How to run an user-defined instance of Reviewernet.org
This is a tutotial to download, create and deploy a user-defined instance of Reviewernet.org. This folder contains both Bash and Python scripts. The only file to run in order to have your instance of RN is download_and_parse.sh.
To run the script succesfully you need:
100 GB of free space on disk to store corpus gzipped partitions (corpus details in the next section)
wget, to download the partitions
python2.7 or higher, for parsing phase
fuzzywuzzy, a fuzzy string matching python library. If python-Levenshtein is also installed, fuzzywuzzy uses Levenshtein Distance (up to 4-10x speedup)
The reference corpus can be found at http://api.semanticscholar.org/corpus/download/. This representation of the full Semantic Scholar corpus offers data relating to papers crawled from the web and subjected to a number of filters.
The papers are provided as a set of json objects, one per line. Papers are grouped in batches and shared as a collection of gzipped files; each file is about 660 MB, and the total collection is about 100 GB.
The provided script, given a list of journals/venues, downloads and parses the partitions in parallel, allowing a maximum number of 10 concurrent downloads and parsers. The only input needed is a set of journal/venue names (see next section for further details on usage), and in the end the output will be:
a folder named datasets which contains the files needed to run a ReviewerNet session.
180 corpus partitions as gzip files
at most 180 parsed files
The completion time of the whole process with a speed connection of 1MB/s is about 10 hours^.
Once exectued, the script checks the presence of the needed libraries, asking the user to install fuzzywuzzy if not present; then the latest corpus manifest is downloaded.
At this point the download&parse process is started: each file listed in the manifest is downloaded - as a gzip - and filtered. The resulting files are stored with the -filtered suffix.
When all partitions have been downloaded and filtered the merging phase can start. Filtered corpus partitions are merged together and the personalized datasets are built.
Eventually the script asks the user whether to delete or not the interemediate files (gzipped/filtered partitions).
At this point he datasets folder will contain the 3 files needed to run ReviewerNet, ending with _pers suffix.
open journals.txt with a text editor and change the content of the JSON object with the names^^ of the journals/venues that will be used to build the topic-based datasets.
execute download_and_parse.sh manifest_link
run a local/remote ReviewerNet session and click on Upload instance to upload the files you've just created;
Use your own instance of ReviewerNet!
^The completion time has been measured on a dual-core laptop connected to a network with 2MB/s bandwidth.
^^Use always https://www.semanticscholar.org/ as reference for journal names and available papers because the same journal/venue is referenced in different ways across different papers. We suggest to put at least three string for each journal in order to have better coverage in the fuzzy search.[see journals.txt for an example]