查询结果:   杨瑞达,林欣,杨燕,贺樑,窦亮.基于混合增强智能的知识图谱推理技术研究[J].计算机应用与软件,2019,36(6):149 - 154.
中文标题
基于混合增强智能的知识图谱推理技术研究
发表栏目
人工智能与识别
摘要点击数
842
英文标题
KNOWLEDGE GRAPH REASONING BASED ON HYBRID-AUGMENTED INTELLIGENCE
作 者
杨瑞达 林欣 杨燕 贺樑 窦亮 Yang Ruida Lin Xin Yang Yan He Liang Dou Liang
作者单位
华东师范大学计算机科学与软件工程学院 上海 200062 国家新闻出版署出版融合发展 华东师大社   
英文单位
School of Computer Science and Software Engineering, East China Normal University, Shanghai 200062, China SAPPRFT Key Laboratory of Publishing Integration Development, East China Normal University, Shanghai 200062, China    
关键词
知识图谱 知识图谱推理 强化学习 混合增强智能
Keywords
Knowledge graph Knowledge graph reasoning Reinforcement learning Hybrid-augmented intelligence
基金项目
国家自然科学基金项目(61773167);上海市科委项目(17511102702);上海市经信委项目(170513);国家新闻出版署出版融合发展(华东师大社)重点实验室开放课题基金
作者资料
杨瑞达,硕士生,主研领域:知识图谱。林欣,教授。杨燕,讲师。贺樑,教授。窦亮,讲师。 。
文章摘要
如今,知识图谱被广泛应用在各个领域,例如问答系统、推荐系统等。而基于知识图谱的应用表现很大程度上依赖于知识图谱本身的知识完备性与准确性。单纯通过人工补齐与审核的方式来构建知识图谱已无法满足超大规模知识图谱的需求。 针对上述问题,提出一种基于混合增强智能的知识图谱推理框架,即同时利用机器模型与人的知识信息来完成知识图谱推理。该框架在基于知识图谱嵌入的向量空间中,利用混合增强智能模型来寻找到实体节点之间的有效路径。与现有方法不同的是,该方法在训练模型时,高效地利用人的知识信息来指导模型的优化。实验表明,该框架在公开数据集上的表现相较于现有方法有一定提升。
Abstract
Nowadays, knowledge graph is widely used in various fields, such as question answering system, recommendation system and so on. The performance of application based on knowledge graph depends on the completeness and accuracy of knowledge graph. It is impossible to construct knowledge graph by manual completion and auditing to meet the needs of large-scale knowledge graphs. To address these challenges, we proposed a knowledge map reasoning framework based on hybrid-augmented intelligence. It uses machine model and human knowledge information to complete knowledge map reasoning simultaneously. The framework used hybrid-augmented intelligent model to find effective paths between entities in vector space based on knowledge graph embedding for knowledge reasoning. Different from the existing methods, this method made efficient use of human knowledge information to guide the optimization of the training model. Experiments on public datasets demonstrated the improvement of the proposed framework compared to existing methods.
下载PDF全文