融合语义信息的知识图谱多跳推理模型

Multi-hop Inference Model for Knowledge Graphs Incorporating Semantic Information

多跳推理模型在知识图谱中充分挖掘和利用实体间的多步关系,组成路径信息,完成知识推理,然而,目前的稀疏知识图谱多跳推理模型大多存在数据稀少及推理路径可靠性较低等问题.为了解决该问题,文中提出融合语义信息的知识图谱多跳推理模型.首先,将知识图谱中的实体和关系嵌入向量空间,作为强化学习训练的外部环境.然后,利用查询关系和推理路径的语义信息,选择相似度最高的(关系,实体)对扩充智能体进行路径搜索的动作空间,以此弥补推理过程中数据稀少的不足.最后,使用推理路径和查询关系的语义相似度评价推理路径的可靠性,并作为奖励函数反馈给智能体.在多个公开稀疏数据集上的实验表明,文中模型明显提升推理性能.

In multi-hop inference models,path information is formed by fully mining and utilizing multi-step relationships between entities in the knowledge graph to accomplish knowledge inference.To solve the problems of sparse data and low reliability of inference paths in most of the existing sparse knowledge graph multi-hop inference models,a multi-hop inference model for knowledge graphs incorporating semantic information is proposed.Firstly,entities and relations in the knowledge graph are embedded into the vector space as the external environment for reinforcement learning training.Then,the semantic information of query relations and inference paths is employed to select the(relation,entity)pair with the highest similarity to expand the action space for path search by the agent,and thus the lack of sparse data in the inference process is compensated.Finally,the semantic similarity between the inference path and the query relation is utilized to evaluate the reliability of the inference path and it is fed back to the agent as a reward function.Experiments on several publicly available sparse datasets show that the inference performance of the proposed model is significantly improved.