Text generator training method and text generator training system

• Main Author: LI JIAFAN
• Format: Patent
• Language: Chinese; English
• Published: 23.08.2019
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a text generator training method and a text generator training system. The method comprises the following steps: carrying out random Gaussian initialization on a hidden layer state vector; reading the hidden layer state vector and the initial character as first input of the text generator; generating a first character through a polynomial sampling method as a first output ofthe text generator; and taking the character with the generation probability meeting a preset condition as the next output of the text generator until an end character is generated. According to theinvention, random Gaussian initialization is carried out on the hidden layer state vector, so that a text generator can be effectively prevented from generating repeated texts, and the diversity of generated samples is improved. 本发明提供一种文本生成器训练方法和文本生成器训练系统，其中方法包括：对隐层状态向量进行随机高斯初始化；读入所述隐层状态向量和起始字符，作为所述文本生成器的第一输入；通过多项式采样方法生成第一个字符，作为所述文本生成器的第一输出；将生成概率满足预设条件的字符作为所述文本生成器的下一输出，直至生成结束字符。本发明中，通过对隐层状态向量进行随机高斯初始化，可以有效避免文本生成器生成重复的文本，提高