Representing Hierarchical Structured Data Using Cone Embedding

Extracting hierarchical structure in graph data is becoming an important problem fields such as natural language processing and developmental biology. Hierarchical structures can be extracted by embedding methods non-Euclidean spaces, Poincaré Lorentz embedding, it now possible to learn efficient taking advantage of the these spaces. In this study, we propose into another type metric space called a cone learning only one-dimensional coordinate variable added original vector or pre-trained space. This allows for extraction information while maintaining properties embedding. The extension has that curvature easily adjusted parameter even when coordinates are fixed. Through extensive empirical evaluation have corroborated effectiveness proposed model. case randomly generated trees, demonstrated superior performance extracting compared existing techniques, particularly high-dimensional settings. For WordNet embeddings, exhibited noteworthy correlation between human outcomes.