The body of scientific literature is growing at an ever-increasing rate. Especially in the field of artificial intelligence (AI), the number of publications is growing every month with currently more than 200 new papers every day in my research area alone. The situation is similar in other fields of science such as astronomy with more than 41.000 new articles per year. As a result, it is increasingly difficult for researchers (like myself) to keep an overview over the current state-of-the-art with the consequence that nearly identical research is advanced in parallel, relevant references are overlooked, or important connections between methods are ignored. This hinders scientific progress as a whole and leads to a suboptimal usage of resources including research funds, compute, energy and intellectual capacity.

While a number of tools (Google Scholar, Semantic Scholar, Connected Papers, Research Rabbit, Arxiv Sanity, etc.) simplify literature retrieval today, they often use simple document representations (e.g., TFIDF, Co-Citation) and are limited to a high-level analysis in the form of graphs or publication lists. Missing is a deep semantic analysis of scientific articles which allows scientists to easily establish connections between the most relevant ideas and the researcher's own interests. Due to the rapid progress in the field of deep neural networks and natural language processing, machine learning opens up new possibilities to better understand, analyze and structure scientific work and thus make research more efficient across disciplines. While tremendous progress has been made in computational language models (e.g., GPT-3), it is surprising that these new tools remain largely unexplored to support researchers themself.

Supported by a Momentum grant of the VolkswagenStiftung, we develop novel models for natural language processing of scientific articles with the goal of supporting the research community at large by developing new algorithms for an intelligent scholar inbox which provides researchers with personalized recommendations about new publications and personalized summaries on a daily basis and analyzes new papers in relation to each scholar's own research.

Scholar Inbox: https://www.scholar-inbox.com/ 

Attempto article: https://uni-tuebingen.de/universitaet/aktuelles-und-publikationen/attempto-online/forschung/newsfullview-attempto-forschung/article/momentum-foerderung-der-volkswagen-stiftung-geht-gleich-drei-mal-an-die-universitaet-tuebingen/