Text mining analysis of 3,843 classroom observation narratives from a K-12 district using LDA topic modeling, TF-IDF inter-rater variation analysis, AFINN sentiment trajectory, and LASSO classification of structured instructional ratings. Built in R with tidytext, topicmodels, and glmnet.

This project applies basic text mining techniques to instructional feedback narratives from a district dataset. Using tokenization and a simplified embedded sentiment lexicon, I quantified positive and negative language within feedback comments and aggregated results at the anonymized observer level. The goal was to examine variation in tone across observers and demonstrate how sentiment analysis workflows can be applied to educational feedback data in R.

This data product explores the instructional intent within Learning Walkthrough (LWT) feedback in the Americas Mid-Atlantic district. By applying the Ladder of Feedback framework through sophisticated text mining techniques—specifically Regular Expression (Regex) pattern matching—this analysis distinguishes between supportive social filler and high-impact coaching. The results illustrate a district profile that is strong in rapport-building but identifies a critical opportunity to increase reflective questioning for teacher growth.