首页> 外国专利> SUPERVISED MACHINE LEARNING DEVICE, SUPERVISED MACHINE LEARNING METHOD, AND SUPERVISED MACHINE LEARNING PROGRAM

SUPERVISED MACHINE LEARNING DEVICE, SUPERVISED MACHINE LEARNING METHOD, AND SUPERVISED MACHINE LEARNING PROGRAM

机译:监督的机器学习设备,监督的机器学习方法和监督的机器学习程序

摘要

An objective of the present invention is, when boundaries of correct answers of a plurality of facts which are appended adjacently thereto differ from boundaries of morphemes which are a result of a morpheme analysis, to efficaciously utilize these correct answers in machine learning. Provided is a supervised machine learning device which extracts facts from test data with respect to correct answers which are appended to training data, comprising: a learning unit which, on the basis of feature values for each morpheme which configures the training data and first new training data, generates models for extracting facts from the training data and extracting first new facts from first new training data; and a generating unit which, on the basis of the models and relations between the facts which are identified by predetermined labels, extracts from the training data first new labels and the first new facts which are identified by the first new labels, and generates the first new training data which includes the extracted first new facts.
机译:本发明的目的是,当在其附近附加的多个事实的正确答案的边界不同于作为词素分析结果的词素的边界时,在机器学习中有效地利用这些正确答案。提供了一种受监督的机器学习设备,该设备从测试数据中提取关于正确答案的事实,该正确答案被附加到训练数据上,包括:学习单元,其基于每个词素的特征值来配置训练数据和首次新训练数据,生成用于从训练数据中提取事实并从第一新训练数据中提取第一新事实的模型;生成单元,其基于由预定标签识别的事实之间的模型和关系,从训练数据中提取第一新标签和由第一新标签识别的第一新事实,并生成第一新的训练数据,其中包括提取的第一个新事实。

著录项

  • 公开/公告号WO2015083293A1

    专利类型

  • 公开/公告日2015-06-11

    原文格式PDF

  • 申请/专利权人 HITACHI LTD.;

    申请/专利号WO2013JP82870

  • 发明设计人 ASANO YU;IWAYAMA MAKOTO;

    申请日2013-12-06

  • 分类号G06F17/27;

  • 国家 WO

  • 入库时间 2022-08-21 15:06:07

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号