文章依据2000~2018年世界知识产权贸易数据,运用社会网络分析法,剖析世界知识产权贸易网络的结构特征,同时在此基础上对该网络的影响因素进行实证分析。结果表明:世界知识产权贸易网络随着时间演进其凝聚性不断增强,网络呈现显著的小世界特征;世界知识产权贸易网络存在明显的“核心–边缘”结构,边缘国(地区)随着时间的推移逐步减少,半边缘国(地区)呈增长趋势,网络中知识产权贸易关联增强;德国、俄罗斯、瑞典始终处于世界知识产权贸易网络的中心地位,在网络中发挥着重要的“桥梁”和“枢纽”作用;地理距离和知识产权保护水平是影响世界知识产权贸易网络的最主要因素,共同语言文化和经济距离在具体年份也能对知识产权贸易产生显著性影响;知识产权贸易网络存在区域异质性特征。Based on the world intellectual property trade data from 2000 to 2018, this thesis uses the social network analysis method to analyze the structural characteristics of the world intellectual property trade network, and makes an empirical analysis on the influencing factors of the network. The results show that: with the evolution of time, the cohesion of world intellectual property trade network is increasing, and the network presents significant small world characteristics;There is an obvious “core-periphery” structure in the world intellectual property trade network. The periphery countries (regions) gradually decrease as time goes on, the semi-periphery countries (regions) show an increasing trend, and the connection of intellectual property trade in the network is strengthened;Germany, Russia and Sweden have always been in the central position of the world intellectual property trade network, playing an important “bridge” and “hub” role in the network;Geographical distance and the level of intellectual property protection are the most important factors affecting the world intellectual proper
传统的基于表示学习的知识推理方法只能用于封闭世界的知识推理,有效进行开放世界的知识推理是目前的热点问题。因此,提出一种基于路径和增强三元组文本的开放世界知识推理模型PEOR(Path and Enhanced triplet text for Open world knowledge Reasoning)。首先,使用由实体对间结构生成的多条路径和单个实体周围结构生成的增强三元组,其中路径文本通过拼接路径中的三元组文本得到,而增强三元组文本通过拼接头实体邻域文本、关系文本和尾实体邻域文本得到;其次,使用BERT(Bidirectional Encoder Representations from Transformers)分别编码路径文本和增强三元组文本;最后,使用路径向量和三元组向量计算语义匹配注意力,再使用语义匹配注意力聚合多条路径的语义信息。在3个开放世界知识图谱数据集WN18RR、FB15k-237和NELL-995上的对比实验结果表明,与次优模型BERTRL(BERT-based Relational Learning)相比,所提模型的命中率(Hits@10)指标分别提升了2.6、2.3和8.5个百分点,验证了所提模型的有效性。
