With　the　accumulation　of　various　kinds　of　text　data,　it　is　no　longer　possible　to　generalize　or　classify　them　by　manual　reading,　so　how　to　use　statistical　models　to　mine　text　data　reasonably　and　effectively　has　become　an　important　issue　in　academic　research　and　practical　work.　This　paper　discusses　three　problems　of　Chinese　text　mining:　word　separation,　keyword　extraction　and　text　classification.　For　the　word　separation　problem,　the　Cascaded　Hidden　Markov　Model　and　the　WDM　that　treats　the　segmentation　between　words　as　missing　data　and　solves　it　with　the　EM　algorithm　are　introduced.　For　the　keyword　extraction　problem,　this　paper　proposes　a　Bayes　factor　and　introduces　CCS　using　sparse　regression.　For　the　text　classification　problem,　the　method　of　building　a　classifier　based　on　the　frequency　of　keywords　and　the　method　of　building　a　classifier　based　on　the　probability　of　the　topic　first　are　introduced.　We　give　the　respective　advantages　of　each　method　by　comparing　the　above　methods　with　two　datasets　using　SVM　and　Random　forest,　and　make　suggestions　of　their　use.　©　2023　SPIE.