After analysing the defects of single MU (most uncertainty)sampling,we put forward a GOS (global optimum search)method and combines MU method with it to jointly implement the query selection.In GOS +MU method,GOS focuses on searching the object globally,under the conditions of limited training samples provided by the application environment and insufficient classifier training,the object selected by this method has high learning value and can fast promote the learning process of classifier;and MU can selects the samples with most uncertainty to supplement training set using current training outcomes of classifier when the GOS fails in sampling.By the simulation on classifying users’reviews on networks products and comparing the effects of other sampling learning methods,the effectiveness of GOS+MU method in compressing the learning cost and improving the training efficiency has been proved.%在分析单一MU （Most Uncertainty）采样缺陷的基础上，提出一种“全局最优搜寻”方法GOS（Global Optimal Search），并结合MU共同完成查询选择。GOS＋MU方法中，GOS着眼全局寻找目标，在应用环境能提供的训练样本数量有限、分类器受训不充分时，该方法选择的对象学习价值高，能快速推进分类器学习进程；MU则能够在GOS采样失效情形下，利用分类器当前训练成果，选择查询不确定性最强的样本补充训练集。通过对网络商品的用户评论进行分类仿真，并比较其他采样学习方法的效果，证明了GOS＋MU方法在压缩学习成本、提高训练效率方面的有效性。
展开▼
