Hyperbolic Centroid Calculations for Text Classification
This addresses a technical bottleneck in NLP for researchers using hyperbolic embeddings, but it appears incremental as it adapts existing centroid concepts to a new space.
The paper tackled the problem of constructing document representations from hyperbolic word embeddings, where traditional averaging is invalid, by investigating hyperbolic centroid schemes for text classification, but no concrete performance numbers were provided.
A new development in NLP is the construction of hyperbolic word embeddings. As opposed to their Euclidean counterparts, hyperbolic embeddings are represented not by vectors, but by points in hyperbolic space. This makes the most common basic scheme for constructing document representations, namely the averaging of word vectors, meaningless in the hyperbolic setting. We reinterpret the vector mean as the centroid of the points represented by the vectors, and investigate various hyperbolic centroid schemes and their effectiveness at text classification.