Entity Embeddings with Conceptual Subspaces as a Basis for Plausible Reasoning
This work addresses the problem of automating conceptual space construction for AI applications, representing an incremental advancement in bridging cognitive models with machine learning.
The paper tackled the lack of automated methods for constructing conceptual spaces in AI by proposing a method that learns entity embeddings from Wikipedia and constrains them into lower-dimensional subspaces, demonstrating experimentally that these subspaces approximate conceptual spaces with features like directions for important attributes and convex regions for natural properties.
Conceptual spaces are geometric representations of conceptual knowledge, in which entities correspond to points, natural properties correspond to convex regions, and the dimensions of the space correspond to salient features. While conceptual spaces enable elegant models of various cognitive phenomena, the lack of automated methods for constructing such representations have so far limited their application in artificial intelligence. To address this issue, we propose a method which learns a vector-space embedding of entities from Wikipedia and constrains this embedding such that entities of the same semantic type are located in some lower-dimensional subspace. We experimentally demonstrate the usefulness of these subspaces as (approximate) conceptual space representations by showing, among others, that important features can be modelled as directions and that natural properties tend to correspond to convex regions.