基于变种概率图模型的文本生成算法

Text generation algorithm based on variant probability graph model

针对文本自动生成问题中通过抽象摘要生成文本的需求特点,提出了一种多层概率图模型用于文本生成。通过多层次摘要等方法使得其能从更高层次理解语义需求,逐层传播信度进行文本生成,并通过线性规划进行不同层间连接权值的训练。对其进行了系统实现后的实验结果表明,所建立的计算模型实现了根据关键词生成文章的目标,效果上显著优于基于seq2seq的文本生成模型,可以用于"概念-文本"生成的实际应用场景。最后,对该模型进一步改进和与现有其他机器学习模型融合的方法进行了展望。

Aiming at the demand characteristics of text generated by abstract abstract in automatic text generation problem, a multi-layer probability graph model for text generation was presented. By means of muhi-level summarization, the semantic requirement could be understood from a higher level, and the text was generated by propagatiing trust degree layer by layer. The weights of different layers were trained by the linear programming. The experimental results of the system show that the computational model achieves the goal of generating the article according to the key words, which is significantly better than that of the text generation model based on seq2seq. So the model can be used for the actual scenario of concept-text generation. Finally, the further improvement of the model and the fusion with other machine learning models were prospected.