This is a python implementation of TextRank for automatic keyword and sentence extraction (summarization) as done in https://web.eecs.umich.edu/~mihalcea/papers/mihalcea.emnlp04.pdf. However, this implementation uses Levenshtein Distance as the relation between text units.
This implementation carries out automatic keyword and sentence extraction on 10 articles gotten from http://theonion.com
- 100 word summary
- Number of keywords extracted is relative to the size of the text (a third of the number of nodes in the graph)
- Adjacent keywords in the text are concatenated into keyphrases
To install the library run the setup.py
module located in the repository's root directory. Alternatively, if you have access to pip you may install the library directly from github:
pip install git+https://github.com/davidadamojr/TextRank.git
Use of the library requires downloading nltk resources. Use the textrank initialize
command to fetch the required data. Once the data has finished downloading you may execute the following commands against the library:
textrank extract_summary <filename>
textrank extract_phrases <filename>
Install the library as "editable" within a virtual environment.
pip install -e .
Dependencies are installed automatically with pip but can be installed serparately.
- Networkx - https://pypi.python.org/pypi/networkx/
- NLTK 3.0 - https://pypi.python.org/pypi/nltk/3.2.2
- Numpy - https://pypi.python.org/pypi/numpy
- Click - https://pypi.python.org/pypi/click