WVRF_CGT: WVRF Automatic Filter Based Cascading Graph Transformer for Information Diffusion in Stock Q&A Platform

Xiaobing Feng, Ling Feng, Pengfei Jiao, Zhiliang Zheng, Qing Bao

Abstract: This article takes the interactive platforms of the Shenzhen Stock Exchange and the Shanghai Stock Exchange as a vertical domain-specific dialogue platform. Firstly, it proposes an automated method for quality rating of Q&A pairs based on WordVectorRandForest (WVRF). By constructing a standard StockQ&APair Q&A pair database, extracting text features using Word2Vec and TF-IDF that the combination of Word2Vec and TF-IDF enhances the precision of Q&A filtering in financial platforms by effectively handling semantic relationships in text data, and combining statistical information such as response time and text length, a random forest model is trained to rate the quality of Q&A pairs. Subsequently, based on the foundation of building and validating a high-quality securities interactive platform Q&A automatic filtering system (AutoStockQ &AFilter), the study further introduces a Cascaded Graph Transformer (CGT) for automatic filtering of stock Q&A platform information diffusion. By evaluating the heat index of stocks using high-quality Q&A pairs in the platform, it provides real-time decision analysis tools for investment. Experimental results show that the average accuracy of this method exceeds 95%, and the precision in information dissemination meets the validation threshold.