[1]韩锦雪,苏美红,周慧媛,等.基于加权概念格与AE的社交知识图谱语义压缩[J].计算机技术与发展,2024,34(11):58-64.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0221]
　HAN Jin-xue,SU Mei-hong,ZHOU Hui-yuan,et al.Semantic Compression of Social Knowledge Graph Based on Weighted Concept Lattice and AE[J].,2024,34(11):58-64.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0221]
点击复制

基于加权概念格与AE的社交知识图谱语义压缩()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
34
期数:
2024年11期
页码:
58-64
栏目:
媒体计算
出版日期:
2024-11-10

文章信息/Info

Title:
Semantic Compression of Social Knowledge Graph Based on Weighted Concept Lattice and AE
文章编号:
1673-629X(2024)11-0058-07
作者:
韩锦雪; 苏美红; 周慧媛; 张素兰
太原科技大学 计算机科学与技术学院,山西 太原 030024
Author(s):
HAN Jin-xue; SU Mei-hong; ZHOU Hui-yuan; ZHANG Su-lan
School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China
关键词:
语义压缩; 社交知识图谱; 加权概念格; 自编码器; 知识图谱
Keywords:
semantic compression; social knowledge graph; weighted concept lattice; autoencoder; knowledge graph
分类号:
TP391
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0221
摘要:
社交知识图谱具有时效性强、更新频繁快等特点,但也导致内容不断扩充,产生了大量冗余信息。 如何删除冗余信息并对其进行语义压缩成为提升社交知识图谱质量的关键。 为此,该文提出一种基于加权概念格与自编码器的社交知识图谱语义压缩方法(SCWCL-AE)。 首先,将社交知识图谱资源描述框架三元组转换为二进制矩阵,获得同一社交知识图谱不同时间版本的形式背景,并通过信息熵确定实体属性权值,构造社交知识图谱加权概念格;提取满足加权外延社交知识图谱支持度的实体属性分类规则,获得重要且使用频率高的实体摘要。 其次,引入无监督自编码器,从无标签的社交知识图谱实体摘要中自动学习抽象特征,以重构损失最小化输出数据,达到压缩知识图谱冗余语义信息的目的。 最后,在Freebase、DBpedia、Zhishi. me 数据集上进行实验,结果验证了该方法对社交知识图谱语义压缩的有效性。
Abstract:
The social knowledge graph possesses characteristics such as strong timeliness and frequent updates,which lead to continuous expansion of content and the generation of a large amount of redundant information. Removing redundant information and semantically compressing it become crucial for improving the quality of the social knowledge graph. Therefore,we propose a method for semantic compression of social knowledge graph based on weighted concept lattice and autoencoder ( SCWCL - AE). Firstly, the resource description framework triples of the social knowledge graph are transformed into binary matrices to obtain the formal context of different versions of the same social knowledge graph at different times,and the entity attribute weights are determined through information entropy to construct the weighted concept lattice of the social knowledge graph. Subsequently,entity attribute classification rules that satisfy the weighted extension support of the social knowledge graph are extracted to obtain important and frequently used entity summaries.Secondly,an unsupervised autoencoder is introduced to automatically learn abstract features from unlabeled entity summaries of the social knowledge graph and minimize reconstruction loss to compress redundant semantic information of the knowledge graph. Finally,experiments conducted on the Freebase,DBpedia,and Zhishi. me datasets validate the effectiveness of the proposed method in semantic compression of social knowledge graph.
更新日期/Last Update: 2024-11-10