The effectiveness of sex offender management policies and supervision approaches relies on the ability of criminal justice professionals to accurately differentiate sexual offenders according to their risk to recidivate. The US Probation and Pretrial Services Office (PPSO) has launched the Risk Tool for Online Child Pornography offenders (RTOCP). specifically designed to predict recidivism among online child pornography offenders.

This paper describes the Child Pornography Risk Assessment Study project, which has successfully demonstrated a proof-of-concept capability, applying NLP and AI technology, to enable the PPSO risk assessment effort by automatically detecting relevant information in offenders' case material, analyzing each sentence in-depth, and generating the corresponding RTOCP surveys. By automating this very time-consuming and painstaking task of reading through thousands of pages of sex offender documents and interpreting the information --an average of 10,000 sentences for every client, the prototype provides probation office personnel with an efficient and consistent method to estimate the likelihood of future criminal behavior in adult male online sex offenders under their supervision. The prototype's error rate was found to be smaller than a human annotators' error rate for this task.

[This document was not public released and is not shareable. Please contact me if interested]