Assessing Supervised Natural Language Processing (NLP) Classification of Violent Death Narratives: Development and Assessment of a Compact Large Language Model (LLM) Approach

ObjectiveThe recent availability of law enforcement and coroner/medical examiner reports for nearly every violent death in the US expands the potential for natural language processing (NLP) research into violence. The objective of this work is to assess applications of supervised NLP to unstructured narrative data in the National Violent Death Reporting System (NVDRS). Materials and MethodsThis analysis applied distilBERT, a compact LLM, to unstructured narrative data to simulate the impacts of pre-processing, volume and composition of training data on model performance, evaluated by F1-scores, precision, recall and the false negative rate. Model performance was evaluated for bias by race, ethnicity, and sex by comparing F1-scores across subgroups. ResultsA minimum training set of 1,500 cases was necessary to achieve an F1-score of 0.6 and a false negative rate of .01-.05 with a compact LLM. Replacement of domain-specific jargon improved model performance while oversampling positive class cases to address class imbalance did not substantially improve F1 scores. Between racial and ethnic groups, F1-score disparities ranged from 0.2 to 0.25, and between male and female victims differences ranged from 0.12 to 0.2. DiscussionFindings demonstrate that compact LLMs with sufficient training data can be applied to supervised NLP tasks to events with class imbalance in NVDRS unstructured police and coroner/medical examiner reports. ConclusionSimulations of supervised text classification across the model-fitting process of pre-processing and training a compact LLM informed NLP applications to unstructured death narrative data.