarks for identifying risk factors in child welfare investigation summaries: domestic violence, substance-related problems, firearms, and opioids (n=500 each). We evaluated seven model sizes (0.6B-32B parameters) in standard and extended reasoning modes, plus a mixture-of-experts variant. Cohen's kappa measured agreement with gold standard classifications established by human experts. Results: The benchmarking revealed a critical finding: bigger models are not better. A small 4B parameter model with extended reasoning proved most effective, outperforming models up to eight times larger. It consistently achieved "substantial" to "almost perfect" agreement across all four benchmark categories. This model achieved "almost perfect" agreement (\k{appa} = 0.93-0.96) on three benchmarks (substance-related problems, firearms, and opioids) and "substantial" agreement (\k{appa} = 0.74) on the most complex task (domestic violence). Small models with extended reasoning rivaled the largest models while being more resource-efficient. Conclusions: Small reasoning-enabled models achieve accuracy levels historically requiring larger architectures, enabling significant time and computational efficiencies. The benchmarking framework provides a method for evidence-based model selection to balance accuracy with practical resource constraints before operational deployment in social work research.

cs.CY↗