Skip to main content
4 minutes reading time (710 words)

Matchbook: A new recommender service to spark scientific collaboration

Blogpost by Tony Ross-Hellauer and Hrvoje Lucić, Know-Center, Graz, Austria
How can researchers, in the spirit of Open Science, move beyond their traditional networks to forge new connections? How can the new worlds of open information made possible by public infrastructures like OpenAIRE further enable Open Science by bringing the right organisations into partnership to stimulate innovation and address the grand societal challenges?

Know-Center is pleased to release Matchbook, a prototype recommender service to spark scientific collaboration, produced in partnership with OpenAIRE.

Our novel recommender service, funded via a small grant (14.5k EUR) from OpenAIRE’s Open Tender Calls, builds upon the OpenAIRE scholarly graph to enable organisations to keyword search for organisations with the strongest records of funding and collaboration success from the ever-expanding range of funders included in OpenAIRE. OpenAIRE’s information on funders and projects, enhanced with links to open access publications and data, is an invaluable resource that has not yet been fully exploited. Drawing information from OpenAIRE on institutions, projects and funders, Matchbook can help scientific institutions interested in forming consortia to identify potential partners with the exact disciplinary strengths and competences they need, rank them according to their previous success in securing funding, and specify searches according to specific national or international funders or even individual funding streams. The service uses Know-Center’s Scalable Recommender Framework to generate recommendations based on three different types of algorithms:
  1. Content-Filtering: recommends institutions based on keywords given by users. The titles and keywords of all OpenAIRE projects are collected and used to calculate a TF-IDF for each project, which is then used to rank results relative to a user’s desired keywords.
  2. Collaborative-Filtering: calculates recommendations for a given organization (based on its organisation id, for example:. Recommendations depend on the project collaboration (algorithm is looking for similar organizations based on common projects). Organization similarity and the probability that an organization will be recommended gets higher with the increase of common projects.
  3. Hybrid-Filtering: uses a combination of the two previous algorithms to recommend partners which not only fulfil specific keywords but also suit the profile of a given partner.
A provisional demonstrator interface for the service has been developed to showcase the service. Via this interface users can query the service using the three algorithms. Organisations (e.g., funders) can also embed the service in their own websites. Our recommender system provides three REST-based Web services (i.e., one service per algorithm) that enable easy customization via GET-parameters (e.g., to set a given partner or keywords). A Swagger-based GUI has been provided for conveniently testing the services and for allowing an easy integration into OpenAIRE by calling the services via JavaScript. Know-Center has also made our python code used to ingest data from the OpenAIRE API available for re-use.

Figure 1:  Matchbook demonstrator interface

The way ahead

Matchbook works as a successful proof-of-concept but would require more development before being able to be offered as a standalone production service – in particular, the recommendations are currently made on a relatively scarce dataset (i.e., just project titles and keywords, collaboration histories of institutions but not always individual departments) which means that performance suffers in comparison to similar recommender services based on richer information. A next step in this research would be to add enriched information - perhaps including text-mining of full project descriptions (also to the work-package level) and linking the publications linked to projects.

In addition, more work could be done to:
  • Add information to make more the algorithms more transparent and better explain how results were calculated and ranked.
  • Develop further use-cases to provide personalized recommendations of individual projects or individuals.
  • Adapt algorithms by considering beyond-accuracy objectives such as optimizing recommendation diversity and serendipity.
  • Integrate Matchbook with VIPER (The Visual Project Explorer), created by Open Knowledge Maps also as a result of the OpenAIRE open tender calls. VIPER is a unique open science application that provides overviews of research projects indexed by OpenAIRE. Know-Center is an organizational member of Open Knowledge Maps, and the two organisations collaborate closely together.
Despite these initial shortcomings, Matchbook has come a long way in a short time. The development team would be thrilled to hear from anyone with feedback, suggestions for future development or ideas for potential future collaboration.

If you have any such ideas, please contact the Matchbook lead Tony Ross-Hellauer.
Stay Informed

When you subscribe to the blog, we will send you an e-mail when there are new updates on the site so you wouldn't miss them.