近年来,基于知识图谱(KG)的服务推荐方法取得了显著进展,但现有方法的性能仍受限于节点信息属性的质量和图结构的局限性,尤其是在实体描述信息不足和大量无关三元组噪声的影响下,推荐准确性受到影响。为此,我们提出了一种结合大语言模型(LLM)的知识结构更新服务推荐方法。基于思维链设计提示策略,进一步增强了大语言模型在扩展知识图数据描述信息中的输出效果,从而全面提升了实体数据的语义表达能力。此外,设计了一种知识感知重构任务,该任务有效识别知识图谱中的关键关联子图,并削弱无用三元组对推荐结果的负面影响,进而优化图结构以提升推荐性能。在真实数据集上进行实验结果表明,我们的方法在面向mashup的API服务推荐优于几种先进的基线方法。In recent years, service recommendation methods based on knowledge graph (KG) have made remarkable progress, but the performance of existing methods is still limited by the quality of node information attributes and the limitations of graph structure, especially under the influence of insufficient entity description information and a lot of irrelevant triplet noise, which affects the recommendation accuracy. Therefore, we propose a knowledge structure update service recommendation method combined with a large language model (LLM). The suggestion strategy based on thought chain design further enhances the output effect of large language model in expanding knowledge graph data description information, thus comprehensively improving the semantic expression ability of entity data. In addition, a knowledge perception reconstruction task is designed, which effectively identifies the key association subgraphs in the knowledge graph, weakens the negative impact of useless triples on the recommendation results, and then optimizes the graph structure to improve the recommendation performance. Experimental results on real data sets show that our approach is superior to several advanced baseline approaches for Mashup-oriented API service recommendations.
多视图聚类(Multi-View Clustering,MVC)旨在利用不同视图间的一致性和互补性来高效处理多视图数据,是大数据分析中重要的研究方向之一.然而,现有方法无法有效学习到多视图信息间的潜在联系,且缺乏考虑视图重要性差异问题.针对上述这些问题,本文提出了一种基于分布对齐变分自编码器的深度多视图聚类方法(Deep Multi-View Clustering based on Distribution Aligned Variational Autoencoder,DMVCDA).首先,针对特定视图我们利用多个变分自编码器从不同视图中提取潜在特征,并对特征的分布进行对齐,以挖掘包含基本信息的潜在特征;然后,引入视图权重参数,获取共享的潜在特征;最后,在潜在特征上建立面向聚类的损失目标,使得学习到的潜在特征更适合聚类任务,从而提高聚类精度.在五个公共多视图数据集上的实验结果表明,我们的模型在精确度(ACC)、标准互信息(NMI)和纯度(Purity)等多个聚类评价指标上均表现出优异的性能.
暂无评论