/ItemSubjector

CLI-tool to easily add "main subject" aka topics in bulk to groups of items on Wikidata

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

ItemSubjector

The purpose of this command-line tool is to add main subject statements to Wikidata items based on a heuristic matching the subject with the title of the item. bild The tool running in PAWS adding manually found main subject QIDs bild Itemsubjector running GNU Screen on a Toolforge bastion with --limit 100000 and --sparql matching the WHO list of essential medicines.

Background

As of september 2021 there were 37M scientific articles in Wikidata, but 27M of them were missing any main subject statement. That makes them very hard to find for scientists which is bad for science, because building on the work of others is essential in the global scientific community.

To my knowledge none of the scientific search engines that are currently used in the scientific community rely on an open graph editable by anyone and maintained by the community itself for the purpose of helping fellow scientists find each others work. Wikipedia and Scholia can fill that gap but we need good tooling to curate the millions of items.

Caveat

This type of matching that ONLY takes the label and not the underlying structured data into account is SUBOPTIMAL. You are very welcome to suggest or contribute improvements so we can improve the tool to help you make better edits.

Features

This tool has the following features:

  • Adding a list of manually supplied main subjects to a few selected subgraphs (These currently include a total of 37M items with scholarly items being the biggest subgraph by far).
  • Matching against a set of items fetched via a SPARQL query.
  • Matching up to a limit of items which together with Kubernetes makes it possible to start a query which collects jobs with items until the limit is reached and then ask for approval/decline of each job. This enables the user to create large batches of jobs with 100k+ items in total in a matter of minutes.
  • Batch mode that can be used together with the above features and be run non-interactively e.g. in the Wikimedia Cloud Services Kubernetes Beta cluster

It supports Wikidata:Edit groups so that batches can easily be undone later if needed. Click "details" in the summary of edits to see more.

Installation

Download the latest release with:

$ pip install itemsubjector

Alternative installation in venv

Download the release tarball or clone the tool using Git.

Clone the repository

git clone https://github.com/dpriskorn/ItemSubjector.git && cd ItemSubjector

Then checkout the latest release.

git checkout vx.x.x where x is the latest number on the release page.

Setup the environment

Make a virtual environment and set it up using poetry. If you don't have poetry installed run: $ pip install poetry

and then setup everying with

$ poetry install --without=dev

to install all requirements in a virtual environment.

PAWS

Note: PAWS is not ideal for batch jobs unless you are willing to keep your browser tab open for the whole duration of the job. Consider using Kubernetes instead, see below

The tool runs in PAWS with no known issues.

  • log in to PAWS
  • open a terminal
  • run git clone https://github.com/dpriskorn/ItemSubjector.git .itemsubjector && cd .itemsubjector && pip install poetry && poetry install --without=dev <- note the dot in front of the directory name that hides it from publication which is crucial to avoid publication of your login credentials.
  • follow the details under Setup below

Wikimedia Cloud Services Kubernetes Beta cluster

Note: this is for advanced users experienced with a SSH console environment, ask in the Telegram WikiCite group if you need help

See Kubernetes_HOWTO.md

Setup

Setup the config by copying config/config.example.py -> config/init.py and enter the botusername (e.g. So9q@itemsubjector) and password (first create a botpassword for your account and make sure you give it the edit page permission and high volume permissions)

  • e.g. cp config_example.py config.py && nano config.py

GNU Nano is an editor, press ctrl+x when you are done and y to save your changes

Use

This tool helps by adding the validated or supplied QID to all scientific articles where the search string appears (with spaces around it or in the beginning or end of the string) in the label of the target item (e.g. scientific article).

Adding QIDs manually

Always provide the most precise subjects first

Run the script with the -a or --add argument followed by one or more QIDs or URLS:

  • poetry run python itemsubjector.py -a Q108528107 or
  • poetry run python itemsubjector.py -a https://www.wikidata.org/wiki/Q108528107

Note since v0.2 you should not add subjects that are subclass of each other in one go. This is because of internal changes related to job handling

Add the narrow first and then the broader like this:

  • poetry run python itemsubjector.py -a narrow-QID && poetry run python itemsubjector.py -a broader-QID

Please investigate before adding broad subjects (with thousands of matches) and try to nail down specific subjects and add them first. If you are unsure, please ask on-wiki or in the Wikicite Telegram group

Disable alias matching

Sometimes e.g. for main subjects like Sweden it is necessary to disable alias matching to avoid garbage matches.

Usage example: poetry run python itemsubjector.py -a Q34 --no-aliases (the shorthand -na also works)

Disable search expression confirmation

Avoid the extra question "Do you want to continue?":

Usage example: poetry run python itemsubjector.py -a Q34 --no-confirmation (the shorthand -nc also works)

Show links column in table of search expressions

This is handy if you want to look them up easily.

Usage example: poetry run python itemsubjector.py -a Q34 --show-search-urls (the shorthand -su also works)

Show links column in table of search expressions

This is handy if you want to look them up easily.

Usage example: poetry run python itemsubjector.py -a Q34 --show-item-urls (the shorthand -iu also works)

Matching main subjects based on a SPARQL query.

The tool can create a list of jobs by picking random subjects from a users SPARQL query.

Usage example for diseases: poetry run python itemsubjector.py -iu --sparql "SELECT ?item WHERE {?item wdt:P31 wd:Q12136. MINUS {?main_subject_item wdt:P1889 [].}}"

This makes it much easier to cover a range a subjects. This example query returns ~5000 items to match :)

Batch job features

The tool can help prepare jobs and then run them later non-interactively. This enables the user to submit them as jobs on the Wikimedia Cloud Service Beta Kubernetes cluster, so you don't have to run them locally if you don't want to.

See the commands below and https://phabricator.wikimedia.org/T285944#7373913 for details.

Note: if you quit/stop a list of jobs that are currently running, please remove the unfinished prepared jobs before preparing new jobs by running --remove-prepared-jobs

List of all options

See $ poetry run python itemsubjector.py -h for all options.

What I learned

  • I used the black code-formatter for the first time in this project and it is a pleasure to not have to sit and manually format the code anymore.

  • I used argparse for the first time in this project and how to type it properly.

  • This was one of the first of my projects that had scope creep. I have removed the QuickStatements export to simplify the program.

  • This project has been used in a scientific paper I wrote together with Houcemeddine Turki

Rewrite 2022:

  • Important to break down methods to 1 method 1 task to increase readability. -> helps reuse in other projects.
  • Important to avoid resetting attributes and instantiate classes instead. -> helps reuse in other projects.
  • Simplify as much as possible to keep the whole thing lean and avoid scope creep. -> helps reuse in other projects. (KISS-principle)
  • Difficult to judge which features are used and which are not. User testing would be nice.
  • UML diagrams are nice. They give a good quick overview.
  • Removing options that no-one seems to use helps keeping it simple. It would be valuable to get better insight of how the program is used by the users. A discussion in github might help in this.

Thanks

During the development of this tool the author got a help multiple times from Jan Ainali and Jon Søby with figuring out how to query the API using the CirrusSearch extensions and to remove more specific main subjects from the query results.

A special thanks also to Magnus Sälgö and Arthur Smith for their valuable input and ideas, e.g. to search for aliases also and to Jean and the incredibly helpful people in the Wikimedia Cloud Services Support chat that helped with making batch jobs run successfully.

Thanks also to jsamwrites for help with testing and suggestions for improvement and for using the tool to improve a ton of items :).

License

GPLv3+