首页> 外国专利> ABSTRACT GENERATION DEVICE, ABSTRACT GENERATION METHOD, AND COMPUTER PROGRAM

ABSTRACT GENERATION DEVICE, ABSTRACT GENERATION METHOD, AND COMPUTER PROGRAM

机译：抽象生成设备，抽象生成方法和计算机程序

页面导航

摘要
著录项
相似文献

摘要

[Problem] To provide a technology for automatically detecting, with high precision, a specified description of interest within a text, and for automatically abstracting the text with a focus on the detected description. [Solution] Provided is an abstract generation device comprising: a text storage device for storing text together with information indicating sections of interest thereof; word vector conversion units 834 and 836 for vectorizing each word of the text, adding to the vectors elements indicating whether the words are the sections of interest, and converting the text to a word vector sequence; LSTMs composed of neural networks carrying out sequence-to-sequence conversions and pre-learned by machine learning to output, in response to each of the word vectors of the word vector sequence being inputted in a prescribed sequence, an abstract of the text composed of the words represented by the word vector sequence; and input units 838–848 for inputting into the neural networks in the prescribed sequence each of the word vectors of the word vector sequence.

机译：[问题]提供一种技术，该技术可以高精度地自动检测文本中的特定关注描述，并以检测到的描述为重点自动抽象文本。 [解决方案]提供了一种抽象生成设备，包括：文本存储设备，用于将文本以及指示其感兴趣的部分的信息一起存储;词向量转换单元834和836，用于向量化文本的每个词，向向量添加指示词是否是感兴趣的部分的元素，并将文本转换成词向量序列; LSTM由神经网络组成，该神经网络执行序列到序列的转换，并通过机器学习进行预学习，以响应于按预定顺序输入单词向量序列的每个单词向量，输出由以下内容组成的文本摘要：单词向量序列表示的单词;以及输入单元838-848，用于以规定的顺序将每个单词矢量序列的单词矢量输入到神经网络。

著录项

公开/公告号WO2018207723A1

专利类型
公开/公告日2018-11-15

原文格式PDF
申请/专利权人 NATIONAL INSTITUTE OF INFORMATION AND COMMUNICATIONS TECHNOLOGY;
展开▼

申请/专利号WO2018JP17606
发明设计人 IIDA RYU;TORISAWA KENTARO;OH JONGHOON;KRUENGKRAI CANASAI;ASAO YOSHIHIKO;ABE NORIYUKI;MIZUNO JUNTA;KLOETZER JULIEN;
展开▼

申请日2018-05-07
分类号G06F17/30;G06F17/27;
国家 WO
入库时间 2022-08-21 11:58:06

相似文献

专利
外文文献
中文文献