Jo?o Silva

 

报告题目：Learning Profiles in Duplicate Question Detection

时间：2018年1月24日10:00—11:00

地点：天赐庄校区本部理工楼321室

 

摘要：Duplicate question detection (DQD) is a natural language processing task that has recently become the focus of active research, where two interrogative segments are considered to be semantically equivalent, and thus duplicate, if they can receive the same answer. In line with this surge of interest, the SemEval challenge, devoted to semantic similarity, was the first to include a task specifically on question similarity in its edition of 2016. Much of the motivation for this research topic is coming from the usefulness of resorting to DQD to support online question answering community forums, and also conversational interfaces, in general. This presentation will cover recent work whereby different methodologies (count-based, SVM and neural network) for DQD were tested to assess how their performance is affected by variation in dataset size.

 

个人简介：Jo?o Ricardo Silva is a postdoctoral researcher at NLX, the Natural Language and Speech Group of the Department of Informatics of the University of Lisbon, Faculty of Sciences. He completed his PhD at the University of Lisbon in 2014. His interests range over several topics in Natural Language Processing. Before and during his MSc, his work focused particularly on the development of foundational shallow processing tools for Portuguese. Later, already under the scope of his PhD, his work concerned the robust handling of out-of vocabulary words. He has collaborated on multiple national and international projects, such as QTLeap, where machine translation is improved by deep language engineering approaches; and most recently on ASSET, where among other tasks a chatbot answer retrieval interface is improved through the application of semantic similarity techniques for duplicate question detection.