Short Text Classification Combining Keywords and Knowledge

With the continuous development of the Internet, short text data are increasing. However, short text has poor classification performance due to sparse features. Nowadays, integrating external knowledge into the classification model has become one of the best solutions to the sparse problem of short text data, but this method is still insufficient. When the mapping result between knowledge and text is not enough, we choose keywords as content feature extensions to compensate for the lack of external knowledge. For the problem of noise generated by external knowledge. We design a context-based attention mechanism, which combines semantic information to make the model more focused on knowledge that has a more positive impact on classification. We conducted experiments on common data sets to verify the effectiveness of the model.