| 中文医学摘要主题建模方法评估 |
| |
| 引用本文： | 王凡,夏晨曦.中文医学摘要主题建模方法评估[J].医学信息学杂志,2018,39(2):60-64. |
| |
| 作者姓名： | 王凡 夏晨曦 |
| |
| 作者单位： | 华中科技大学同济医学院医药卫生管理学院 武汉 430030,华中科技大学同济医学院医药卫生管理学院 武汉 430030 |
| |
| 基金项目： | 中央高校基本科研业务费华中科技大学青年创新基金项目"面向社交网络的情感分析与观点挖掘方法研究"（项目编号：2015AC0 28）。 |
| |
| 摘 要： | 以中文医学论文摘要为语料,采用3种特征筛选方法以及两种主题模型方法的组合进行主题建模,结果表明建模方法中LDA的预测能力和拟合度优于CTM,而特征筛选方式中IDF拟合度、TF预测能力更好。
|
| 关 键 词： | 主题模型 评估 中文 医学摘要 |
| 修稿时间： | 2017/10/16 0:00:00 |
Evaluation of Topic Modeling Methods in Chinese Medical Abstracts |
| |
| WANG Fan and XIA Chen-xi.Evaluation of Topic Modeling Methods in Chinese Medical Abstracts[J].Journal of Medical Informatics,2018,39(2):60-64. |
| |
| Authors: | WANG Fan and XIA Chen-xi |
| |
| Institution: | School of Medical and Health Management of Tongji Medical College, HUST, Wuhan 430030, China and School of Medical and Health Management of Tongji Medical College, HUST, Wuhan 430030, China |
| |
| Abstract: | Taking abstracts of Chinese medical papers as corpus, the paper adopts the combination of 3 feature selection methods and 2 topic modeling methods to carry out topic modeling. The result shows that in the modeling methods, prediction ability and fitting degree of LDA are better than that of CTM, but in the feature selection methods, fitting degree of IDF and prediction ability of TF are better. |
| |
| Keywords: | Topic model Evaluation Chinese Medical abstract |
|
| 点击此处可从《医学信息学杂志》浏览原始摘要信息 |
| 点击此处可从《医学信息学杂志》下载免费的PDF全文 |
