fursovia/geometric_embedding

"Zero-Training Sentence Embedding via Orthogonal Basis" paper implementation

Geometric Embedding Algorithm (GEM)

This is an implementation of Geometric Embedding Algorithm, a simple and robust non-parameterized approach for building sentence representations. See the paper for more details.

The work is done as a project for NLA course at Skoltech.

Example

from gem import SentenceEmbedder
from embeddings import get_embedding_matrix

sentences = ["We come from the land of the ice and snow",
            "From the midnight sun where the hot springs blow"]
            
embedding_matrix, vocab = get_embedding_matrix('glove.6B.300d.txt')
embedder = SentenceEmbedder(sentences, embedding_matrix, vocab)

embedded_sentences = embedder.gem(window_size=3, sigma_power=3)

Data used

• GloVe embeddings
• LexVec embeddings

Team

• Alexey Bokhovkin
• Eugenia Cheskidova
• Ivan Fursov
• Ruslan Rakhimov