Here, we experimented with various models to predict popularity of online news articles, quantified by shares. The booming of online media highlights the business needs to foresee text influences prior to publications. We studied data features through heavy visualizations, based on which we deployed greedy search, recursive feature eliminations, and tree-based feature ranking to select key features. For regression, we adopted Multiple Linear Regression, Lasso Regression, and XGBoost to predict numeric shares. We used Logistic Regression, Decision Trees and various Ensemble methods for binary and multi-class classification tasks. Classifiers are fine tuned with grid search and cross validated to improve performances. We aspire in the future to preprocess raw data and incorporate more recent articles to adapt the model better on recently published articles.
tanghengyu / ds_ga1001finalproject Goto Github PK
View Code? Open in Web Editor NEWGroup 4Two