Semi-Supervised Learning and Domain Adaptation in Natural Language Processing (Synthesis Lectures on Human Language Technologies) por Anders Sogard

Titulo del libro : Semi-Supervised Learning and Domain Adaptation in Natural Language Processing (Synthesis Lectures on Human Language Technologies)
Fecha de lanzamiento : May 30, 2013
Autor : Anders Sogard
Número de páginas : 104
ISBN : 1608459853
Editor : Morgan & Claypool Publishers
Semi-Supervised Learning and Domain Adaptation in Natural Language Processing (Synthesis Lectures on Human Language Technologies) por Anders Sogard

Descargue o lea el libro de Semi-Supervised Learning and Domain Adaptation in Natural Language Processing (Synthesis Lectures on Human Language Technologies) de Anders Sogard en formato PDF y EPUB. Aquí puedes descargar cualquier libro en formato PDF o Epub gratis. Use el botón disponible en esta página para descargar o leer libros en línea.

Anders Sogard con Semi-Supervised Learning and Domain Adaptation in Natural Language Processing (Synthesis Lectures on Human Language Technologies)

Páginas:104Géneros:12:UY:Computerscience12:UYQL:Naturallanguage&machinetranslationSinopsis:Thisbookintroducesbasicsupervisedlearningalgorithmsapplicabletonaturallanguageprocessing(NLP)andshowshowtheperformanceofthesealgorithmscanoftenbeimprovedbyexploitingthemarginaldistributionoflargeamountsofunlabeleddata.Onereasonforthatisdatasparsity,i.e.,thelimitedamountsofdatawehaveavailableinNLP.However,inmostreal-worldNLPapplicationsourlabeleddataisalsoheavilybiased.Thisbookintroducesextensionsofsupervisedlearningalgorithmstocopewithdatasparsityanddifferentkindsofsamplingbias.Thisbookisintendedtobebothreadablebyfirst-yearstudentsandinterestingtotheexpertaudience.MyintentionwastointroducewhatisnecessarytoappreciatethemajorchallengeswefaceincontemporaryNLPrelatedtodatasparsityandsamplingbias,withoutwastingtoomuchtimeondetailsaboutsupervisedlearningalgorithmsorparticularNLPapplications.Iusetextclassification,part-of-speechtagging,anddependencyparsingasrunningexamples,andlimitmyselftoasmallsetofcardinallearningalgorithms.Ihaveworriedlessabouttheoreticalguarantees(thisalgorithmneverdoestoobadly)thanaboutusefulrulesofthumb(inthiscasethisalgorithmmayperformreallywell).InNLP,dataissonoisy,biased,andnon-stationarythatfewtheoreticalguaranteescanbeestablishedandwearetypicallyleftwithourgutfeelingsandacatalogueofcrazyideas.Ihopethisbookwillprovideitsreaderswithboth.ThroughoutthebookweincludesnippetsofPythoncodeandempiricalevaluations,whenrelevant.TableofContents

Los más vendidos Libros