基于词向量的评价搭配抽取算法研究 | |
杨令铎; 史海波; 周晓锋 | |
刊名 | 小型微型计算机系统 |
2016 | |
卷号 | 37期号:10页码:2269-2272 |
关键词 | 搭配抽取 词向量 神经网络 条件随机域 最大熵 |
ISSN号 | 1000-1220 |
其他题名 | Research on the Algorithm of Evaluation Collocation Extraction Based on Word Vector |
通讯作者 | 杨令铎 |
产权排序 | 1 |
中文摘要 | 传统中文评价搭配抽取采用的最大熵和条件随机域等算法依赖于人工选取特征,且对前期语义标注精度要求较高.本文提出一种使用词向量代替传统语义特征进行搭配抽取的方法.其中词向量通过深度学习模型在大规模语料上进行无监督学习得到.实验中将词向量及语义特征分别作为三种机器学习模型的输入,结果表明使用词向量在神经网络模型中取得了较好的效果,其精度、召回率都比使用语义特征最好情况高出接近3%,同时,我们发现随着无监督学习训练语料的增大,得到的词向量也越来越实用. |
英文摘要 | Maximum entropy and conditional random field or other algorithms used for collocation extraction in the traditional assessment of Chinese language rely on manual selection of characteristics and have a high demand for semantics marking precision at the preliminary stage. In this paper,an alternative approach is suggested which substitutes term vector for the traditional semantic characteristics in collocation extracting. Specifically,the term vectors are acquired by an in-depth model completing unsupervised learning from a large corpus. In testing,the term vectors and the semantic characteristics are separately entered as inputs into three machine learning models. The results indicate that better outcomes are produced when term vectors are used in the neural network model in the sense that both the precision and recall rate are higher by nearly 3% than the best outcomes that are achievable with semantic characteristics. We also note that as the size of the corpus used for unsupervised learning training increases the resulting term vectors become more and more pragmatic. |
收录类别 | CSCD |
语种 | 中文 |
CSCD记录号 | CSCD:5833022 |
内容类型 | 期刊论文 |
源URL | [http://ir.sia.cn/handle/173321/19398] |
专题 | 沈阳自动化研究所_数字工厂研究室 |
推荐引用方式 GB/T 7714 | 杨令铎,史海波,周晓锋. 基于词向量的评价搭配抽取算法研究[J]. 小型微型计算机系统,2016,37(10):2269-2272. |
APA | 杨令铎,史海波,&周晓锋.(2016).基于词向量的评价搭配抽取算法研究.小型微型计算机系统,37(10),2269-2272. |
MLA | 杨令铎,et al."基于词向量的评价搭配抽取算法研究".小型微型计算机系统 37.10(2016):2269-2272. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论