NLP: Predicting Political Ideologies

Completed May 2026

What I Built

Built an NLP pipeline in Python to classify political party of U.S. presidents from 10,818 executive orders including text cleaning and spaCy-based NER to find linguistic patterns
Benchmarked 5+ modeling approaches including Naive Bayes, Logistic Regression, LinearSVC,
Decision Trees, custom CNN with GloVe embeddings, and fine-tuned distilBERT/legalBERT transformers, using features such as bag-of-words, TF-IDF, and Word2Vec representations
Achieved 86%+ F1 with fine-tuned BERT models versus 57.5% uniform baseline, and ran controlled tests (with/without NER) to show partisan language patterns persist independent of named entities
Conducted exploratory data analysis (term frequency, TF-IDF word clouds, token-length trends across presidencies) to surface partisan linguistic differences and changes in complexity over two centurie