Emotional　conflict　is　a　psychological　concept,　often　used　to　describe　the　complex　emotional　state　of　an　individual.　This　paper　introduces　this　concept　to　the　natural　language　level,　and　investigates　methods　for　recognizing　emotional　conflicts　in　texts　based　on　Large　Language　Model.　To　this　end,　we　focus　on　how　to　use　Large　Language　Models　to　generate　samples　that　match　emotional　conflict.　Specifically,　the　basic　dataset　is　generated　by　GPT4-trubo(mainly　through　prompts　bootstrapping).　We　explore　a　number　of　prompt　design　methods,　e.g.,　sample　prompts,　streaming　sample　prompts　and　keyword　scenario　prompts.　In　addition,　we　will　show　the　Q&A　templates　and　some　of　the　model　responses　for　each　prompt　method,　as　a　guide　to　the　choices　in　data　generating　step.　At　the　end,　the　machine　synthesized　dataset　is　manually　verified　to　ensure　the　data　quality.　The　experiments　use　the　open-source　model　ChatGLM3-6B　on　our　manually-tuned　dataset　for　three　sets　of　performance　evaluations.　The　three　sets　use　Zero-Shot　Q&A,　Prompts　Optimization,　and　LoRA　fine-tuning　as　evaluating　method.　And　we　use　F1　Score,　Accuracy,　Precision,　and　Recall　performance　as　the　evaluating　metrics.　©　2024　Copyright　held　by　the　owner/author(s).