1: \begin{abstract}

2: A hyperbolic space has been shown to be more capable of modeling complex networks

3: than a Euclidean space.

4: This paper proposes an explicit update rule along geodesics in a hyperbolic space. 

5: The convergence of our algorithm is theoretically guaranteed, and the convergence rate is better than the conventional Euclidean gradient descent algorithm.

6: Moreover, our algorithm avoids the ``bias'' problem of existing methods using the Riemannian gradient.

7: Experimental results demonstrate the good performance of our algorithm in the \Poincare embeddings of knowledge base data.

8: \end{abstract}

9: