Abstract: With the widespread application of industrial cloud robots,the explosive growth of service information has caused service information overload.Quick and efficient recommending high-quality and reliable services to users from a large number of manufacturing services is very difficult.A recommendation method for industrial cloud robot manufacturing capability service based on knowledge representation learning was proposed,which started from the construction of industrial cloud robot manufacturing capability service knowledge graph,and analyzed industrial cloud robot manufacturing task information,manufacturing capability information,service function information and non-function information.To improve the speed of knowledge recommendation,the knowledge representation learning was introduced.It represented entities and relationships as low-dimensional dense vectors to enable efficient computation of semantic associations between entities and relationships.Aiming at the problem that the random replacement method used to generate negative samples in most translation models leaded to the low prediction accuracy of the model,a new negative sample generation strategy was proposed,which was combined with TransE to verify the negative sample generation strategy on the FB15K.The experimental results showed that it had higher accuracy in link prediction.

Key words: industrial cloud robotics, knowledge graph, TransE algorithm, service recommendation

摘要： 随着工业云机器人的广泛应用,服务信息爆炸式增长造成服务信息过载,为了快速有效地从海量制造服务中为用户推荐优质可靠的服务,提出一种基于知识表示学习的工业云机器人制造能力服务推荐方法,从工业云机器人制造能力服务知识图谱的构建出发,对工业云机器人制造任务信息、制造能力信息、服务功能信息和非功能信息进行描述;为提高知识推荐的速度,引入知识表示学习,将实体和关系表示为低维稠密的向量,实现实体和关系的语义联系的高效计算。针对大多数翻译模型负样本生成采用随机替换的方法导致模型预测精度低的问题,提出了一种新的负样本生成策略,将其与TransE结合,在FB15K数据集上验证了负样本生成策略的有效性,实验结果证明本文所提方法在进行链接预测时具有更高的准确率。

关键词: 工业云机器人, 知识图谱, TransE算法, 服务推荐