Abstract: Link prediction is to discover and recover missing information in a knowledge graph (KG). Combining external knowledge and employing some specified methods to fulfill link prediction is the topic with great attention and key problem in KG research. Taking e-commerce application as the background, this paper combines the KG that has been constructed to describe user interest with external data, and adopts Bayesian network (BN), an important probabilistic graphical model, as the framework for representing and inferring the similarities among commodities as well as corresponding uncertainties. This paper constructs the BN to reflect the similarities by statistic computations upon commodity properties, and evaluates the authenticity of the links between commodity and user nodes based on the probabilistic inference mechanism of BN. Consequently, the real and complete KG can be obtained, as the basis of personalized recommendation and correlation query processing. The experimental results established on real data show the effectiveness of the model and algorithms proposed in this paper.

Key words: knowledge graph, link prediction, Bayesian network, similarity, probabilistic inference

摘要： 结合外部知识，使用特定方法进行知识图谱的链接预测，即知识图谱中缺失信息的发现和还原，是目前知识图谱领域研究的热点和关键。以电子商务应用为背景，基于已经构建好的描述用户兴趣的知识图谱，结合外部数据集，以贝叶斯网这一重要概率图模型作为不同商品之间相似性及其不确定性的表示和推理框架，通过对商品属性进行统计计算，构建反映商品之间相似关系的贝叶斯网，进而基于概率推理机制，定量地判断商品节点与用户节点之间存在链接的真实性，得到真实和完整的知识图谱，为个性化推荐和关联查询提供依据。建立在真实数据上的实验结果表明，提出的模型和算法是有效的。

关键词: 知识图谱, 链接预测, 贝叶斯网, 相似性, 概率推理