一种基于语义增强和指导路由机制的方面级情感三元组抽取方法

目前，细粒度情感分析已在观点挖掘、文本过滤等域获得广泛应用，通过细粒度情感分析，能完成更精准的文本理解和结果判断.其中，包含方面、观点和情感极性的情感三元组抽取任务是一个具有代表性的细粒度情感分析任务，且大多数相关研究是基于管道模型和端到端模型开展的.然而，一方面，管道模型本质为两阶段模型，存在错误传播的问题；另一方面，端到端模型也无法充分利用句子中各组成之间的联系，存在高层次语义关系捕获能力欠缺的问题.为解决以上问题，本文对句法和语义知识进行特征补充，提出一个基于语义增强和指导路由机制的情感三元组抽取方法(ASTE-SEGRM).首先，基于键值对网络学习源文本的句法特征和词性特征.区别于以往的建模方式，本文所提方法动态捕捉不同句法及词性类型的重要程度，并赋予不同的权重，以实现语义增强；其次，受启发于迭代路由机制，引入指导路由机制构建神经网络，使用先验知识指导情感三元组的抽取；最后，在四个基准数据集上的实验结果证明，本文所提方法优于数个基线模型.

An aspect sentiment triplet extraction method based on semantic enhancement and guided routing mechanism

Fine-grained sentiment analysis is widely used in fields such as opinion mining and text filtering to achieve more accurate text understanding and result determination. The Aspect Sentiment Triplet Extraction (ASTE) task is a representative fine-grained sentiment analysis task, and most of the related research is based on either the pipeline model or end-to-end model. However, the pipeline model suffers from error propagation as a two-stage model, and the end-to-end model does not make full use of the connections between the constituents in a sentence and lacks the ability to capture high-level semantic relations. To address the these issues, this paper features complementary syntactic and semantic knowledge and proposes a sentiment triplet extraction method based on semantic enhancement and guided routing mechanisms (ASTE-SEGRM). Firstly, the syntactic features and lexical features of the source text are learned based on Key-Value Pair Neural Network (KVMN). Secondly, inspired by iterative routing mechanism, a guided routing mechanism is introduced to build a neural network that uses a priori knowledge to guide the extraction of sentiment triplets. Finally, experimental results on four benchmark datasets demonstrate that the proposed approach outperforms several baseline models.