This paper presents a Bayesian logistic regression approach for text categorization that avoids overfitting and produces sparse predictive models. The method uses a Laplace prior to encourage sparsity in the model parameters, leading to compact classifiers that are as effective as those produced by support vector machines (SVMs) or ridge logistic regression with feature selection. The authors describe their model fitting algorithm, open-source implementations (BBR and BMR), and experimental results. The paper begins with an introduction to the challenges of maximum likelihood logistic regression in high-dimensional text data, where the number of predictor variables often exceeds the number of observations. The authors then describe their Bayesian approach, which uses a prior distribution that favors sparsity in the model parameters. They present the basics of their Bayesian approach, the fitting algorithm, and the data sets and methods used in their experiments. The authors compare their Bayesian logistic regression approach to other methods, including SVMs and ridge logistic regression with feature selection, on five text categorization data sets. They find that their approach produces classifiers that are competitive with state-of-the-art text categorization algorithms, including widely used feature selection methods. The paper also discusses the use of different priors, including Gaussian and Laplace priors, and describes the CLG algorithm for ridge logistic regression. The authors modify the CLG algorithm to fit lasso logistic regression models and discuss the selection of hyperparameters. They also describe the text representation methods used in their experiments, including TF-IDF weighting and cosine normalization. The authors evaluate the effectiveness of their approach using metrics such as error rate and F1 measure. They find that their Bayesian logistic regression approach outperforms ridge logistic regression and SVMs on most data sets. The paper concludes with directions for future work.This paper presents a Bayesian logistic regression approach for text categorization that avoids overfitting and produces sparse predictive models. The method uses a Laplace prior to encourage sparsity in the model parameters, leading to compact classifiers that are as effective as those produced by support vector machines (SVMs) or ridge logistic regression with feature selection. The authors describe their model fitting algorithm, open-source implementations (BBR and BMR), and experimental results. The paper begins with an introduction to the challenges of maximum likelihood logistic regression in high-dimensional text data, where the number of predictor variables often exceeds the number of observations. The authors then describe their Bayesian approach, which uses a prior distribution that favors sparsity in the model parameters. They present the basics of their Bayesian approach, the fitting algorithm, and the data sets and methods used in their experiments. The authors compare their Bayesian logistic regression approach to other methods, including SVMs and ridge logistic regression with feature selection, on five text categorization data sets. They find that their approach produces classifiers that are competitive with state-of-the-art text categorization algorithms, including widely used feature selection methods. The paper also discusses the use of different priors, including Gaussian and Laplace priors, and describes the CLG algorithm for ridge logistic regression. The authors modify the CLG algorithm to fit lasso logistic regression models and discuss the selection of hyperparameters. They also describe the text representation methods used in their experiments, including TF-IDF weighting and cosine normalization. The authors evaluate the effectiveness of their approach using metrics such as error rate and F1 measure. They find that their Bayesian logistic regression approach outperforms ridge logistic regression and SVMs on most data sets. The paper concludes with directions for future work.