一种引入元路径相似性度量的材料实体检索方法  

Material entity retrieval method introducing similarity measure based on meta-path

作　　者：黄华泽 胡紫璇 游进国[1,2] 黄星瑞 陶静梅 易健宏[3] Huang Huaze;Hu Zixuan;You Jinguo;Huang Xingrui;Tao Jingmei;Yi Jianhong(Faculty of Information Engineering&Automation,Kunming University of Science&Technology,Kunming 650500,China;Yunnan Key Laboratory of Artificial Intelligence,Kunming 650500,China;Faculty of Material Science&Engineering,Kunming University of Science&Technology,Kunming 650093,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]云南省人工智能重点实验室,昆明650500 [3]昆明理工大学材料科学与工程学院,昆明650093

出　　处：《计算机应用研究》2024年第9期2781-2786,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(62062046)。

摘　　要：近年来,随着材料数据的积累以及“材料基因组计划”的普及,面对大量需要处理和管理的材料数据,快速准确地检索并获取相应信息已成为一个重要问题。传统的检索方法由于仅能查询某一材料的相关信息,并且存在检索结果不全面、无法处理复杂语义关系等问题,难以获取相似程度较高的材料。为了快速、准确地找到与某种材料相似的材料,提出可度量不同节点的加权材料相似度计算模型WM-PathSim。首先,使用metapath2vec学习材料节点的嵌入表示;其次,引入TFIDF-CBOW模型学习材料路径实例的存在概率,进而计算不同元路径的权重;最后,加权求和符合条件的元路径得到最后的相似性度量,来预测不同材料之间的相似程度。在真实数据集上的结果表明,在不同的路径关系中,所提模型相比于基线方法在性能上有较大提升,其AUC和precision指标分别提升了0.37~5.02百分点和1~7.33百分点,说明所提模型得到材料间的相似程度更加准确和有效,从而能够获得相似材料。In recent years,with the accumulation of material data and the popularization of the“material genome project”,it has become an important issue to retrieve and obtain the corresponding information quickly and accurately in the face of a large amount of material data that needs to be processed and managed.However,traditional retrieval methods can only query information related to a certain material,and there are problems such as incomplete retrieval results and inability to handle complex semantic relations,making it difficult to obtain materials with a high degree of similarity.In order to find materials similar to a certain material quickly and accurately,this paper proposed a weighted material similarity calculation model WM-PathSim that could measure different nodes.Firstly,it learned the embedding representation of material nodes by using metapath2vec.Secondly,it introduced the TFIDF-CBOW model to learn the existence probability of material path instances,and then calculated the weights of different meta-paths.Finally,it obtained the weighted summation of eligible meta-paths as the final similarity measure to predict the similarity between different materials.The results on the real datasets show that the proposed model has a greater performance improvement compared with the baseline method in different path relations,and its AUC and precision metrics are improved by 0.37~5.02 percentage points and 1~7.33 percentage points,respectively.It indicates that this model is more accurate and effective in obtaining the degree of similarity between materials,and thus enabling the acquisition of similar materials.

关 键 词：材料相似度 metapath2vec TFIDF-CBOW 元路径权重 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]