This　paper　is　based　on　the　famous　log-periodic　power　law　model　(LPPL)　in　financial　physics　to　warn　of　the　collapse　of　China＇s　Shanghai　Composite　Index　and　GEM　Index　in　June　2015.　In　view　of　the　existing　research　using　the　LPPL　model　to　warn　of　market　crash,　only　the　historical　trading　data　of　the　market　are　considered.　For　the　first　time,　investor　sentiment　factors　are　incorporated　into　the　modeling　process　of　LPPL　model　to　improve　the　early　warning　effect　of　LPPL　model.　Using　the　text　mining　technology　combined　with　semantic　analysis　methods　to　grasp　the　financial　media＇s　stock　evaluation　report　for　word　frequency　statistics,　in　order　to　build　the　medium　sentiment　index.　The　further　modified　expression　of　the　crash　probability　function　in　the　LPPL　model　is　represented　as　a　function　of　historical　trading　data　and　medium　sentiment,　and　thus　constructing　an　LPPL-MS　combination　model　to　warn　of　stock　market　crash.　The　empirical　results　show　that　the　LPPL-MS　combination　model　constructed　in　this　paper　has　higher　warning　accuracy　than　the　LPL　model,　and　its　prediction　crash　time　is　closer　to　the　actual　crash　time　of　the　Shanghai　Index　and　GEM　Index,　and　its　fitting　results　have　passed　the　relevant　test.