基于神经网络的吟诗作对技术研究与应用摘 要 自古以来,中华民族每年都有吟诗作对的传统习俗,是中国的传统文化。在吟诗作对时讲究平仄要协调,对仗要工整,朗诵起来要有很强的美感,被很多人所喜爱。如何在运用机器语言来实现这一特殊的语言形式受到了广泛的关注。因此本文便是自然语言处理方面的技术,在RNN的基础上进行“吟诗作对”的研究与应用,设计与实现了一个根据指定序列生成目标序列的模型。因此根据实际需求是一项序列的自动生成工作。通过给定的输入序列,生成输出序列。在这种情况下,有人提出了一种编码和解码的模型。编码就是将文字以根据实际情况用相等长度的向量表示,解码就是将编码的序列再进行转化。本文采用双向循环神经网络进行编码,采用循环神经网络进行解码,在进行词向量处理的时候也要将上下文的关系考虑进去,同时考虑上下联对应位置的字的相关性,因此在解码阶段采用了注意力机制来完成这一规则。同时为了防止神经网络出现过拟合的情况在实验过程中加入Dropout机制,提高生成模型的效果。最后通过用静态页面的方法,输入上联经过模型自动生成下联。本文经过RNN的模型在实际应用中得到了很好的效果,自动生成的下联可以与上联的格式相同、文字相对应。关键词: 循环神经网络;对联生成;dropout;注意力机制The method and application of reciting poems and composing pairs based on Neural NetworkabstractSince ancient times, the Chinese nation has the traditional custom of reciting poems and composing right every year, which is the traditional culture of China. When reciting poems, we should pay attention to the coordination of the level and the tone, the alignment should be neat, and the recitation should have a strong aesthetic feeling, which is loved by many people. How to use machine language to realize this special language form has been widely concerned. Therefore, this paper is the technology of natural language processing. On the basis of RNN, the research and application of "reciting poems and making pairs" is carried out, and a model of generating target sequences according to specified sequence...