Abstract
Social media generates vast amounts of spatio-temporal sequential data. However, current methods often ignore the complex spatio-temporal correlations within these data. This oversight makes it difficult to fully capture the dynamic features of the data. To delve deeply into the concealed feature attributes within timely spatio-temporal sequence data from social media, this study introduces a Spatio-Temporal Graph Wavelet Neural Network (ST-GWNN). This model captures spatio-temporal correlations across time and space by combining spatial graphs from multiple time intervals. On this basis, we have developed a spatial feature extraction layer using the Graph Wavelet Neural Network (GWNN). This layer learns localized representations of node features to identify spatial dependencies. In GWNN, graph wavelet transformation reduces computational complexity and improves operational efficiency compared to Spectral CNN. Furthermore, the sparse representation of node features is enhanced via localized learning, thereby improving network performance. The effectiveness of the model is verified using four distinct social media datasets. Experimental results underscore the notable advantages of the proposed model in the realm of timely time-series data association mining, showcasing its capacity to better capture spatio-temporal dynamics and uncover the underlying association mining within the data. In comparison to alternative models, the approach outlined in this paper exhibits substantial improvements in terms of accuracy and efficiency, affirming the efficacy and innovation of the model.