Phrases support and simpler FAISS installation

• Adding the possibility to specify a set of phrases from a dictionary which will be identified in the input corpus, e.g. target entities or locations or terms of domain of interest (using the -phrases key).

• Now FAISS can be installed using conda in much-much simpler and platform independent way.

Adding new pre-computed models

• Add new pre-computed models for English, German, and Russian languages
• Moving the old pre-computed models to the new format
• Adding documentation on how to convert pre-trained word embedding to sense embeddings
• Updating the Quick Start examples and usage examples
• Adding an easier user interface of the word sense disambiguation class

Rewriting all modules of the project in Python and porting it to Python 3

• Remove dependency to the binary C word2vec implementation replacing it with the gensim word2vec implementation
• Remove dependency to the Java implementation of ChineseWhispers algorithm and word sense induction: now using the chinese_whispers module from pipy and word sense induction inside the project
• Migration to Python 3
• Removing old slower implementation of the algorithm for computation of word graph with much faster one based on the Facebook FAISS library
• Addressing consistency issues with the command line arguments
• Reimplementing the word pooling mechanism
• Adding make files for easy installation and training test models
• Adding a iPython notebook with a QuickStart guide

New stable release

This release introduces a makefile for installing of the dependencies and several refactorings in the vector pooling stage.

The version published in the original paper

Pelevina et al., (2016). Making sense of word embeddings.