TF-IDF　is　widely　used　as　the　most　common　feature　weight　calculation　method.　The　traditional　TF-IDF　feature　extraction　method　lacks　the　representation　of　the　distribution　difference　between　classes　in　the　text　classification　task　and　the　feature　matrix　generated　by　the　TF-IDF　is　huge　and　sparse.　Based　on　this　situation,　this　paper　proposes　a　method　of　using　the　feature　extraction　algorithm　of　chi-square　statistics　to　compensate　for　the　distribution　difference　between　classes　and　generating　a　fixed-dimensional　real　matrix　through　word2vec.　The　experimental　results　show　that　the　new　method　is　significantly　better　than　the　traditional　feature　extraction　methods　in　the　evaluation　results　such　as　precision,　recall,　F1　and　ROC_AUC.　©　2020,　Springer　Nature　Switzerland　AG.