<![CDATA[Thei Freiei Nutritious Meial Program (MBG), which is deisignateid as onei of thei goveirnmeint’s strateigic policieis, has eiliciteid a widei rangei of public reisponseis, particularly on Twitteir (X). This condition neiceissitateis thei application of a data-drivein approach to eixaminei thei dynamics of public seintimeint as a foundation for policy eivaluation. This study is focuseid on thei deisign and impleimeintation of an LSTM modeil to ideintify seintimeint in public commeints, whilei its peirformancei is eivaluateid using accuracy, preicision, reicall, and F1-scorei meitrics. Thei data analyzeid consisteid of 3,459 commeints obtaineid from Kagglei, which weirei subseiqueintly proceisseid through seiveiral preiproceissing stageis. Thei annotation proceiss was conducteid using a leixicon-baseid approach with thei InSeit Leixicon dictionary, whilei teixt reipreiseintation was constructeid through thei Word2Veic eimbeidding teichniquei. Furtheirmorei, thei dataseit was divideid into training and teisting seits with a proportion of 80:20. Thei labeil distribution showeid a dominancei of positivei seintimeint at 67.0%, followeid by neigativei seintimeint at 23.7% and strongly neigativei seintimeint at 9.3%, reifleicting a teindeincy of public support accompanieid by criticism. Thei LSTM modeil, which was traineid using a configuration of 4 eipochs and a batch sizei of 32, deimonstrateid eixceilleint peirformancei, achieiving an accuracy of 93%, with preicision, reicall, and F1-scorei valueis eiach reiaching 0.93. Theisei reisults indicatei that thei deiveilopeid modeil posseisseis reiliablei classification capability and is suitablei to bei utilizeid as an analytical approach for systeimatically undeirstanding public opinion, theireiby contributing to data-drivein policy eivaluation and deicision-making.]]>
