基于双向循环神经网络的评价对象抽取研究  

Bidirectional Recurrent Neural Network for opinion target extraction

作　　者：陈鹏[1] 徐冰[1] 杨沐昀[1] 李生[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2017年第1期71-75,共5页Intelligent Computer and Applications

基　　金：国家自然科学基金(61402134);中央高校基本科研业务费专项资金资助(HIT.NSRIF.2015068)

摘　　要：评价对象抽取的研究难点在于如何精确地表示大范围的上下文信息。本文针对微博观点句,采用了基于双向循环神经网络(BRNN)的方法来抽取评价对象并对评价对象的情感倾向进行判定。BRNN的隐藏层对上下文进行了抽象,如果经过良好地训练,就能在循环处理句子时有效地表示远距离的有序上下文信息,而无需对上下文窗口长度进行限定。本文选择了词、词性、依存句法树以及产品词典等特征构建了BRNN模型。通过实验发现,上述4种特征组合获得了最优实验结果,通过与CRF模型的对比,本文提出的方法在相互覆盖模式下F值比CRF模型高出0.61%,验证了本文方法的有效性。本文方法在COAE2015任务3的资源受限评测任务中,获得了最好结果。The challenge of opinion target extraction is how to represent wide-range context accurately. The paper focused on opinion sentences from microblog and employed a method based on Bidirectional Recurrent Neural Network （ BRNN） to extract opinion targets and judge their emotion tendency. The hidden layer of BRNN can abstract context. If the model is trained well, it can represent wide - range, ordered context effectively without the limitation of the length of the context window. The paper chooses the word, the part of speech, the dependency parsing and the product dictionary as features to build the BRNN model. The results of experiment show that the combination of four features mentioned above achieves the best performance. Compared with CRF model, the F value of overlap mode increased by 0.61% using the method proposed in this paper, which validates the method is effective. The method proposed in this paper gets the best result in COAE2015 task3 with limited resources.

关 键 词：情感分析 评价对象抽取 双向循环神经网络模型 条件随机场模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]