一种利用近邻和信息熵的主动文本标注方法

由于大规模标注文本数据费时费力,利用少量标注样本和大量未标注样本的半监督文本分类发展迅速.在半监督文本分类中,少量标注样本主要用来初始化分类模型,其合理性将影响最终分类模型的性能.为了使标注样本尽可能吻合原始数据的分布,提出一种避开选择已标注样本的K近邻来抽取下一组候选标注样本的方法,使得分布在不同区域的样本有更多的标注机会.在此基础上,为了获得更多的类别信息,在候选标注样本中选择信息熵最大的样本作为最终的标注样本.真实文本数据上的实验表明了提出方法的有效性.

An Active Labeling Method for Text Data Based on Nearest Neighbor and Information Entropy

As it is quite time-consuming to label text documents on a large scale,a kind of text classification with a few labeled data is needed.Thus,semi-supervised text classification emerges and develops rapidly.Different from traditional classification,semi-supervised text classification only requires a small set of labeled data and a large set of unlabeled data to train a classifier.The small set of labeled data is used to initialize the classification model in most cases.Its rationality will affect the performance of the final classifier.In order to make the distribution of the labeled data more consistent with the distribution of the original data,a sampling method is proposed to avoid selecting the K nearest neighbors of the labeled data to be new candidate labeled data.With the help of this method,the data located in various regions will have more opportunities to be labeled.Moreover,in order to obtain more category information from the very few labeled data,this method compares the information entropy of the candidate labeled data and the datum with the highest information entropy is chosen as the next datum to be labeled manually.Experiments on real text data sets suggest that this approach is very effective.