• 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):
    • 1
    • 2

• Next steps:

  • Zekun & Wensi:

  *(1) fix what remains to be fixed in the graph building code (syns, hyper/hypo, stopwords...) *(2) test that it works as expected on the Microsoft (English) and LINA (French, just added) sentence compression data sets *(3) implement path ranking function based on original Filipova edge weights, keyphrase scores (for coverage) and language model (for fluency)

  • Antoine: perform community detection (grouping of related sentences) and community cleaning (elimination of redundant and non-informative sentences) in an unsupervised way
  • next face-to-face progress meeting to be determined before Christmas break. I am always available via email for questions.

• 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