Identifying language associated with probation absconding using natural language processing: a proof-of-concept study

The objective of this proof-of-concept study was to apply natural language processing (NLP) techniques to case notes of probationers to identify textual features associated with absconding from supervision. A text regression method called concise comparative summaries (CCS) was applied to case notes from N = 2,000 misdemeanor and felony offenders on probation in Tarrant County, Texas. Half of the probationers had completed probation, while the other half had absconded. There were 10 words and phrases associated with probation absconding (e.g., “technical violation”, “failed pay”, “cannabinoids”) and 14 associated with probation completion (e.g., “paid fees”, “alcohol use”, and “everything going well”). Results affirm the utility of using NLP methods to identify key words and phrases that are associated with probation absconding and completion and may help generate hypotheses for future research on absconding.