Speaker: Josef Glas (192-02 Computer Graphics)
Students and researchers regularly need to review and summarize the state-of-the-art of a given research direction for papers, proposals, or state-of-the-art reports. In a systematic review process, researchers parse all papers of a given set of digital libraries and include or reject papers for further consideration based on some more or less ill-defined criteria. From these papers, relevant information is extracted, categorized, and summarized. This is mostly done manually. The goal of this work is to design and implement a prototype for an innovative research tool that helps users to discover and categorize text fragments of scientific papers. The tool should provide a smart user interface using semi-supervised machine learning that gradually learns what aspects the user is interested in and recommends suitable text fragments that fit these categories. This goes beyond current approaches, which focus mostly on similarities on document level. With regard to methodology there are two key challenges that is first, designing an appropriate topic modelling approach by combining supervised and unsupervised techniques, and second, designing a suitable User Interface.