基于跨语言图神经网络模型的属性级情感分类

目前,在属性级情感分类任务上较为成熟的有标注数据集均为英文数据集,而有标注的中文数据集较少.为了能够更好地利用规模庞大但却缺乏成熟标注数据的中文语言数据集,针对跨语言属性级情感分类任务进行了研究.在跨语言属性级情感分类中,一个核心问题为如何构建不同语言的文本之间的联系.针对该问题,在传统的单语言情感分类模型的基础上,使用图神经网络模型对跨语言词-词、词-句之间的关系信息进行建模,从而有效地刻画两种语言数据集之间的联系.通过构建单语词-句之间的联系和双语词-句之间的联系,将不同语言的文本关联起来,并利用图神经网络进行建模,从而实现利用英文数据集预测中文数据集的跨语言神经网络模型.实验结果表明：相较于其他基线模型,所提出的模型在F1值指标上有着较大的提升,从而说明使用图神经网络建立的模型能够有效地应用于跨语言的应用场.

Cross-lingual Aspect-level Sentiment Classification with Graph Neural Network

Most of the mature labeled dataset of aspect-level sentiment analysis are in English, it is quite rare in some low-resource language such as Chinese. For the sake of utilizing the vast but unlabeled Chinese aspect-level sentiment classification dataset, this study works on cross-lingual aspect-level sentiment classification. Nevertheless, the most central and difficult problem in cross-lingual mission is how to construct the connection between the documents in two languages. In order to solve this problem, this study proposes a method using graph neural network structure to model the connection of multilingual word-to-document and word-to-word, which could effectively model the interaction between the high-resource language (source language) and low-resource language (target language). The connections include multilingual word-to-document connection and monolingual word-to-document connection are constructed to tie the source language data and target language data, which are modeled by graph neural network to realize using English labeled dataset as trainset to predict Chinese dataset. Compared with other baseline model, the proposed model achieves a higher performance in F1-score, which indicates that the presented work does contributing to the cross-lingual aspect-level sentiment classification.