• Relevant papers can be found in the corresponding folder

• Some useful links:

  • list of stopwords (English) from the SMART information retrieval system can be found here
  • Boudin and Morin 2013 repo with graph-building code
  • using WordNet in Python
  • language modeling in Python (to be used for the path ranking and selection step):
    • Python NLTK language models
    • Language Modeling with Python's Theano
  • Microsoft Concept Graph (as a possible alternative to WordNet for hypo/hypernyms)

• Next steps:

  *(1) integrate word attraction force, do not consider proper nouns (NNP) for synonym/hypernym/entailment replacement *(2) try embedding-based clustering of keywords *(3) re-organize the code so that all tuning parameters are easily accessible *(4) run experiments for traditional documents and meetings

• Word embeddings resources:

• note that the Word Mover's Distance paper and the CoreRank (EMNLP 2016) paper are already in the 'papers' folder.

• gensim library in Python for word2vec

• training visualization demo (for Chrome)

• interesting blogpost

• introduction slides with links to the important papers