| 李少杰,陈曙东,郝悦星,欧阳小叶,龚立晨.基于卷积神经网络的高效知识表示模型[J].高技术通讯(中文),2020,30(9):901~907 |
| 基于卷积神经网络的高效知识表示模型 |
|
| |
| DOI：doi：10.3772/j.issn.1002-0470.2020.09.004 |
| 中文关键词: 知识图谱； 知识表示； 卷积神经网络（CNN）； 知识补全； 维度变换； 信息交互 |
| 英文关键词: knowledge graph, knowledge representation, convolutional neural network (CNN), knowledge graph completion, dimensional transformation, information interaction |
| 基金项目: |
|
| 摘要点击次数: 2692 |
| 全文下载次数: 1624 |
| 中文摘要: |
| 为提升知识表示的有效性和可靠性，提出一种基于卷积神经网络（CNN）的知识表示模型（ConvKE）。ConvKE采用维度变换策略来提升卷积滑动窗口在三元组矩阵上的滑动步数以及三元组内实体和关系在更多维度上的信息交互能力。ConvKE还通过2-D卷积滑动窗口提升感受野来捕获三元组更多维度上的整体信息。通过采用知识补全任务来评估ConvKE模型的效果，实验结果证明了ConvKE在2个基准数据集WN18RR、FB15K-237的平均排名（MR）指标上取得了较好的结果。 |
| 英文摘要: |
| In order to improve the validity and reliability of knowledge representation, this paper proposes a knowledge representation model based on convolutional neural network (CNN). Convolutional knowledge embeddings (ConvKE) uses a dimensional transformation strategy to increase the number of sliding steps of the convolution sliding window on the triple matrix and the information interaction capabilities of the entities and relations within the triple in more dimensions. ConvKE also captures the overall information of the triples in more dimensions by using the 2D convolution sliding windows with enhancing receptive field. By using the link prediction task to evaluate the performance of ConvKE, the experimental results show that ConvKE has achieved good results on the mean rank (MR) metric of the two benchmark datasets WN18RR and FB15K-237. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
