Domain-Informed Negative Sampling Strategies for Dynamic Graph Embedding in Meme Stock-Related Social Networks

TL;DR

This paper introduces novel negative sampling strategies tailored for dynamic graph embedding in meme stock-related social networks, improving the modeling and analysis of complex social interactions influencing stock movements.

Contribution

The paper proposes new negative sampling methods based on real social network analysis and financial knowledge, specifically designed for meme stock social networks.

Findings

Proposed strategies outperform existing baselines in model evaluation.

Enhanced training of dynamic graph embedding models for meme stock networks.

Better capture of complex interaction patterns in social networks.

Abstract

Social network platforms like Reddit are increasingly impacting real-world economics. Meme stocks are a recent phenomena where price movements are driven by retail investors organizing themselves via social networks. To study the impact of social networks on meme stocks, the first step is to analyze these networks. Going forward, predicting meme stocks' returns would require to predict dynamic interactions first. This is different from conventional link prediction, frequently applied in e.g. recommendation systems. For this task, it is essential to predict more complex interaction dynamics, such as the exact timing. These are crucial for linking the network to meme stock price movements. Dynamic graph embedding (DGE) has recently emerged as a promising approach for modeling dynamic graph-structured data. However, current negative sampling strategies, an important component of DGE, are designed for conventional dynamic link prediction and do not capture the specific patterns present in meme stock-related social networks. This limits the training and evaluation of DGE models in such social networks. To overcome this drawback, we propose novel negative sampling strategies based on the analysis of real meme stock-related social networks and financial knowledge. Our experiments show that the proposed negative sampling strategies can better evaluate and train DGE models targeted at meme stock-related social networks compared to existing baselines.