Mohamed Yahya (original) (raw)

Senior Software Engineer & Research Scientist
AI Group, Bloomberg

Formerly: Doctoral student @ Max Planck Institute for Informatics,
advised by Gerhard Weikum

🔥 في 08/11/2019 أطلقنا العنصر 14، قناة يوتوب علميّة عن تكنولوجيا الحاسوب في حياتنا اليوميّة. المزيد ...

I am broadly interested in the problem of facilitating user access to knowledge hiding within structured and textual data. My work lies at the intersection of NLP, IR and Databases. I worked on various approaches to executable semantic parsing for question answering over large knowledge graphs [EMNLP 2012, CIKM 2013, WWW 2017, EMNLP 2017]. Because structured sources are never complete and QA technology will always be imperfect when translating questions into some formal meaning representation, .. I've also looked at the problem of_combining structured and textual data_ to improve the robustness of question answering [EMNLP 2014, WSDM 2016, VLDB 2016, EDBT 2019]. I've also worked on the problem of deploying QA systems starting with very little training data through a never ending learning approach [WWW 2018]. To help improve the state of the art in factoid QA to match what actual users ask, I helped release ComQA, a dataset of factoid questions gathered from community QA website, annotated with answers and paraphrase clusters [NAACL 2019]. Recently, I've been interested in the problem of improving the usability of QA systems based on semantic parsing through semantically driven auto-completion. The resulting completion systems complete to those and only those questions that the underlying QA system understands, facilitating both discovery and expectation management [SIGIR 2018, arXiv 2019]. Finally, because all work and no play makes one a dull boy, I played with fun problem of reverse-Jeopardy!: automatically generating Jeopardy! episodes from knowledge bases [WWW 2015, WebSci 2016, ICTIR 2017]. (more)