Week 7: Text Classification

Using supervised machine learning to classify text documents — predicting outcomes from language with Naive Bayes and regularized regression.

Learning Objectives

By the end of this week, you will be able to:

Explain the generative assumptions behind Naive Bayes classification
Build a text classification pipeline using {tidymodels} and {textrecipes}
Apply tf-idf feature engineering as a preprocessing step for classification
Evaluate model performance using accuracy, ROC AUC, and confusion matrices
Tune a Lasso regularization penalty and interpret the resulting word-importance coefficients

Lecture Materials

Slides

🖥 Slide deck from this week’s lecture

View Slides

Code & Data

💻 R scripts used during class:

Lab5

📂 Assignment

Lab 5 — Text Classification Graded · 11 pts

Due: May 19 @ 11:59 pm

In this lab you will build a supervised text classifier to predict whether a climbing incident report describes a fatal or non-fatal accident. You will fit a Naive Bayes baseline, then compare its performance to other models on held-out test data.

Download Lab 5

📚 Text Classification in Environmental Science — Key Citations

Curated readings organized by thematic area. Useful background for situating the classification methods from this week in applied environmental research contexts.

Climate Change & Misinformation

Citation	Year	Topic	Keywords
Coan et al. — Nature Climate Change	2021	Climate Denial Detection	contrarian claims, binary classification, SVM, misinformation, media framing
Shyrokykh et al. — PLOS ONE	2023	Climate Tweet Classification	short text, Naive Bayes, SVM, BERT, Twitter, climate relevance
Webersinke et al. — arXiv	2021	ClimateBERT	domain-adapted language model, climate-relevant text detection, transfer learning

Biodiversity & Conservation

Citation	Year	Topic	Keywords
Kulkarni et al. — Methods in Ecology & Evolution	2021	Threatened Species News Filtering	binary classification, CITES, news relevance, random forest, information retrieval
Grasso et al. — arXiv	2024	EcoVerse Dataset	eco-relevance, annotated Twitter dataset, binary classification benchmark

Environmental Health & Pollution

Citation	Year	Topic	Keywords
Patel et al. — PLOS ONE	2017	Exposure Assessment Literature Screening	text mining, PubMed, binary relevance classification, systematic review automation
Bellinger et al. — BMC Public Health	2017	Air Pollution Epidemiology Screening	ML, data mining, systematic review, air pollution, study relevance classification

← Week 6: LDA Topic Modeling