引入外部知识的社交平台立场检测模型

Social Platform Stance Detection Model Infusing External Knowledge

针对社交平台的用户评论进行立场检测,旨在对某一特定话题下的用户评论进行立场分类。现有立场检测研究只关注评论文本的内部语义特征,而忽略了与评论文本相关的外部知识。通过将评论的关键信息映射到知识图谱中,以检索与该关键信息有关的外部知识,并将外部知识引入到模型中进行立场检测任务,该方法可通过提供可能对立场检测任务至关重要的背景知识来提升分类效果。在构建立场检测模型时,除考虑评论的文本特征外,采用门控图神经网络方法融合评论之间的结构信息,从而提取相关评论的相互影响情况。实验结果表明,与已有解决该问题的立场检测模型相比,该模型取得了较好的检测结果。将评论的文本特征与结构信息相结合并引入外部知识,可有效提升模型的立场检测性能。

Stance detection for user comments on social platforms aims to classify the stance of users′reviews towards a specific topic.Existing stance detection studies only focus on the internal semantic features of reviews text,while ignoring the external knowledge associated with the text of the review.Retrieve external knowledge related to the key information of a comment by mapping it to a knowledge graph and introduce the external knowledge into the model for the stance detection task,which can improve the classification result by providing background knowledge that may be crucial to the stance detection task.In addition to considering the textual features of reviews when constructing the stance detection model,employs a gated graph neural network approach to fuse the structural information between reviews which can capture the interactions of related reviews.The experiment shows that the model achieves better stance detection results compared with existing stance detection models for solving this problem.By combining the textual features and structural information of reviews and introducing external knowledge,the performance of the stance detection model can be effectively improved.