Predicting user interaction behavior on government microblogs: A machine learning approach

Abstract

Nowadays, the Chinese government usually publishes policies via social media, where everyone can respond to the messages. Exploring the interaction behaviors of millions of users can help the government collect the users' opinions and make decisions. We use the machine learning algorithm Multilayer Perceptron Network model to predict the user's interactive behaviors under their comments/replies. We found that three kinds of features are useful in predicting user interaction behaviors: The Doc2vec word vector features, the user attributes, and the user history posting information. The experiments show the effectiveness of the neural network-based model, which provides a way to optimize the formulation and implementation of public policies.