融合知识图谱多维度信息的电力科研成果推荐算法OA
A Recommendation Algorithm for Electric Power Research Results by Fusing Multi-dimensional Information from Knowledge Graphs
随着电力行业研究的迅速发展,科研工作者需要从海量文献中筛选相关文献.现有的推荐系统往往忽视了文献间深层次的语义联系和专业领域的内在差别,导致推荐结果的准确性和个性化程度不足.针对这一问题,本文提出了一种基于知识图谱的电力科研成果推荐算法.首先,本文构建了用于电力科研领域推荐算法数据集,具体为构建出电力科研成果知识图谱,其中电力关键词作为推荐主体,文献标题作为推荐对象,并从标题中提取研究内容和方法信息以丰富知识图谱结构,利用查询的电力科研关键词来推荐所需的电力科研文献.其次,本文在基于知识图谱的意图网络(KGIN)模型的基础上,聚合了电力关键词和推荐文献邻居节点的信息,提出一种改进的KGIN模型,该模型可以更好地表达关键词和推荐文献之间的关系,融合了意图感知信息、关系感知信息、语义信息和高阶结构信息,实现了关键词和推荐文献更丰富的嵌入表示,获得了更好的推荐效果.最终实验结果表明,相较于本文所述的一些推荐系统基线模型,本文模型明显提升了对电力科研成果推荐的效果.
With the rapid development of research in the electric power industry,researchers need to filter relevant liter-ature from a huge amount of literature.Existing recommendation systems often ignore the deep semantic links between the literature and the intrinsic differences within the professional field,resulting in insufficient accuracy and personalisation of the recommendation results.To address this problem,this paper proposes a knowledge graph-based recommendation algo-rithm for electric power research results.Firstly,this paper constructs a recommendation algorithm dataset for the field of electric power research,specifically to construct a knowledge graph of electric power research results,in which the electric power keywords are used as the recommendation subject,the title of the literature as the recommendation object,and the re-search content and method information is extracted from the title in order to enrich the structure of the knowledge graph,and the query electric power research keywords are used to recommend the required electric power research literature.Secondly,based on the knowledge graph-based intent network(KGIN)model,this paper aggregates the information of the neighbour nodes of the electric power keywords and the recommended literature,and puts forward an improved KGIN model,which can better express the relationship between the keywords and the recommended literature,and integrates the intent-aware in-formation,the relationship-aware information,the semantic information,and the higher-order structural information,and achieves a richer embedding representation of the keywords and the recommended literature,and obtains a better representa-tion of the keywords and the recommended literature.rich embedding representation,and obtains better recommendation effect.The final experimental results show that compared with the baseline models of some recommendation systems de-scribed in this paper,the model in this paper obviously improves the effect of recommending scientific research results on e-lectric power.
徐晓轶;毛艳芳;吕晓祥
国网江苏省电力有限公司南通供电分公司,江苏南通 226300国网江苏省电力有限公司南通供电分公司,江苏南通 226300国网江苏省电力有限公司南通供电分公司,江苏南通 226300
信息技术与安全科学
知识图谱推荐系统图卷积神经网络电力科研成果
knowledge graphrecommender systemgraph convolutional neural networkelectric power research results
《计算技术与自动化》 2026 (1)
43-49,7
国网江苏省电力有限公司科技项目(J2023051)
评论