基于可信标签扩展传递的跨领域倾向性分析*

针对传统监督分类方法不能很好地处理不同领域中服从不同分布的数据这一问题进行了研究，提出了一种基于可信标签扩展传递的半监督分类算法。情感种子词与目标领域待标注词之间按照相似度进行标签传递，将具有可信标签的词迭代移入情感种子词集实现扩展，结合目标领域词的先验情感分计算出最终情感分，从而有效地实现跨领域倾向性分析。实验表明，本文的方法能够大幅度提高跨领域情感分析的准确率。

Cross-domain opinion analysis based on the bootstrapping and propagation of trust-worthy labels

Due to relying too much on abundant labeled data from the training set, the traditional supervised classification methods could not perform well when processing imbalanced data from various domains. In order to solve this problem, this paper proposed a half-supervised classification algorithm based on the bootstrapping and propagation of trust-worthy labels by integrating the LPA algorithm and the concept of Bootstrapping. It conducted label propagation between the seed words and the unlabeled words from the target domain according to their similarities. Then it chose the words that were with trust-worthy labels to extend the set of seed words iteratively. After that, it further improved the sentiment scores of the words from the target domain by use of their prior scores. Experimental results indicate that the proposed algorithm could improve the performance of cross-domain opinion analysis dramatically.