epapagia/dynamicword2vec

Dynamic Word Embeddings for Evolving Semantic Discovery code.

DynamicWord2Vec

Paper title: Dynamic Word Embeddings for Evolving Semantic Discovery.

Paper links: https://dl.acm.org/citation.cfm?id=3159703 https://arxiv.org/abs/1703.00607

Files:

/embeddings

• embeddings in loadable MATLAB files. 0 corresponds to 1990, 1 to 1991, ..., 19 to 2009. To save space, each year's embedding is saved separately. When used in visualization code, first merge to 1 embedding file.

/train_model

• contains code used for training our embeddings

• data file download: https://www.dropbox.com/s/nifi5nj1oj0fu2i/data.zip?dl=0

  /train_model/train_time_CD_smallnyt.py

  • main training script

  /train_model/util_timeCD.py

  • containing helper functions

/other_embeddings

• contains code for training baseline embeddings

• data file download: https://www.dropbox.com/s/tzkaoagzxuxtwqs/data.zip?dl=0

  /other_embeddings/staticw2v.py

  • static word2vec (Mikolov et al 2013)

  /other_embeddings/aw2v.py

  • aligned word2vec (Hamilton, Leskovec, Jufarsky 2016)

  /other_embeddings/tw2v.py

  • transformed word2vec (Kulkarni, Al-Rfou, Perozzi, Skiena 2015)

/visualization

• scripts for visualizations in paper

  /visualization/norm_plots.py

  • changepoint detection figures

  /visualization/tsne_of_results.py

  • trajectory figures

/distorted_smallNYT

• code for robust experiment
• data file download: https://www.dropbox.com/s/6q5jhhmxdmc8n1e/data.zip?dl=0

/misc

• contains general statistics and word hash file