In　this　paper,　we　will　continue　to　use　the　new　method　in　the　2019　version　to　continue　the　work　of　the　2020　TREC　Incident　Streams　System　task[1].　Social　media　has　become　an　indispensable　part　of　human　life,　such　as　Twitter,　Weibo　and　so　on.　When　natural　disasters　occur,　such　as　fires,　earthquakes,　flash　floods,　tsunamis,　mudslides　and　other　natural　disasters　or　shootings,　robberies　and　other　emergencies,　if　only　through　media　reports,　the　time　of　the　event　will　be　very　slow,　leading　to　some　preventable　loss.　People　like　to　post　disaster　situations　or　events　on　social　media.　The　purpose　of　the　task　is　to　filter　such　natural　disasters　or　emergencies　by　classifying　the　text　on　twitter.　Similarly,　each　tweet　is　prioritized　and　the　tagged　information　is　reported　to　the　relevant　personnel　according　to　different　priorities.　Let　the　staff　know　about　the　progress　of　the　incident　to　help.　This　article　will　introduce　the　framework　and　methods　of　the　classification　system,　as　well　as　the　experimental　results.　©　2020　29th　Text　REtrieval　Conference,　TREC　2020　-　Proceedings.　All　Rights　Reserved.