USMA Admissions and Natural Language Processing

The United States Military Academy (USMA) at West Point is renowned forproducing Army Officers entrusted with the critical mission of leading Soldiersinto combat. USMA expects each graduate to serve as a leader of character,prepared to lead Soldiers in the United States Army. Through data collected atUSMA, this research provides a way to analyze the character of college appli-cants (prior to admission) using Natural Language Processing (NLP) techniquesand machine learning algorithms. We extract NLP variables from letters of rec-ommendation that were written about college applicants in an effort to predictthe number of negative Cadet observation reports (NCOR) they receive persemester, which we use as a proxy measure for poor character. We provide ev-idence for a positive relationship between the number of NCORs that a Cadetreceives per semester and recommendations with high average words per re-sponse and a higher than average proportion of negations. However, our resultsdemonstrate that the approach of using basic NLP techniques is insufficient foradmissions departments to achieve the very difficult task of assessing collegeapplicants for downstream character issues.