Evaluation of syntactic phrase indexing - CLARIT NLP track report

Evaluation of syntactic phrase indexing - CLARIT NLP track report

ID:37657795

大小:173.39 KB

页数:11页

时间:2019-05-27

Evaluation of syntactic phrase indexing - CLARIT NLP track report_第1页
Evaluation of syntactic phrase indexing - CLARIT NLP track report_第2页
Evaluation of syntactic phrase indexing - CLARIT NLP track report_第3页
Evaluation of syntactic phrase indexing - CLARIT NLP track report_第4页
Evaluation of syntactic phrase indexing - CLARIT NLP track report_第5页
资源描述:

《Evaluation of syntactic phrase indexing - CLARIT NLP track report》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库

1、EvaluationofSyntacticPhraseIndexing

2、CLARITNLPTrackReport1122ChengxiangZhai,XiangTong,NatasaMilic-Frayling,DavidA.Evans1LaboratoryforComputationalLinguisticsCarnegieMellonUniversityPittsburgh,PA15213fcz25@andrew.cmu.edu,xt22@andrew.cmu.edug2CLARITECHCorpora

3、tion5301FifthAve.Pittsburgh,PA15232-2124fnatasa@clarit.com,dae@clarit.comg1IntroductionTM1TheCLARITNLPtracke ortisfocusedonevaluatingtheusefulnessofsyntacticphrasesfordocumentindexing.TheCLARITsystemhasseveralNLPtechniquesintegratedwiththevectorspaceretrieva

4、lmodel[Evansetal.91,Evansetal.95].TheNLPtechniquesusedinCLARITincludemorphologicalanalysis,robustnoun-phraseparsing,andautomaticconstructionof rstorderthesauri,amongothers.OnemainfeatureofCLARITindexingisthatitemphasizesphrase-basedindexingwithdi erentoption

5、sfordecomposingnounphrasesintosmallerconstituents,includingsinglewords.InpastTRECs,thedefaultmodeforindexinginvolvedfullnounphrases,singlewords,2andoccasionallyselectedsub-phrases.Whilesomeearlyexperimentshaveshownthee ectivenessofnounphrasesforindexing[Evan

6、setal.91],thereisnodirectevaluationoftheire ectivenessinthecontextofTREC.Inparticular,thecontributionofsmallsub-phrasestoretrievalperformancehasnotbeenevaluatedoutsidethecontextofoverallsystemperformance.TheversionoftheCLARITsystemthatweusedintheexperimentsh

7、asitsNLPcomponenttightlyintegratedwiththerestofthesystem.ThisdoesnotalloweasyevaluationoftheindividualNLPcomponents.We,therefore,developedseparateNLPmodulesandusedCLARITasaretrievalengineonlytoevaluatethee ectivenessofphrase-basedindexing.2Phrasesforindexing

8、MostcurrentIRsystemsuseword-basedindexingoftensupplementedwithphrasesthatareobtainedusingsimplestatisticalapproaches,suchasafrequencycountingofadjacentwordpairs.Intuitively,itseemsthatsyntacticphrases(i.e.,phraseswithcertainsyntacticrelations)canrepresentdoc

9、umentcontentmorepreciselythansinglewordsandphrasesdiscoveredusingsimplestatisticalmethods,andarethuspotentiallyusefulforindexing[Evansetal.91,Evansetal.95,Strzalkowskietal.95,Evans&Zhai96,Hullet

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。