Research on short text information mining and classification methods for social media

The internet era has brought massive information output and dissemination, and social media, mainly represented by WeChat and Weibo, have gradually become an important part of people's daily life. As the data of short texts generated by social media are growing, how to extract and classify useful information from these texts has become a pressing problem. The study designs a co-occurrence information model to build a graph structure of short texts and classifies them by combining a graph convolutional network and introducing an attention mechanism. The outcomes demonstrate that the precision of the upgraded model is 82.94% and 90.03% in the datasets MR and HR, respectively, with better classification outcomes. The precision of the model is basically stable at 80% and above, up to 90% under the change of training data size. The error rate is only 8.66% and the time required is 29.85% in the classification of short textbooks in microblogging platform. The precision and operational efficiency provide a new technical and methodological reference for the information processing of social media.