Hierarchical approaches to Text-based Offense Classification

Researchers working with administrative crime data often must classify offense narratives into a common scheme for analysis purposes. No comprehensive standard currently exists, nor is there a mapping tool to transform raw descriptions into offense types. This paper introduces a new schema, the Uniform Crime Classification Standard (UCCS), and the Text-based Offense Classification (TOC) tool to address these shortcomings. The UCCS schema draws from existing efforts, aiming to better reflect offense severity and improve type disambiguation. The TOC tool is a machine learning algorithm that uses a hierarchical, multilayer perceptron classification framework, built on 313,209 hand-coded offense descriptions from 24 states, to translate raw descriptions into UCCS codes. We test how variations in data processing and modeling approaches affect recall, precision, and F1 scores to assess their relative influence on model performance. The code scheme and classification tool are collaborations between Measures for Justice and the Criminal Justice Administrative Records System.