Topic segmentation with an aspect hidden Markov model(2)

We present a novel probabilistic method for partially unsupervised topic segmentation on unstructured text. Previous approaches to this problem utilize the hidden Markov model framework (HMM). The HMM treats a document as mutually independent sets of words

TopicSegmentationwithanAspectHiddenMarkovModel

DavidM.Blei

UniversityofCalifornia,Berkeley

Dept.ofComputerScience

Berkeley,CA,94720

PedroJ.Moreno

CambridgeResearchLaboratory

CompaqComputerCorporation

CambridgeMA02142-1612

July2001

Abstract

Wepresentanovelprobabilisticmethodforpartiallyunsupervisedtopicsegmen-tationonunstructuredtext.PreviousapproachestothisproblemutilizethehiddenMarkovmodelframework(HMM).TheHMMtreatsadocumentasmutuallyindepen-dentsetsofwordsgeneratedbyalatenttopicvariableinatimeseries.WeextendthisideabyembeddingtheaspectmodelfortextintothesegmentingHMM.Indoingso,weprovideanintuitivetopicaldependencybetweenwordsandacohesivesegmentationmodel.WeapplythismethodtosegmentunbrokenstreamsofNewYorkTimesarti-clesaswellasnoisytranscriptsofradioprogramsonSPEECHBOT1,anonlineaudioarchiveindexedbyanautomaticspeechrecognitionengine.WeprovideexperimentalcomparisonsbetweenourtechniqueandtheHMMapproach.OurresultssuggestthatthistechniquecanperformaswellastheHMMmethodandinsomecasesevenbetter.