Tackling complex real-world problems needs interdisciplinary research and an open and easy to navigate knowledge landscape.
kmapper is a related article tool supporting interdisciplinarity—developed by me.
Using IBM Watson's NLU service, kmapper analyzes research articles to extract key DBpedia concepts. An article's top concepts are used to find related content across different scientific disciplines and the results are visualized by the kmapper map. Working with high-level semantic concepts avoids discipline-specific vocabulary and leads to the discovery of interconnections across research from very different disciplines.
The article of interest is located in the middle of the map. Each colored line represents a scientific discipline or unique combination of disciplines among related articles. Related articles with a higher degree of relatedness are located closer to the article of interest than less related articles.
Currently, there is an HTML snippet at the bottom of each article page which let’s you embed a preview image of the according kmapper visualization on any website (e.g. at the original article URL).
The snippet for above article example would look as follows:
<script id="kmapper_widget" src="https://kmapper.herokuapp.com/embed.js" doi="10-5585-geas-v4i3-387" width="240" height="240"></script><div id="kmapper_badge"></div>
And this is what the according preview image embedded on a website would look like:
kmapper also allows to map a user's own text against currently kmapper-indexed articles to find related research (please write an e-mail to email@example.com for access credentials and testing the approach).
See an example map for text about Sustainable Development Goal 17 here.
The current prototype consists of a corpus of 589,465 open access articles with English abstracts from a total of 3,997 different journals - obtained from the Directory of Open Access Journals. 126,398 unique concepts have been extracted from the abstracts using Watson's NLU service. Each article-concept relation has a local relevance score and for each unique concept a global relevance score has been calculated based on the concept's frequency in the corpus. For simplicity, scientific disciplines of articles have been derived from the journal's LCC classification, resulting in 1,058 unique combinations of scientific disciplines.