The　spread　of　fake　news　on　social　media　seriously　jeopardizes　the　authenticity　of　online　content.　Such　misinformation　has　the　potential　to　affect　public　opinion,　government　credibility,　and　more.　Therefore,　automatic　detection　of　fake　news　is　an　important　and　challenging　problem　that　has　not　been　well　solved　today.　Fake　news　has　three　universally　recognized　characteristics:　the　text　of　an　article,　the　user　response　it　receives,　and　the　source　users　promoting　it.　Existing　work　mainly　focuses　on　formulating　a　solution　for　a　certain　feature,　which　cannot　detect　fake　news　more　accurately.　In　this　work,　we　propose　a　model　that　combines　all　three　features　to　achieve　more　accurate　and　automated　predictions.　We　propose　a　model　called　TSRI,　which　consists　of　four　modules:　Text,　Source,　Response,　and　Integrate.　The　first　module　is　based　on　news　text　content,　using　the　BERT+LSTM　model,　while　considering　the　semantic　representation　of　sentences　and　the　long-distance　dependencies　between　sentences.　The　second　module　uses　GAT　to　model　the　network　of　replying　and　retweeting　users　based　on　social　network　features.　The　third　module　is　based　on　the　user＇s　response　to　the　news,　using　LDA　to　find　conflicting　opinions　in　the　news　tweets,　and　based　on　the　identified　tweets＇　opinions,　constructs　the　credibility　of　the　tweets　and　spreads　the　network,　and　finally　obtains　the　module＇s　value　through　iterative　deduction.　result.　The　last　module　combines　the　above　three　and　outputs　whether　the　news　is　true　or　false.　©　2023　IEEE.