基于多模态知识图谱的跨平台信息推荐仿真  

Cross-Platform Information Recommendation Simulation Based on Multimodal Knowledge Graph

作　　者：陈海红 申广忠[2] CHEN Hai-hong;SHEN Guang-zhong(College of Mathematics and Computer Science Chifeng University,Chifeng Inner Mongolia 024000,China;Dalian Jiaotong University College of Software,Dalian Liaoning 116020,China)

机构地区：[1]赤峰学院数学与计算机科学学院,内蒙古赤峰024000 [2]大连交通大学软件学院,辽宁大连116020

出　　处：《计算机仿真》2024年第10期463-467,共5页Computer Simulation

基　　金：内蒙古自治区高等学校科学研究项目(NJZY22189)。

摘　　要：不同平台中的信息以不同的格式和结构存在,且存在质量、可用性等方面的差异,导致信息推荐效率较低,误差较大。于是提出一种基于多模态知识图谱的跨平台信息推荐算法。建立多模态知识图谱,通过基于注意力机制的卷积神经网络抽取各个平台文本语义标签,采用人工标注方法进一步抽取信息的语义标签,将抽取的各个平台的不同语义标签融合处理,利用最长公共子序列(LCS)算法完成不同平台信息的相似度计算,选择相似度最高的平台构建候选跨平台信息推荐列表。通过聚类识别出每个用户的兴趣,建立用户兴趣矩阵,计算用户当前的兴趣度,利用用户兴趣度建立转移概率矩阵,同时结合候选跨平台信息推荐列表实现信息推荐。实验结果表明,所提算法的跨平台信息推荐结果准确性和覆盖率均得到有效提升,同时推荐结果也更加合理。Dfferent platforms store information in different formats and structures,resulting in low efficiency of information recommendation and large errors.Therefore,a cross-platform information recommendation algorithm based on multimodal knowledge graphs was proposed.At first,a multimodal knowledge graph was built,and then the convolutional neural network based on attention mechanism was used to extract semantic labels of text from all platforms.Furthermore,an artificial labeling method was adopted to extract semantic labels and fuse them.Meanwhile,the longest common subsequence(LCS)algorithm was used to calculate the similarity between information on different platforms.Then,the platform with the highest similarity was selected to construct a candidate list for cross-platform information recommendation.Based on the cluster,all the users'interests were identified.Moreover,a matrix of user interest was established to calculate the current interest degree of users.Finally,a transfer probability matrix was built based on user interests.In combination with the candidate list for cross-platform information recommendation,the information recommendation was realized.Experimental results show that the accuracy and coverage rate of the crossplatform information recommendation results obtained by the proposed algorithm has been effectively improved,and the recommended results are more reasonable.

关 键 词：多模态 知识图谱 跨平台信息 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]