TeamoninthisvideowillcontinuelookingathowtodotextclassificationcontinuingfromwhatwelearnedinpartonewillbeclassifyingtheIMDbdatasetstobuild a modelthatcanunfairif a moviereviewispositiveornegative.
Uptothispoint, we'vedonethepreprocessingofthedata, gettingitinto a raiseofnumericvaluesthatcanthenbeusedtotrain a neuralnetwork.
Buttodesign a neuralnetworkthatcanlearnfromthis, wehavetousesomethingcalledanembeddinginanembedding.
A wordisconvertedinto a vectorinmultidimensionalspace, withthetheorythatwordsofsimilarsentimentwillhave a sortofsimilardirectioninthatspace.
Nowyoumightthink, Wait a second.
Howdoes a wordgetconvertedinto a vector?
Whatwouldthatlooklike?
Well, let's lookat a verysimplifiedexample.
Sosayyou're a fanofRegencyAargh!
RomanceslikethoseofJaneAusten.
Yeah, I know, I know.
Takecharactersfromprideandprejudiceonplotthemon A toteachartwhereoneaccessisthegenderderivedfromtheirtitle, andtheotherisanestimationoftheirpositioninsocietybasedontheirtitle.
Sowe'lllookatoneofmyfavorites, MrCollins.
Nowhe's obviously a male, andfromhistitle, MrHe's probablynotnobility, sowe'llplothiminblue.
Therandomone, whichwasmadeupof a jumbleofrandomwords, scored 10.34 whichyouprobablyexpect, buttheonewiththereviewismadeupentirelyofthewordBrilliantwill, ofcourse, be a positivereviewandyoucanseeitscores a perfectone, andthat's it.
Inthesevideos, yousawhowtobuild a taxsentimentclassifications.