Incorporating Neighborhood Information and Sentence Embedding Similarity into a Repost Prediction Model in Social Media Networks

Abstract

Predicting repost behaviors within social media networks plays an important role in human activities analysis and influence maximization decision making. Traditional methods for repost prediction can be categorized into stochastic diffusion based models and user profile or content features based machine learning models. In this paper, we propose a new framework combining user profile, content similarity and the neighborhood information around each target link as input features to make the prediction. Here neighborhood information can be interpreted as the combination of neighbors’ user profile. Two different kinds of graph based combination models are introduced in the article. After collecting the input features, we implement the state-of-the-art machine learning methods, e.g., Logistic Regression, K-nearest Neighbors, Gaussian Naive Bayes, Deep Neural Network, Random Forest, XGBoosting and Stacking Model to predict repost probability. We evaluate our model on real dataset Weibo to compare the performance with different features and machine learning methods.

Appendices

Appendix

A Prediction Performance Measures

Measure	Definition	Formula
Accuracy	The ratio of correctly predicted observations to the total observations	$\frac{\left( TP+TN\right) }{\left( TP+FP+FN+TN\right) }$
Precision	The ratio of correctly predicted positive observations to the total predicted positive observations	$\frac{TP}{\left( TP+FP\right) }$
Recall	The ratio of correctly predicted positive observations to all the observations in actual class	$\frac{TP}{\left( TP+FN\right) }$
F1 Score	The weighted average of Precision and Recall	$\frac{2 \cdot Precision\cdot Recall}{\left( Precision+Recall\right) }$
ROCAUC	Compute area under the receiver operating characteristic curve which is True Positive Rate against False Positive Rate curve from prediction scores	–