Can we do better than Co-Citations? Bringing Citation Proximity Analysis from idea to practice in research paper recommendation
This event took place on Wednesday 11 October 2017 at 11:30
Scholarly publications are increasing exponentially every year, creating challenges for researchers to stay in touch with new relevant articles in their domain. Recommender Systems can serve dual purpose. To researchers, it can help them to stay in touch with the latest developments in their field. To authors, it can broaden their audiences resulting in increased number of reads and therefore more effective dissemination of knowledge. Citation Proximity Analysis (CPA) is based on the Co-Citation approach and the underlying heuristic of CPA is that the closer the documents are cited together the more likely they are related. In this work, a step by step scalable approach is developed for building CPA-based recommender systems. In this approach, three new novel proximity functions are introduced, extending the basic assumption of Co-Citation analysis to take the distance between the cocited documents into account. A CPA based recommender system was built from a corpus of more than 350,000 full-texts articles and a user survey was conducted to perform an initial evaluation. And, two of our three proximity functions used within CPA outperformed CoCitation based baseline approach by 25%.
Watch the webcast replay >>