Iryna Gurevych. Gerhard Weikum and Prof. Simon Razniewski. During my PhD, I worked on discovering informative negative statements from encyclopedic and commonsense open-world knowledge bases.
Jeff Z. Daria Stepanova where I worked on repairing inconsistent knowledge bases using pretrained language models. Current topics of interest: AI for mental health, analyzing the impact of research papers, commonsense knowledge, cultural NLP. Previous topics of interest: searching and curating web-scale knowledge graphs, mining salient negative knowledge.
Links Google Scholar - Resume. Simon Razniewski, Prof. Aditya Bikram Biswas Hiwi , Mang Zhao Hiwi , Volunteer: VLDB ' Identified from Wikidata, we construct a datasets about 7. Every community comes with interseting findings such as demographic data and exceptional members. We present a web portal to showcase the Uncommonsense system. Users can browse interesting negative statements about every day concepts such as elephant, pancake, and acne.
We introduce UnCommonSense, a method for discovering expressive negative statements about everyday concepts. The method significantly outperforms the state-of-the-art on informativeness and recall. We present a method to repair incorrect statements in existing knowledge bases by replacing incorrect triples with likely correct ones, thus avoiding information loss. Our method explores the power of LM probes and shows that context retrieval from the knowledge base itself can significantly boost the probing.
We extend previous methods on negation inference by introducing the order-oriented peer-based inference method, which shows an improvement in informativeness. We review Wikidata's attempts to allow negative knowledge and discuss the gains challenges arising from implementing a negation-inference system.
