eWEning star has been around for few months now and we are already busy. I am very much impressed by the local academic community in Groningen as several collaborations have already been established!
Computational Linguistics of the Faculty of Arts
One of the partners will be the Information Science group with the lead of an absolutely magnificent scientist (and person!) Malvina Nissim who has not only been elected as the University of Groningen Lecturer of the Year 2016, but also gave a very insightful TEDx talk about Serendipity beyond stereotypes. Her examples gave me the first sparks of getting interested in AI and understanding its potential when used in a smart way and for greater good (so forget all the doomsday stories of AI).
As eWEning star wants to build a platform where the user finds the relevant documents faster, it is important to have an understanding of the content of each paper. If we would assess each paper independently, it would take us probably years, even when just focusing on papers related to wind industry (which will be the focus of eWEning star in the Proof-of-Concept phase).
Therefore, teaching AI to categorise the papers according to our preference can have a tremendous effect. But just like a child, AI also learns through examples: we need to show it what a paper with, for example, “economic-aspect” looks like. After it has learnt it (and other STEER aspects), we can feed it a LARGE amount of papers and it can categorise them much faster than humans ever could.
When we are able to categorise the papers in different ways, we are able to open a flexible “menu” for the users: “What types of documents are you interested in? Just let us know the boundaries and we will get you all that are available.” And THAT makes all the difference compared to using search based on keywords.
The problem was well concluded by Angers A. and Petrillo M. in their paper “Clustering and classification of reference documents from large-scale literature searches” in 2017 :
“Finding relevant scientific information in the published literature is an ever-growing challenge, as the amount of articles and reviews published increase constantly. Broad searches can produce a large amount of hits that can be difficult to process. On the other hand, focused searches can produce manageable numbers of results, with the risk of missing key references and introducing bias depending on the terminology used.”
How much faster could we get forward when we would have access to the relevant information immediately? If different stakeholders would know faster what is already studied and known by academia and industry, it would be easier to make the NEXT step in the energy transition, also by policy makers and investors who rarely are the actual experts of the latest developments.
Science, Business and Policy program
As I am a big fan of cross-industrial and multidisciplinary topics, it was fantastic to run into this program at the University of Groningen, with the guidance of Campus of Groningen. I have a solid trust that understanding different perspectives will increase industries’ chances on innovation and development as mentioned in an earlier blog. Multidisciplinary approach (or rather interdisciplinary, if you are into details in definitions) has also been implemented by other universities such as MIT.
So I am only delighted that in the SBP Program one of the tasks is to collaborate with companies and establish projects where students get to find solutions for real challenges or questions that companies have. For eWEning star the project-team of students will assess market potential of the digital platform: How many universities, research institutes and other stakeholders are busy with Renewable Energy? Considering how huge the industry already is and how it is expected to grow, the need for efficient information search can only be expected to increase.
Five superheroes working in categorisation
eWEning star is currently planning to have three different categories for the papers, with total of 15 different aspects to build the Proof-of-Concept. Considering that minimum 100 papers are needed for each aspect in order to teach the AI at this stage to recognize the differences between them, some good amount of papers are involved. This is a task too big to handle alone, so the papers are getting manually assessed by 5 additional superheroes. Each paper will be assessed by two different heros in order to make sure the same aspects are recognised from the same paper. The “child AI” requires consistency in its training, just like real children.
As mentioned in our website, we are open for collaborations regarding categorising research papers. In case your organisation wants to be part of this initiative with your own papers, please take contact.