基于语义依存分析的图网络文本分类模型

针对目前已有的文本分类方法未考虑文本内部词之间的语义依存信息而需要大量训练数据的问题，提出基于语义依存分析的图网络文本分类模型TextSGN。首先对文本进行语义依存分析，对语义依存关系图中的节点（单个词）和边（依存关系）进行词嵌入和one-hot编码；在此基础上，为了对语义依存关系进行快速挖掘，提出一个SGN网络块，通过从结构层面定义信息传递的方式来对图中的节点和边进行更新，从而快速地挖掘语义依存信息，使得网络更快地收敛。在多组公开数据集上训练分类模型并进行分类测试，结果表明，TextSGN模型在短文本分类上的准确率达到95.2%，较次优分类法效果提升了3.6%。

Text classification model with graph network based on semantic dependency parsing

Due to the problem of the existing text classification methods which left out the semantic dependency information between words and required a lot of training data, this paper proposed a graph network text classification model TextSGN based on semantic dependency parsing. The model first performed semantic dependency parsing on the text, then in the semantic dependency graph, it performed word embedding and one-hot encoding on the nodes(single words) and edges(dependencies). In a further step, this paper proposed a SGN block to mine the semantic dependencies rapidly. The SGN block defined the way of information transmission from the structure level, updated the nodes and the edges in the graph to mine the semantic dependencies quickly and make the network converge faster. The experimental results on a set of open datasets show that Text-SGN achieves 95.2% accuracy in short text classification, which is 3.6% higher than the sub-optimal classification.