关键词: 知识图谱补全, 表示学习, FOL规则, 可解释, 稀疏实体

Abstract:

In order to enhance the prediction accuracy and interpretability of knowledge graph representation, by improving the IterE framework consisting of three modules：representation learning, rule learning and rule fusion, a knowledge graph representation learning method jointing FOL rules is proposed for many representation learning algorithms, aiming at the rule learning and fusion module, rules confidence calculation method is improved based on the triples scoring function to expand applicability, and soft labels calculation method is improved to relax fusion requirements and expand fused data increment, to iteratively implement representation update rules and rules enhanced representation. Link prediction and generating explanations experiments show that adding logic rules makes base model improve the prediction accuracy and interpretability, and it helps to improve the representation of sparse entities in more sparse data sets.

Key words: knowledge graph completion, representation learning, FOL rules, interpretability, sparse entities