Abstract: Aiming at the problem of fine-grained sentiment analysis in the direction of Natural Language Processing (NLP), in order to explore the influence of Pre-trained Language Models (PLM) with structural biases on the end-to-end sentiment triple extraction task, and solve the problem of low fault tolerance rate of aspect semantic feature dependence that is common in previous studies. Combining aspect-aware attention mechanism and Graph Convolutional Network (GCN), the Aspect-aware attention Enhanced Graph Convolutional Network (AE-GCN) model was proposed for aspect sentiment triple extraction task. Firstly, multiple types of relations were introduced for the aspect sentiment triple extraction task. Then, these relations were embedded into the adjacent tensor between words in the sentence by using the double affine attention mechanism. At the same time, the aspect-aware attention mechanism was introduced to obtain the sentence attention scoring matrix, and the aspect-related semantic features were further mined. Next, sentence was converted into a multi-channel graph through the graph convolutional neural network, to learn a relation-aware node representation by treating words and relation adjacent tensors as edges and nodes, respectively. Finally, an effective word pair representation refinement strategy was used to determine whether a word pair matches, which was used to consider the implicit results of aspect and opinion extraction. Experimental results on the ASTE D1 reference dataset show that F1 values of the model on the 14res, 14lap, 15res and 16res sub-datasets are improved by 0.2, 0.21, 1.25 and 0.26 percentage points compared with the Enhanced Multi-channel Graph Convolutional Network (EMC-GCN) model. Experimental results on the ASTE-D2 benchmark dataset show that the F1 value of the proposed model on the 14lap, 15res and 16res sub-datasets are increased by 0.42, 0.31 and 2.01 percentage points compared with the EMC-GCN model. A large number of experimental results on benchmark datasets show that the proposed model has great improvement in accuracy and effectiveness compared with the EMC-GCN model.

Key words: Natural Language Processing (NLP), sentiment analysis, sentiment triplet extraction, aspect-aware attention, Graph Convolutional Network (GCN)

摘要： 针对自然语言处理(NLP)方向的细粒度情感分析问题，为了探索携带结构偏差的预训练语言模型(PLM)对端到端式情感三元组抽取任务的影响，解决之前研究中普遍存在的方面语义特征依赖容错率低的问题，结合方面感知注意力机制和图卷积网络(GCN)提出了用于方面情感三元组抽取任务的方面感知注意增强图卷积网络(AE-GCN)模型。首先，在方面情感三元组抽取任务中引入多种类型的关系；然后，采用双仿射注意力机制将这些关系嵌入到句子中单词之间的相邻张量中，并在此引入了方面感知注意力机制以获取句子注意评分矩阵，深入挖掘与方面相关的语义特征；接着，图卷积神经网络通过将单词和关系相邻张量分别视为边和节点，将句子转换为多通道图以学习关系感知节点表示；最后，使用一种有效的词对表示细化策略确定词对是否匹配，以考虑方面和意见抽取的隐含结果。在ASTE-D1基准数据集上的实验结果表明，所提模型在14res、14lap、15res和16res子数据集上的F1值相较于增强型多通道图卷积网络（EMC-GCN）模型提升了0.2、0.21、1.25和0.26个百分点；在ASTE-D2基准数据集上的实验结果表明，所提模型在14lap、15res和16res子数据集上的F1值相比于EMC-GCN模型提升了0.42、0.31和2.01个百分点。在基准数据集上的大量实验结果表明，所提模型相较于EMC-GCN模型在准确率和有效性方面有了较大改进。

关键词: 自然语言处理, 情感分析, 情感三元组抽取, 方面感知注意力, 图卷积网络