基于松散条件下协同学习的中文微博情感分析

传统的协同学习算法需要2个充分冗余的特征视图，而在多数情况下达不到特征充分冗余的要求，为此提出松散条件下的协同学习框架.利用支持向量机算法和长短期记忆网络（LSTM）算法分别建立基于向量空间模型的微博特征视图和基于语义相关的词向量特征视图，在2个视图上进行协同学习.针对未标注样本的选择，提出结合主动学习中的不确定策略和协同学习中的最高置信度策略的选择策略，从不同角度充分利用未标注样本中包含的信息量.实验结果表明，在中文微博情感极性研究领域，提出的选择策略与传统选择策略相比，能够提高分类器的性能，并且利用松散条件下的协同学习框架实现微博情感分析性能.

Microblog sentiment analysis based on collaborative learning under loose conditions

Aiming at the facts that two completely redundant feature views are required in the traditional collaborative learning and the redundancy of features are not reached in most cases, a collaborative learning framework under loose conditions was proposed. The support vector machine algorithm and the long short-term memory algorithm were used to establish the microblog feature view based on the vector model and the word vector model. The collaborative learning was conducted on these two models. A new selection strategy of unmarked samples which combined the uncertain strategy in active learning and the maximum certainty-factor was proposed. The information contained in unlabeled samples was fully used. The experimental results show that compared with the traditional selection strategy, the selection strategy improves the quality of categorizer and manages to complete Chinese microblog sentiment analysis with the proposed collaborative learning framework under loose conditions.