How can young researchers find suitable conferences for their research project? Are there other conferences that fit a research project well, outside of the scope of well-established researchers?
These questions were tackled by three students of the University of Mannheim (Andreea Iana, Steffen Jung, Philipp Naeser), who created a recommendation system that suggests publication opportunities. The service aims to recommend possible conferences where a new paper can be submitted to, and is based on the Springer Nature Scigraph data.
The SN SciGraph information on past conferences and publications served as the main dataset used for training the recommender system. Additionally, this past-oriented dataset has been enriched with data on future conferences, their submission dates, and other related information mined from WikiCFP, a Web page which publishes calls for papers, in order to provide information on upcoming conferences.
The authors implemented different recommendation techniques based on various features of a paper, such as the author names, abstract, or keywords. The findings show that the best results are achieved by using keywords, in particular the product market codes in SciGraph. However, since these keywords are often not easy to select by laymen users, recommendations based on the paper abstract provide a user-friendly alternative with slightly lower performance.
While the results are promising, further improvements of the models and alternatives of utilizing the information in SciGraph for recommendations can be experimented with. For example, a graph-embedding model could be trained on the whole SciGraph in order to create recommendations based on the proximity in the graph. Moreover, additional features, such as citations contained in a paper, could be included in the system. The main limitation of the current approach is that the system can only recommend conferences contained in SciGraph, and thus, excludes conferences with proceedings not published by Springer (e.g. ACM, IEEE). Similarly, new conferences not yet contained in SciGraph cannot be recommended. To address these limitations, further work could focus on utilizing additional features from WikiCFP, as well as on changing the recommendation target and interplay between the SciGraph and WikiCFP datasets, namely by recommending CFPs in WikiCFP, and by using the SciGraph data for only training the models.
A prototype to test the system will be available soon.
Please sign in or register for FREE
If you are a registered user on Research Data at Springer Nature, please sign in