PLOS Cloud Explorer
The open access (OA) publishing movement promises to enrich the public domain with knowledge and scholarship previously confined to restricted-access journals. Despite the major potential benefits of OA to the public understanding of science, academic journals are still largely inaccessible on a cultural level. Here we focus on the challenge of getting a summary-level picture of a particular area of research. Our goal is to create an experience that allows both researchers and curious members of the general public to explore research trends and interactions between research topics. We are not aware of any current interactive tool that facilitates such interactions.
We used open-access content published in the Public Library of Science (PLOS) journals. The PLOS Search API, which provides access to full article data and metadata, allowed us to extract article titles, abstracts, publication dates, and subject areas (keywords) that are part of the PLOS Thesaurus.
Our initial plan was to present word clouds drawn from aggregated abstract text for a chosen subject area. We thought abstracts, which are summaries of articles, would reflect the research topics in their text. However, we encountered difficulties when we realized that the most frequent words in scientific literature are not necessarily topical.
For an detailed account of our process up to this point, with code examples for every step, see this notebook.
Change of Course
After exploring the data obtainable through the PLOS API, we saw that each article is associated with a rich set of subject terms, and the position of those terms within the PLOS subject area polyhierarchy (i.e., thesaurus) is also specified. We realized that these subject terms constitute meaningful and well-organized textual and structural information, which we could work with much more fruitfully than with abstract text.
Since the PLOS website interface currently only allows users to view the subject hierarchy either through a large list or individual branches, we decided to create an interactive tree that illuminates a larger structural context of the relationships between research areas.
In addition to the tree visualization, we also compute a word cloud for each subject area, which shows what else those articles that are tagged with that subject are about. This is computed by counting the frequencies of other subject terms in the set of articles that contain the selected subject area.
The tree and word cloud visualizations can be interactively filtered by restricting the publication date range.
What's behind the scenes
We downloaded bulk data from PLOS API
In order to collect a full dataset from the PLOS Search API, we used a Python script to download data for ~120 thousand acticles -- all articles ever published in PLOS journals and accessible through the API -- while respecting the limits on the frequency and number of API calls. After some trial and error, a successful bulk download took only three hours.
The resulting dataset (last updated: 2014-07-21) contains the following for each article:
- article DOI
- journal name
- publication date
- abstract text (not all articles)
- subject areas
These data describe article subject areas in a very useful way. The "subject area" field is a list of strings, and each string encodes a path through the polyhierarchy. For example:
[u'/Computer and information sciences/Information technology/Data processing', u'/Computer and information sciences/Information technology/Data reduction', u'/Biology and life sciences/Ecology/Biodiversity', u'/Ecology and environmental sciences/Sustainability science', u'/Biology and life sciences/Organisms/Animals/Vertebrates/Mammals', u'/Computer and information sciences/Computer networks', u'/Computer and information sciences/Computing methods/Cloud computing', u'/Ecology and environmental sciences/Ecology/Biodiversity', u'/Biology and life sciences/Organisms/Animals/Vertebrates']
The entire PLOS subject area polyhierarchy was kindly provided to us as a spreadsheet with ~17 thousand rows, one node per row. It is now also available on GitHub.
We transformed the data into two JSON objects
We transformed the article data (a ~400 MB pickled DataFrame) into a JSON object containing a list of articles (indexed by DOI) and the data about them. Due to the large file size, we excluded several pieces of information that we weren't immediately using in our data visualization (though we could add these back later). Currently, each article in the JSON object has the publication date (truncated to year only) and the list of subject area paths that were provided by the API. In addition, for each article, we calculated the set of top-level (root) and lowest-level (leaf) terms that appear in its subject area field. We explain how these are used below.
We transformed the subject area polyhierarchy from its spreadsheet representation into a JSON object with a tree structure. For each node in the tree, we also calculated the number of articles that include that node in their subject area paths (using string search). This allows us to visualize the "size" of each node in the tree.
We used D3.js, JQuery, and Dimple.js to create a dashboard for interactive exploration of the PLOS data. This dashboard presents two control elements:
- an interactive SVG visualization of the subject area tree, allowing sub-trees to be selected;
- a graph of the total number of articles over the time, allowing a specific time span to be selected.
For the selected time span and subject area, the dashboard displays:
- a bar graph of the number of articles per year matching any of the subjects under that node;
- a histogram of the distribution of top-level subject terms;
- a word cloud composed of the leaf-level subject terms associated with all the articles.
Try the web app! (On MacOS, works best on Chrome. On Windows, try Firefox.)
Our visualization reveals interesting things about the research articles in PLOS journals. By selecting a subject area node, you can see from the word cloud and the histogram that research areas are highly interconnected. You can observe and explore trends in the number of articles over time for a given set of subjects (using the time series graphs), and also trends in the associations among subjects (using the histogram and word cloud).
Reproducing our results
You can reproduce our work with the following steps.
- Clone our Git repository.
- To run the web app, execute
python -m SimpleHTTPServerin a shell from the root directory of the repository, and visit http://localhost:8000/ in your browser. Boom!
- If you want to reproduce our data collection and analysis, first get a PLOS API Key and make a file called
ipython_notebooks/settings.pywhich contains the statement
PLOS_KEY = u'your_key'. (Replace
your_keywith your key.)
- Make sure you have all the packages listed in
- Run the Python code in the notebooks linked above, under What's behind the scenes. Note that this will take many hours due to the volume of data.
- You may want to download the latest version of the PLOS Thesaurus spreadsheet, in order to generate an up-to-date subject area tree. PLOS periodically updates the Thesaurus.
- You'll need to generate an article data set using the code for making API calls.
- Then you'll need to regenerate the JSON objects.
We would like to be able to show users specific articles that match the subject areas and time ranges that they are exploring. We would need to reintroduce some of the article metadata that we filtered out to reduce the size of the dataset: article titles, authors, and journals.
List of technologies/packages used
- IPython notebook
Who made this? Who did what?
PLOS Cloud Explorer is by Anna Swigart, Colin Gerber, and Akos Kokai.
Anna worked on bulk data download, communication with PLOS staff, web app coding, and web app design. Colin worked on bulk data download, article data analysis, web app coding, and web app design. Akos worked on article data analysis, article & thesaurus data transformation, and documentation.
This program is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.
A copy of the GNU Affero General Public License is included in this repo.
All PLOS content (article data, subject area thesaurus) is licensed under the Creative Commons Attribution (CC BY) license.