# Incorporating topical stance into signed bipartite networks for user retweet prediction

## TL;DR

This paper improves retweet prediction by combining user-topic networks and stance analysis using a new neural network method.

## Contribution

The novel Topic Stance-integrated Graph Attention Neural Network (TSGAT) enhances retweet prediction accuracy by incorporating topic stance and network structure.

## Key findings

- The proposed TSGAT outperforms benchmarks like SGCN in predicting retweet behavior.
- Incorporating topic stance and network structure improves prediction accuracy.
- The method effectively captures user characteristics and emotional polarity.

## Abstract

Social networks accelerate information dissemination, and retweet behavior is an important way of user interaction. User retweet prediction analyzes user characteristics to predict retweet behavior and emotional polarity, which can help platforms understand user emotional tendencies and can be applied to scenarios such as public opinion analysis. However,most existing studies on tree like propagation chains focus on link existence and rarely combine positive negative polarity with topic semantics. This paper constructs a signed bipartite network using user and topic nodes, proposes a Topic Node weighting-based Topical Stance Representation (TNTSR) method, and develops a Topic Stance-integrated Graph Attention Neural Network (TSGAT) for retweet prediction. Experiments on social platform datasets show that both methods outperform benchmarks like SGCN in predicting retweet behavior and polarity. This research effectively leverages topic stance and network structure, enhancing the accuracy of user retweet prediction.