Introduction

EmbeddingsAnalysis is a package containing algorithms and utilities for embeddings processing. At this point, only word embeddings are de facto supported however other types (i.e. graph embeddings) could be used as well.

Processing methods

The package implements the following embeddings processing algorithms:

• Artetxe et al. "Uncovering divergent linguistic information in word embeddings with lessons for intrinsic and extrinsic evaluation", 2018
• Vikas Raunak "Simple and effective dimensionality reduction for word embeddings", NIPS 2017 Workshop

and utilities:

• word vector compression through CompressedWordVectors (uses QuantizedArrays.jl)
• saving WordVectors, CompressedWordVectors objects to disk in either binary or text format
• convert ConceptNet objects to WordVectors objects
• selection of a subset of word vectors starting from a seed vocabulary
• simple word vector based translation (requires aligned embeddings)

Installation

Installation can be performed from either outside or inside Julia with:

$ git clone https://github.com/zgornel/EmbeddingsAnalysis.jl

and

using Pkg
Pkg.clone("https://github.com/zgornel/EmbeddingsAnalysis.jl")

respectively.