SKLCS Seminar
Title: Learning Representation for Fine-Grained Text Analysis
Speaker: Lizhen Qu (Macquarie University, Australia)
people.mpi-inf.mpg.de/~lqu
Time: 31st July 2015, 15:00
Venue: Seminar Room (334), Level 3, Building 5,
Institute of Software, Chinese Academy of Sciences (CAS),
4 Zhongguancun South Fourth Street, Haidian District, Beijing 100190
Abstract:
My talk will consist of two parts. In the first part of the talk I
will present Senti-LSSVM model for sentiment-oriented relation
extraction. This task aims to jointly extract both sentiments (e.g.
Paul likes Nexus 5.) and comparisons (e.g. Paul thinks Nexus 5 is
better than Galaxy S5.) from sentences. The corresponding outputs are
directed hyper-graphs and the lexical features are learned with
recursive neural networks.
In the second part, I will introduce my recent work at NICTA on
applying deep learning techniques to a number of natural language
processing tasks, including identification of multi-word expressions,
named entity recognition, part-of-speech tagging, and chunking. We
find that deep learning techniques perform especially well on
cross-domain tasks. We have achieved 10% improvement over competitive
baselines on named entity recognition for novel types.
