Abstract
Stock price prediction is an important research topic in the financial field. However, there are two challenges, modeling the graph-structured stock comments effectively and tackling the severe distribution shift of the stock price data. To exploit the structural characteristics of stock comments and address the issue of data distribution shift, this study proposes a novel model, CommentGCN-OPERA, for stock price prediction. In terms of the social media textual data, to effectively process the inherently graph-structured stock comments and their interrelationships, we propose the CommentGCN module for modeling the information contained in these comments. Regarding the stock price data, the OPERA (OPEning pRice normAlization) module alleviates the issue of data distribution shift by employing input normalization and output denormalization operations, thereby reducing the bias of the stock price data. Experimental results on 6 stocks demonstrate that CommentGCN-OPERA outperforms 5 baseline models in most cases. Additionally, the ablation study further validates the independent effectiveness of the CommentGCN and OPERA modules within the model.