Can an algorithm show you who to collaborate with?

researchcollaborationCollaboration is widely seen as a good thing (when it’s done well at least), but knowing who can add value to your work is often tricky.  There have been some attempts to employ serendipity to the occasion, with tools such as Lunch Roulette aiming to match employees up at random for a lunch date.

Other systems take a slightly more scientific approach.  For instance, I covered a paper earlier this year that used an algorithm to match up delegates at a conference.  Users of the system were broadly positive, with around 50% revealing that they had struck up productive partnerships as a result.

Another paper has set out to explore the same issue, but this time with research scientists the target of their attentions.  The paper focuses on the use of research networking systems (RNSs) that aim to model the research community and suggest suitable collaborators.

Now, I should begin by saying how valuable research collaborations are.  A recent paper highlighted just how frequent, and indeed how global, research collaborations are now.  The study found that the scientific impact of a paper rose when the authors were from a number of different nations.

So can RNSs help this?  The research set out to explore how researchers use such tools, and to use this knowledge to construct a functional collaboration finder tool.

The team developed a number of prototypes to test their hypothesis.  Prototype 1 relied upon the researchers existing contact list, attempting to match these contacts with publication and author information found within an RNS.

The 2nd prototype relied more on seniority as its guiding light.  So, for instance, junior researchers would be more likely to team up with those of similar status.  The prototype used metrics such as number of publications, grant funding and so on to build up its user profiles.

Each prototype was trialled amongst the users enrolled onto the study, with each asked a series of questions about their experience, before being asked to complete a number of search related tasks, using both the prototypes and their preferred methods of searching.

Results

It emerged that researchers typically deploy a range of strategies for finding willing and useful collaborators, with the average researcher collaborating with 32 others.  Most of these were found through existing social networks.

Whilst networking lore suggests this might not be that effective, a study from earlier this year revealed that might not be the case.  The paper found that weak ties were not much use when it came to collaboration.

When weak ties existed between colleagues, whether along goal or personal orientated lines, there was shown to be no significant impact upon the performance of their team.  That was certainly not the case with stronger ties however, with a clear link between the strength of the ties and the performance of the team.

When solving problems in a competitive environment, the study revealed, it does not matter how many people someone knows or networks with — what really matters are the strongest ties in the network. This has implications for the organization of teams of scientists, engineers, and a host of others tackling today’s most complex problems.

Maybe that is something for the RNS developers to consider.

Related

Facebooktwitterredditpinterestlinkedinmail

One thought on “Can an algorithm show you who to collaborate with?

Leave a Reply

Your email address will not be published. Required fields are marked *

Captcha loading...