Identify relevant scientific papers with simple machine learning techniques
Copy shakespeare.py, data and content_sources to your pythonpath.
To intsall an example knowledge set, copy examples' contents to $HOME/.shakespeare
Depends on bibtexparser, feedparser scikit-learn packages, which can be installed via pip
pip install bibtexparser scikit-learn feedparser
-
fetch functions for the following journals
- Phys Rev A-X
- PRL
- PNAS
- Nature + Nature:Stuff
- Science
- Small
- ACS Nano, Nano Letters
- Soft Matter
- Langmuir
- Angewandte Chemie
- JCP, JCP B
-
Fetch functions for arXiv
-
support for BibTex Files
-
Naive bayes training and classification
The very first thing to do is to let the code know where 'bad stuff' is
./shakespeare.py -g good.bib -k examples/ --overwrite-knowledge --train
Train naive_bayes algorithm
./shakespeare -g thegoodstuff.bib -b thebadstuff.bib -k examples --train
Find papers from nature nano and PNAS
./shakespeare.py -j natnano pnas -o cool_papers.md
Find papers from the arxiv cond-mat.soft and math, then review the algorithms selection
./shakespeare.py -a cond-mat.soft math --feedback
Help printout
usage: shakespeare.py [-h] [-o OUTPUT] [-b [BIBFILES [BIBFILES ...]]]
[-j [JOURNALS [JOURNALS ...]]] [-a [ARXIV [ARXIV ...]]]
[--all_sources] [--all_good_sources] [--train]
[-g GOOD_SOURCE] [-m METHOD] [-k KNOWLEDGE]
[--overwrite-knowledge] [--feedback] [--review_all]
optional arguments:
-h, --help show this help message and exit
-o OUTPUT, --output OUTPUT
output file name. only supports markdown right now.
-b [BIBFILES [BIBFILES ...]], --bibtex [BIBFILES [BIBFILES ...]]
bibtex files to fetch
-j [JOURNALS [JOURNALS ...]], --journals [JOURNALS [JOURNALS ...]]
journals to fetch. Currently supports physreve
physrevd jchemphysb physreva physrevc pnas nature
jchemphys science natmat physrevb acsnano jphyschem
nanoletters natphys prl small angewantechemie langmuir
physrevx natnano.
-a [ARXIV [ARXIV ...]], --arXiv [ARXIV [ARXIV ...]]
arXiv categories to fetch
--all_sources flag to search from all sources.
--all_good_sources flag to search from good sources. Specfied in your
config file.
--train flag to train. All sources beside "--train-input-good"
are treated as bad/irrelevant papers
-g GOOD_SOURCE, --train_input_good GOOD_SOURCE
bibtex file containing relevant articles.
-m METHOD, --method METHOD
Methods to try to find relevent papers. Right now,
only all, title, author, and abstract are valid fields
-k KNOWLEDGE, --knowledge KNOWLEDGE
path to database containing information about good and
bad keywords. If you are training, you must specifiy
this, as it will be where your output is written
--overwrite-knowledge
flag to overwrite knowledge,if training
--feedback flag to give feedback after sorting content
--review_all review all the new selections. Otherwise, you will
only review the good selections
- Train a bunch and see if this is worth any more time
- Make an nice installer
- Add support for a config file for setting defaults (which journals to search, etc)