Abstract: A short text multi-label classification algorithm combining similarity graph and random walk model is proposed. Firstly, the sample data and labels are used as nodes to create a similarity graph, and the weight between the sample and the label is calculated with the help of an external know- ledge base to obtain the matching degree between the predicted sample and the label set. Secondly, the multi-label data are mapped into a multi-label dependency graph. A random walk is performed on the graph, and the previous matching degree is used as the initial prediction value to calculate the probability distribution of each node. When the probability distribution tends to be stable, the probability distribution of the node is the probability distribution of the label, and then the label set of the predicted text is determined. The experimental results show that the proposed method achieves better performance in the classification of multi-label texts. Compared with similar algorithms, the classification performance is significantly improved.

Key words: multi-label short text classification, similarity graph, restart random walk, WordNet