In this study a model of Siamese topological neural networks, which consisted of a pair of hierarchical neural networks each with a lowdimensional internal layer, is proposed. Through similarity learning, the objective of the proposed siamese network is to learn low-dimensional topological representations of a given similarity between pairwise highdimensional inputs. The low-dimensionality of the internal layer allows human to visualize the structure of the high-dimensional data in the context of their similarities, for example, label-similarity or ranksimilarity. Different from many similarity learning techniques, dimensionality reduction is integrated in the proposed siamese networks allowing flexible context-oriented visualization analysis for high-dimensional data.