现有方面级情感分析研究大多数往往从文本数据本身进行情感分析，而没有充分利用领域知识，忽略了语义依存信息的重要性，使得方面表示受噪声信息影响严重，出现噪声词注意权重高的可能。针对以上问题，结合领域知识，提出了一种剪枝算法和语义-注意力机制相结合的方法(Pruning And Semantic Attention，PASA)针对服务领域特定方面进行情感分类。方法一方面结合领域知识对文本对应的语义依存树进行剪枝实现方面信息降噪，另一方面，通过利用语义-注意力机制进行增强并精确捕获方面的上下文描述信息，从而实现对方面情感极性的判断。为了验证所提出方法的正确性和有效性，在物流数据集、酒店评论数据集及SemEval 2014 的Restaurant数据集进行了大量实验，结果表明，所提出的方法相对于其它方法具有明显优势，在垂直领域具有较好的应用前景。
Aspect-level Sentiment Classification Based on Semantic-Attention Mechanism
Most of the existing aspect-level sentiment analysis studies tend to conduct sentiment analysis from the text data itself,but fail to make full use of domain knowledge and ignore the importance of semantic dependency infor-mation,which makes aspect description be seriously affected by noise information and may lead to high attention weight of noise words.To solve this problem,we propose a novel approach Pruning And Semantic Attention(PASA)to analyze sentiment targeting specific aspects of service domain by combining attention mechanism and semantic de-pendency tree pruning algorithm.In PASA,the semantic dependency tree corresponding to the text is pruned by com-bining domain knowledge to decrease the number of noise words unrelated to the target on one hand.On the other hand,the sentiment polarity of the target object is judged by the context description information closely related to as-pect words which is enhanced and accurately captured with the semantic and attention mechanism.Extensive experi-ments were carried out three datasets,one private logistics dataset and two public datasets in the service domain inclu-ding a hotel comment dataset and a restaurant dataset SemEval 2014.The results show that the proposed method has obvious advantages over other methods and has a good application prospect in the vertical domain.
NETL
NSTL
万方数据
