Unleashing Expert Opinion from Social Media for Stock Prediction

TL;DR

This work tackles the noise in social-media signals for stock prediction by introducing a dynamic expert tracing system that identifies both true and inverse experts and filters out bots, paired with a Dual Graph Attention Network (DualGAT) to propagate sparse expert signals across inter-stock relationships. The approach transforms binary expert predictions into meaningful return signals, pre-trains a temporal baseline with MS-LSTM, and uses a dual-graph attention mechanism to fuse industry and correlation-based relationships, significantly improving predictive accuracy and trading metrics. Key contributions include an up-to-date StockTwits dataset, robust expert identification achieving up to ${\rm ACC}=72.8\%$ on short horizons, and a propagation framework that expands coverage to ~90% of stocks within two hops while synergizing with traditional financial features. The method demonstrates strong cross-market generalization and scalable efficiency, offering a practical path to leverage social-media expertise in quantitative investment strategies; code is available at the provided GitHub link.

Abstract

While stock prediction task traditionally relies on volume-price and fundamental data to predict the return ratio or price movement trend, sentiment factors derived from social media platforms such as StockTwits offer a complementary and useful source of real-time market information. However, we find that most social media posts, along with the public sentiment they reflect, provide limited value for trading predictions due to their noisy nature. To tackle this, we propose a novel dynamic expert tracing algorithm that filters out non-informative posts and identifies both true and inverse experts whose consistent predictions can serve as valuable trading signals. Our approach achieves significant improvements over existing expert identification methods in stock trend prediction. However, when using binary expert predictions to predict the return ratio, similar to all other expert identification methods, our approach faces a common challenge of signal sparsity with expert signals cover only about 4% of all stock-day combinations in our dataset. To address this challenge, we propose a dual graph attention neural network that effectively propagates expert signals across related stocks, enabling accurate prediction of return ratios and significantly increasing signal coverage. Empirical results show that our propagated expert-based signals not only exhibit strong predictive power independently but also work synergistically with traditional financial features. These combined signals significantly outperform representative baseline models in all quant-related metrics including predictive accuracy, return metrics, and correlation metrics, resulting in more robust investment strategies. We hope this work inspires further research into leveraging social media data for enhancing quantitative investment strategies. The code can be seen in https://github.com/wanyunzh/DualGAT.

Figures (3)

• Figure 1: Comparison of stock price movement prediction accuracy (%) achieved by experts identified through different expert identification methods versus naive sentiment aggregation methods. Accuracy is measured by how well subsequent sentiment predictions from identified experts align with actual stock price movements.
• Figure 2: An overview of our proposed method. We first identify experts and inverse experts from social media based on historical prediction performance across different market regimes, as shown in the left part illustrating prediction patterns of different user types (experts/inverse experts, bots, spammers, lucky users, mob; see Sec.IV). Then we identify and extract these expert signals which are sparse in nature. To address the sparsity issue, we propose a dual graph attention network (DualGAT) that takes both sparse expert signals and market and fundamental features as input. DualGAT incorporates relational information among stocks from multiple aspects to propagate expert signals across related stocks, expanding expert signal coverage and finally outputs the daily cross-sectional return predictions for each stock.
• Figure 3: The workflow of DualGAT. On the far left, it illustrates the construction process of both the correlation and industry graphs, where the orange nodes represent stocks with initial expert signals. As seen in the figure, DualGAT facilitates the effective propagation of expert signals across all stocks.