Seq2Seq中文文本摘要在金融知识引擎系统中的应用

Application of Chinese Text Summarization by Seq2Seq Model in Engine System of Finance Knowledge

针对一个基于中文文本摘要的金融知识引擎系统,研究了Seq2Seq模型在系统中的应用。首先构建Seq2Seq模型,将研报等重要的数据输入模型的Encoder端,从Decoder端输出摘要。在seq2seq模型中加入了Attention(注意力)机制,也就是在产生输出的时候,对关系较大的输入输出数据赋以较大权重,再根据关注的区域产生下一个输出。最后通过LawRouge评价器对生成的金融数据进行效果评价。

For an engine system of finance knowledge, the application of Seq2 Seq model in the system is researched. The summarization is extracted from the Decoder by importing important data such as research reports from the Encoder of Seq2 Seq model. In the model, Attention mechanism is provided to give a large relative data a larger weight. According this the output data is exported. At last the LawRouge is used to evaluate the summarization from the Seq2 Seq model.