基于翻译嵌入模型的可视化推荐方法  

Research on Visualization Recommendation Method Based onTranslation Embedding Model

作　　者：张晓蓉[1] 刘崎鹏 廖竞[1] 陈元松[1] 李杨 李琦 ZHANG Xiaorong;LIU Qipeng;LIAO Jing;CHEN Yuansong;LI Yang;LI Qi(School of Computer Science&Technology,Southwest University of Science and Technology;Sichuan Gas Turbine Establishment,Areo Engine Corporation of China;Sichuan Changhong Electrical Appliances Co.,Ltd,Sichuan Hongxin Software Co.,Ltd,Mianyang 621000,China)

机构地区：[1]西南科技大学计算机科学与技术学院 [2]中国航空发动机集团有限公司四川燃气涡轮研究院 [3]四川长虹电器股份有限公司四川虹信软件股份有限公司,四川绵阳621000

出　　处：《软件导刊》2024年第11期32-38,共7页Software Guide

基　　金：国家自然科学基金项目(61872304,61802320)。

摘　　要：可视化推荐与自动生成可视化可以帮助无数据可视化背景的用户快速创建有效的数据可视化。针对基于规则和机器学习的方法中工作量大和可解释性差的问题,提出一种基于知识图谱嵌入模型的方法,采用缓存自对抗负采样学习实体和关系的嵌入,构建知识图谱推理可视化设计。首先,使用基于规则的方法从数据集中抽取特征,构建知识图谱;其次,采用改进的TransH(即sTransH)模型来学习实体与关系的嵌入,解决一对多/多对一的状况;最后,根据实体的嵌入投影和平移向量来推理可视化设计选择。实验结果表明,sTransH模型在精度、平均排名(MR)、排名比例(Hit@2)等指标评估中均优于现有的TransE、TransH、RotatE和TransE-adv基线方法,验证了该模型在可视化推荐方面的有效性。Visualization recommendation or automatic visualization generation can help users with no background in data visualization to quickly create effective data visualizations.Aiming at the problems of heavy manual effort and poor interpretability in existing rule-based meth⁃ods and machine learning-based methods,this paper proposes a knowledge graph embedding model based method,which uses cache self-ad⁃versarial negative sampling to learn embeddings of entities and relations,and constructs a knowledge graph to reason about visualization de⁃signs.Firstly,a rule-based method is used to extract features from the data set to construct a knowledge graph.Secondly,the improved TransH(sTransH)model is used to learn the embedding of entities and relations to solve the one-to-many/many-to-one situation.Finally,we reason about visualization design choices based on the embedded projections and translation vectors of entities.Experimental results show that sTransH outperforms the existing TransE,TransH,RotatE,and TransE-adv baseline methods in terms of accuracy,mean rank(MR),and rank ratio(Hit@2),which verifies the effectiveness of the proposed sTransH model in visualization recommendation.

关 键 词：数据可视化 可视化推荐 知识图谱嵌入 翻译模型 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]