Learning path offers personalized guidance in order to improve learners' performance in e-learning systems. With the growth of learning resources, the workload of manually planning learning paths is increasing accordingly, which becomes a burden to experts and teachers. Knowledge graph is capable of describing the characteristics of knowledge and learning resources, providing inspirations to generate a learning path. In this paper, we propose a method of a learning path generator based on knowledge graph, which firstly generates a sequence of knowledge points by the self-designed topological ranking algorithm and then serializes the learning objects by using ant colony optimization. Results of comparative experiment reveal that there is no difference in learning time and performance between the experimental group and control group, and that the automatically generated learning path is highly similar to the expert path with the same learning outcomes.