Multi-Entity Extraction Framework for Academic Documents (with default extraction tools)
To create a new index using the existing index mapping in Elasticsearch, you can use the following Elasticsearch query:
PUT /new_index
{
"settings": {
"index": {
"number_of_shards": 1,
"number_of_replicas": 0
}
},
"mappings": {
"_doc": {
"properties": {
// Copy the mapping properties from the existing index here
}
}
}
}In the above query, replace new_index with the name of the new index that you want to create. The settings section specifies the number of shards and replicas for the new index. The mappings section specifies the mapping properties for the new index.
To copy the mapping properties from an existing index, you can use the following Elasticsearch query:
GET /existing_index/_mappingThis will return the mapping properties for the existing_index. Copy the properties section and paste it into the mappings section of the PUT query for the new_index.
Note: You can also copy the entire mapping from the existing index and use it to create the new index, but this may result in unnecessary fields and properties being included in the new index. It's recommended to copy only the necessary mapping properties
-
Set the appropriate settings in /src/extractor/python_wrapper/properties.config and pdfmef/src/extractor/csxextract/config.py
-
Go to /src/extractor/ and run
python main.py
-
Delete a paper from an index. Use the script under
script/delete_papers.py.
The dockerfile in /docker enables setting up base image based on Ubuntu 18.04. While most of the required dependencies will be installed by this method, there will still be some manual configuration needed to get the extractor running. Refer Dependencies section for more details. Below are the required commands to get the docker setup running.
Change to docker directory
cd docker
Build the Docker Image from dockerfile using
docker build -t pdfmef-image:latest .
Spin up a container from the above image
docker run -it -p 8888:8888 -v <shared-dir-host>:<shared-dir-container> pdfmef-image bash
The above command will run an ubuntu base image with most of the required dependencies pre-installed. All the code will be found in folder /pdfmef-code within container. Additionally, one can share ports using -p option and directory using -v option as depicted in the above command. Here <shared-dir-host> represents absolute path of shared directory in the host and <shared-dir-container> is path of shared directory in container.
- Python 2.7 (make sure to use pip2.7)
- subprocess32 package (
pip install subprocess32 --user) - xmltodict (
pip install xmltodict --user) - MySQLdb (
pip install mysqldb-rich) - defusedxml (
pip install defusedxml) - requests (
pip install requests)
- Clone this repo to your local machine anywhere
- From the project src directory, run
python setup.py install --user
(The --user option is optional, I just like to install packages only for my user account personally)
Run, from the extraction framework root directory:
python -m extraction.test.__main__
If using Python 2.7 you can run more simply:
python -m extraction.test
- extraction framework python library (on python path (run
python setup.py install --userfrom its root directory) - defusedxml python library (run
pip install defusedxml --userto install) - requests python library (run
pip install requests --userto install)
Grobid is used to extract header information from the PDF files. Grobid should be running as a service somwhere. (Run ./gradlew run from grobid main directory if not already running. See Grobid's Github project for more complete installation instructions.) The URL for Grobid can be configured in csxextract/config.py.
PDFBox is used to get a plain text representation of the PDF files. The PDFBox jar needs to be on the machine somewhere. The default expected location is ~/bin but this can be configured in csxextract/config.py.
The PDF Classifier .jar file is used to classify PDFs as academic or not. It should be on the local machine somewhere, as well as its associated acl file and training data file. The paths to these three files can be configured in csxextract/config.py. These files an be found in resources/classifier.
ParsCit is used to extract citation information from the PDF files. The path to it's citeExtract.pl script should be configured in csxextract/config.py. The default expected location is ~/bin/pars_cit/bin/citeExtract.pl.
Installation of ParsCit can be tricky. See its INSTALL doc for full instructions details. Also important is the Troubleshooting page which has answers for common problems.
A message like "Can't locate XML/Twig.pm in @INC (@INC contains: ...)" means that a Perl library is missing. Missing libraries can be installed with cpan.
Also note the question "When running citeExtract.pl I get some errors complaining about the wrong ELF class of the binaries. How can I fix this?" After Step 1 in the install instructions, the following commands should be run:
$ cp -Rf * ../../.libs
$ cp crf_learn ../../.libs/lt-crf_learn
$ cp crf_test ../../.libs/lt-crf_testFinally, the step marked as "optional" in the install instructions might actually be necessary when installing ParsCit.
pdffigures2 is used to extract figures and tables from PDF files along with related figure and table metadata. It should be installed as directed by the pdffigures2 Github page. The path to the pdffigures2 binary can be configured in csxextract/config.py
A Java jar is used to extract algorithms from PDFs. The required files can be found at resources/algextract/. Store the algo_extractor.jar file and the perl/ directory on your system where you wish. Then, in csxextract/config.py, set the ALGORITHM_JAR_PATH variable to the location of the algo_extractor.jar and the ALGORITHMS_PERL_PATH variable to the location of the perl/ directory.
Finally, make sure you have the Lingua::Stem Perl module installed. This can be installed with cpan.