Keywords_extraction_with_GOW

• First we present an example of the methods used to extract keywords (see Graph of words and keywords extraction.ipynb and K-truss_code_example.ipynb)
• Then we give a code to compute the k_core and obtain the graphs of directories of files or all files in directories containing sub-directories (see K_core_corpus.py)
• We also give an implementation of the K-truss algorithm (see K-truss_code.py)
• We make a time analysis to see the evolution of some words through time, in order to detect events related to them.

Libraries

• Networkx to create and vizualize graphs
• Spacy to preprocess the text

Papers implemented :

• The k-core is directly taken from Networkx library.
• The k-truss is implemented following https://arxiv.org/abs/1205.6693
• The density and inflexion methods are implemented following https://www.aclweb.org/anthology/D16-1191

Graph of words and keywords extraction.ipynb

This notebook is dedicated to people who want to extract keywords from text document or corpus documents using a graph approach.

The goal of this notebook is to extract keywords from a text file using four different approachs :

• Best Coverage keywords extraction - http://www2013.w3c.br/proceedings/p715.pdf
• Div Rank keywords extraction - http://clair.si.umich.edu/~radev/papers/SIGKDD2010.pdf
• K-core Number - Python Library Networkx
• K-Truss - https://arxiv.org/abs/1205.6693

Through a french summary of Games of Thrones, we bring an example of the outputs of the four different approaches.

K-truss_code_example.ipynb

This jupyter notebook is an example of the following script

K-truss code

Two functions are implemented.

• The first one compute the K-truss of each node in G, the maximum non empty subgraph, the k from the maximum non empty subgraph and the necessary informations to compute the density and inflexion method.
• The second one gives the k-truss subgraph of the graph, where k is given as an input