Transcript
Page 1: Double Embeddings and CNN-based Sequence Labelinghxu/ACL_poster.pdf · Double Embeddings and CNN-based Sequence Labeling for Aspect Extraction Hu Xu1, Bing Liu1, Lei Shu1, Philip

Theimplementationofthepaperisavailableathttps://www.cs.uic.edu/~hxu/

ArchitectureofDE-CNN:redvectorsarepaddingvectors

Double Embeddings and CNN-based Sequence Labeling for Aspect Extraction HuXu1,BingLiu1,LeiShu1,PhilipS.Yu1,21DepartmentofComputerScience,UniversityofIllinoisatChicago2InstituteforDataScience,TsinghuaUniversity

{hxu48, liub, lshu3, psyu}@uic.edu

Ther-thCNNfilterforthei-thwordinlayerl

AspectExtractionisanimportanttaskinfine-grainedsentimentanalysis.Weproposeasimpleandfastapproachwithoutusinganysophisticatedfeaturesandmodels.Thecontributionsarein2folds:Doubleembedding:weusetwotypesofpre-trainedembeddingsforaspectextraction:generalpurposeembeddingsanddomainspecificembeddings.CNN:weuseCNNforsequencelabeling,whichisparallelandfasterthanserialLSTM.WeadaptCNN(e.g.,dropmax-poolinglayer)togetbetterresults.

Top Related