Automated Domain-based Keyword Detection

Keyword detection for a particular domain, based on the PRDualRank framework.

Contributors

Work done during Fall 2019, Spring 2020, Summer 2020, Fall 2020 at FORWARD Lab @ UIUC, by Dipro Ray and Shuhan Wang.

Previous Maintainers

Dipro Ray ([email protected])

Where do I find X?

• All source code files lie within .py files in the ./final_stuff/ directory. The latest version of the scripts are final_framework_v9_precision.py and final_framework_v9_recall.py. All previous versions (v1 through v9) are also in the same directory. Most of the later versions differ primarily in the kind of ranking function used.
• Metrics and associated scripts and ipynbs lie in the ./development_ipynbs/ directory. You can find the code to compute keyword precision and recall here.
• Input data, and output data are stored in ./final_stuff/data/ and ./final_stuff/outputs/ respectively.
• Archives can be found in the ./old/ subdirectory of each folder, as well as the ./metrics/ directory.

How do I run the framework?

If you'd like to use spaCy's GPU capability, make sure you have access to a GPU. (For Nvidia users, use nvidia-smi to check GPU info.) Then, install CUDA Toolkit 10.2 (note the version! spaCy/cupy isn't compatible with the latest 11.0 version yet, as far as I know.) Use nvcc --version to ensure CUDA drivers (as well as the correct version) have been installed. Execute pip install -U spacy[cuda102]. Then, try:

~$ python3
Python 3.7.4 (default, Aug 13 2019, 20:35:49)
[GCC 7.3.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import spacy
>>> spacy.require_gpu()
True

If the last line is "True" and returns no errors, you're good to go with respect to spaCy GPU compatibility. Before proceeding onto the next steps, execute python -m spacy download en_core_web_sm.

1. cd into the repository directory and run pip install -r requirements.txt
2. cd into the final_stuff subdirectory. This will be the main working directory henceforth.
3. final_framework_v9_precision.py or final_framework_v9_recall.py is the script to be run to execute the latest framework. (FYI: It uses helper scripts: prdualrank.py, wikiscore.py, extractor_helpers.py`.)
4. Place your input data in ./data/ directory. Make sure to sanitize the input! Lower case all text, and remove all non-alphanumeric text except periods.
5. An example command to run the script is: python3 final_framework_v6.py arxiv_titles_and_abstracts_short.txt 350 750 test_run.txt 9
   1. arxiv_titles_and_abstracts_short.txt: This parameters indicates that the input data file is ./data/arxiv_titles_and_abstracts_short.txt
   2. 350: The number of patterns to be extracted in each iteration
   3. 750: The number of keywords to be extracted in each iteration
   4. test_run.txt: This means that the output will be stored in ./outputs/test_run.txt
   5. 9: This refers to the scoring method (look in final_framework_v6.py for details). For now, this parameter is always set to 9.
   6. An extra parameter iter_num exists, but it needs to be set within the source code of final_framework_v6.py. It affects the number of iterations to be run.
6. As mentioned earlier, your results will be stored in the designated output file.
7. Push your results, with a meaningful file name that indicates date, time, framework version, scoring method, corpus (and ensure it's in the outputs directory).