Adam Jatowt, en séjour recherche au L3i jusqu’à 18 juillet, fera un séminaire labo pour présenter ses travaux le jeudi 21 juin de 10h30 à 12h probablement en salle 018.
Résumé de la présentation et petite biographie ci-dessous. Notez la date et venez nombreux.
Across-time Term Similarity Computation and Explanation
Recently, large amounts of historical texts have been digitized and made accessible to the public. However, accessing them is difficult due to the terminology gap problem resulting from semantic evolution of terms. Consequently, users trying to search in long-term document archives may not know correct terms when formulating their queries. In this talk, I will first describe our approach for estimating and explaining the across-time similarity of terms. Given an input term (e.g., iPod) and the target time (e.g. 1980s), the task is, first, to find its closest analog that existed in the target time (e.g., Walkman) and, then, to output evidences of the similarity between the query and the returned answer (e.g., portable music devices). Finally, I will briefly introduce our online system for interactively explaining semantic term change over time using Google Books n-grams datasets.
Adam Jatowt is an Associate Professor at the Social Informatics Department at Kyoto University. He has received his Ph.D. in Information Science and Technology from the University of Tokyo, Japan in 2005 on the topic of temporal document summarization. After completing his PhD, he worked for a year as a postdoctoral researcher at National Institute of Information and Communications Technology (NICT). He has many research interests in an area of information retrieval and knowledge extraction from document collections. Adam is however especially interested in the convergence of IR and text mining with a digital history area. He has been also working on text comprehensibility estimation. More recently, he started research in city street attribute extraction and pedestrian route recommendation. Adam has been serving as a PC co-chair of IPRES2011, SocInfo2013, ICADL2014 and JCDL2017 conferences as well as a tutorial co-chair of SIGIR2017, and co-organizer of three NTCIR evaluation tasks. He regularly serves as a PC member of WSDM, CIKM, SIGIR, JCDL, WISE, TPDL, DASFAA, SocInfo and COLING.
Email : adam dl.kuis.kyoto-u.ac.jp
Webpage : http://www.dl.kuis.kyoto-u.ac.jp/ adam/