In　this　paper,　we　study　a　stock　prediction　method　based　on　false　information　identification　and　machine　learning.Firstly,　a　web　crawler　and　the　Tushare　data　interface　tool　are　used.　The　textual　data　and　stock　price　data　required　for　the　study　are　obtained.　Next,　the　data　is　pre-processed　such　as　tokenizing,　stemming,　etc.　Next,　the　paper　uses　bag-of-words,　POS　tagging　and　word2vec　methods　for　feature　selection.　The　performance　of　various　classifiers　was　compared,　and　the　better　performing　logistic　regression　classifier　was　chosen　to　calculate　the　truth　probability　score　and　determine　the　true/false.　Finally,　two　stock　price　prediction　models,　LSTM　and　GRU,　are　written　and　their　prediction　results　are　compared　in　real　stock　data.　The　prediction　results　of　GUR　were　found　to　be　more　accurate　and　to　fit　better　over　time.　©　2022　IEEE.