金融领域机器阅读理解模型

Financial Reading Machine Reading Comprehension Model

近些年机器阅读理解相关的研究越来越多,但只针对某一领域的研究相对较少,因此通过网络爬虫获取数据,在金融领域上提出一种多重注意力机制的端到端的抽取式机器阅读理解模型,该模型的主要特点是先对数据重构,然后对融合后的文档作自注意力机制(self-attention),加深问题与文章的关联,突出文章中与问题关联较深的特征,最后联合多篇文章再作自注意力机制突出文章间的关联性,结果表明该模型提高训练效率BLEU-4和Rouge-L分别为38.37、44.65。

In recent years,more and more researches related to machine reading comprehension have been conducted,but relatively few studies have focused on a certain field.Therefore,data obtained through web crawlers has proposed an end-to-end multi-attention mechanism in the fi⁃nancial field.Extractive machine reading comprehension model.The main feature of the model is to reconstruct the data first,then to make self-attention mechanism for the fused documents to deepen the relationship between the problem and the article,and to highlight the deep⁃er relationship between the article and the problem.Finally,to combine multiple articles and then make self-attention mechanism to high⁃light the relationship between the articles.The results show that the model improves the training efficiency.The BLEU-4 is 38.37 and the Rouge-L is 44.65.