摘要
在B-CNN模型各个特征通道内引进比例因子,结合正则化激活方式构建稀疏层,完成特征通道筛选,利用改进B-CNN构建异构网络大数据知识表示模型,通过维度变换方式增加卷积滑动窗口的滑动步数,提高数据内实体与关系的信息共享作用,利用可变粒度策略分割有效数据知识三元组的细粒度数据,实现异构网络大数据知识扩充。实验证明:该算法在表示数据知识时三元组预测准确比例较高,归一化互信息与调整兰德指数均较高,收敛速度较快,数据知识表示效果和扩充效果较好。
The scale factor was introduced into each characteristic channel of B-CNN,the sparse layer was constructed in combination with the regularization activation method,the characteristic channel screening was completed,the heterogeneous network big data knowledge representation model was constructed by using the improved B-CNN,the sliding steps of convolution sliding window were increased by dimension transformation,and the information sharing function of entities and relationships in the data was improved.The variable granularity strategy was used to segment the fine-grained data of effective data knowledge triples to realize the knowledge expansion of big data in heterogeneous networks.Experiments show that the algorithm has a high proportion of triple prediction accuracy,high normalized mutual information and adjusted Rand index,fast convergence speed,and good data knowledge representation and expansion effect.
作者
张伟华
王海英
ZHANG Weihua;WANG Haiying(College Mechanical and Electrical Engineering, Zhengzhou Business University, Gongyi 451200, China)
出处
《兵器装备工程学报》
CSCD
北大核心
2022年第6期290-294,共5页
Journal of Ordnance Equipment Engineering
基金
河南省科技攻关计划项目(202102210357)
郑州商学院新工科创新融合团队项目(2021-CXTD-05)。
关键词
B-CNN模型
异构网络
大数据
知识扩充
比例因子
可变粒度
B-CNN model
heterogeneous network
big data
knowledge expansion
proportional factor
variable size