A new AI-based recommendation system for preprints

Recent statistics show that the number of papers published in all fields of science is growing exponeitially. The majority of these new papers are published on preprint servers such as arXiv, biorxiv, medrxiv, etc… Researchers, to keep up with new research in their field and competition, need to read at least the title and abstracts of these published preprints and identify the ones that match their interests. As the number of new preprints can be very high (~1000 daily in exact sciences), the time requested just for identifying which papers could be interesting (not even reading the papers) can be very high.

 Tools presently available to try to identify the most relevant papers for a given researcher are in general based on keyword-based recommendation systems (e.g. google scholars’ alert system). The efficiency of these systems, although sometimes interesting, relies heavily on keywords manually entered by users (and, to some extent, by papers’ authors), with the danger of missing important part of the literature by simply not selecting the best keywords. In addition, some of these systems are also based on user’s feedback and are, in general, very bad at the beginning of their use.

The consequence of this situation is simple: many researchers do not devote the necessary time for scanning the most recent papers, and simply rely on conferences, network, social media, etc. in order to get to know new papers. This is highly inefficient, strongly biased, and can have, in the case of private sector research, strong economic consequences.

We developed a new AI-based tool that identifies for each registered users, the most relevant preprints matching their research interests. Once registered, our users receive each monday an email giving the three most relevant papers made available the previous week, as well as the list of all published papers in their (arXiv) field, ranked from the most interesting to the least interesting.