优化 字幕优化流程

VideoAnalysisCore/AICore/GPT/GTP_Analysis_1.cs

@@ -31,7 +31,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
     public class GTP_Analysis_1 : IBserGPTWorkflow
     {
         private readonly GeminiGPTClient geminiClient;
-        private readonly DeepSeekGPTClient chatClient;
+        private readonly DeepSeekGPTClient deepSeekClient;
         private readonly BestAIClient chatGPTClient;
         private readonly Repository<CourseGradingCriteria> criteriaDB;
         private readonly RedisManager redisManager;
@@ -51,7 +51,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
             Repository<KnowledgeInfo> knowledgeInfoDB, Repository<VideoKonwPoint> videoKonwPointDB, SimpLetexClient simpLetexClient,
             Repository<VideoQuestion> videoQuestionDB, OssClient ossClient, Repository<VideoQuestionKonw> videoQuestionKonwDB, RedisManager redisManager, BestAIClient chatGPTClient, GeminiGPTClient geminiClient)
         {
-            chatClient = moonshotClient;
+            deepSeekClient = moonshotClient;
             criteriaDB = criteria;
             this.videoTaskDB = videoTaskDB;
             this.knowledgeInfoDB = knowledgeInfoDB;
@@ -90,7 +90,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
               $"输出内容只返回json格式({checkResFormat1})" +
               $" 格式 (方法点Id|方法点名称) " +
               $"提供的知识点名称({knows})。";
-            await redisManager.AddTaskLog(taskInfo.Id, DateTime.Now + "=>2.开始分析视频内容知识点");
+            await redisManager.AddTaskLog(taskInfo.Id, "=>2.开始分析视频内容知识点");
             VideoKnowRes[] konwRes;
             var knowOK = false; 
             for (int i = 0; i < 4; i++)
@@ -190,7 +190,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
             var pptFormat = taskInfo.VideoType == AttachmentsInfoType.复习
                 ? "这堂课是习题课,所讲解内容几乎都是试题。"
                 : string.Empty;
-            var checkResFormat = """{"Score":打分(number),"Evaluation":评价(string)""";//,"Data":优化后的分段(array)}""";
+            var checkResFormat = """{"Score":打分(number),"Evaluation":评价以及扣分原因(string)""";//,"Data":优化后的分段(array)}""";
             var checkMessage =
                 $"""
                 请你担任一位专业的视频内容分析教研老师，擅长评估视频内容的结构和逻辑流暢度。
@@ -226,11 +226,11 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                 return JsonSerializer.Deserialize<SenseVoiceRes[]>(taskInfo.CaptionsAI);
             var subject = taskInfo.Subject.ToString();
             var newCaptionsList = new List<SenseVoiceRes>(captionsArr.Length);
-            var spanCount = 50;
+            var spanCount = 75;
             var totalCount = captionsArr.Length / spanCount + 1;
 
             await Parallel.ForAsync(0, totalCount,
-                new ParallelOptions() { MaxDegreeOfParallelism = 20 },
+                new ParallelOptions() { MaxDegreeOfParallelism =15 },
                 async (s, c) =>
             {
                 var cArr = captionsArr
@@ -242,17 +242,20 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                 var nowCaptionStr = cStrArr.ToJson();
                 var resFormat = """[string(修改结果)]""";
                 var postMessages =
-                   $"这是一堂中国{subject}课堂的字幕,由结果是语音识别提供。" +
-                   $"字幕内容与{subject}学科{sections}章节相关。" +
-                   $"但是识别出来的字幕包含错误的关键字 例如应当是'数列'被识别为'树列'。" +
-                   $"所以需要你帮我修复其中的错别字,修复公式排版。" +
-                   $"请注意,只允许对字幕进行修改。" +
-                   $"输出内容只返回json格式({resFormat})" +
-                   $"字幕内容(JSON字符串)" +
-                   $"`{nowCaptionStr}`" +
-                   $"字幕结束。" +
-                   $"最后请确保输出字幕条数与输入字幕条数一致!!! 如果不一致则重新优化并且确保字幕条数一致!!!!";
-                var resData = await chatClient.ChatAsync<string[]>(taskInfo.Id.ToString(), postMessages, "优化字幕", "deepseek-chat", 3000);
+                      $"角色设定：你是一位专业的中国{subject}学科专家，负责校对关于{sections}内容的课堂教学字幕。\n\n" +
+                      $"任务描述：\n" +
+                      $"请根据上下文逻辑，对输入的语音识别（STT）字幕进行深度优化。具体要求如下：\n" +
+                      $"1. 逻辑纠错：结合{subject}学科背景，利用前后文语义修正所有错误词汇。不仅要修正同音错别词（如：树列改为数列），还要修正因识别模糊导致的语义断裂或学科术语错误。\n" +
+                      $"2. 断句与标点：优化字幕的标点符号，并根据老师说话的语感和学科逻辑重新调整断句位置。确保每一条字幕在学术表达上自然、通顺，修复由于语音停顿造成的断句不当。\n" +
+                      $"3. 公式规范：将字幕中提到的数学或科学公式统一转化为规范的 LaTeX 格式。\n\n" +
+                      $"强制约束：\n" +
+                      $"- 数量对齐：输出的字幕条数（Array Length）必须与输入的字幕条数完全一致，严禁合并、拆分或删除任何条目。\n" +
+                      $"- 纯净返回：只允许返回 JSON 格式的字符串，严禁包含任何前言、后缀或解释性文字。\n" +
+                      $"- 数据格式：JSON 结构必须严格符合：{resFormat}\n\n" +
+                      $"待优化字幕内容：\n" +
+                      $"{nowCaptionStr}\n\n" +
+                      $"最终核对：请确保输出 JSON 中包含的字幕条数与输入的字幕条数完全对应。";
+                var resData = await deepSeekClient.ChatAsync<string[]>(taskInfo.Id.ToString(), postMessages, "优化字幕");
                 if (resData.Count() != cArr.Count())
                 {
                     resData = cStrArr.ToArray();
@@ -335,7 +338,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
         /// <returns></returns>
         private async Task<SenseVoiceRes[]> AnalysisVideoQuestions(VideoTask taskInfo, List<KnowledgeInfo> knowledgeInfos)
         {
-            await redisManager.AddTaskLog(taskInfo.Id, DateTime.Now + $"=>{taskInfo.Id} 提取试题");
+            await redisManager.AddTaskLog(taskInfo.Id, $"=>{taskInfo.Id} 提取试题");
             if (taskInfo is null || string.IsNullOrEmpty(taskInfo.PPTKeyFrame))
                 return null;
             var farmeArr = JsonSerializer.Deserialize<int[]>(taskInfo.PPTKeyFrame);
@@ -376,7 +379,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                             break;
 
 #if DEBUG
-                        await redisManager.AddTaskLog(taskInfo.Id, DateTime.Now + $"=>{taskInfo.Id} 提取{knowInfoArr.StartTime}秒试题的试题内容");
+                        await redisManager.AddTaskLog(taskInfo.Id, $"=>{taskInfo.Id} 提取{knowInfoArr.StartTime}秒试题的试题内容");
 #endif
 
                         //var knowArr=JsonSerializer.Serialize(knowInfoArr.Select(s => new { s.KnowPointId, s.KnowPoint }));
@@ -394,7 +397,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                            $"输出内容只返回json格式为({resFormat})" +
                            $"以下是试题内容" +
                            $"`{sRes.Result.res.value}`";
-                        var resData = await chatClient.ChatAsync<VideoQuestionOSSDto[]>(taskInfo.Id.ToString(), postMessages, "提取试题");
+                        var resData = await deepSeekClient.ChatAsync<VideoQuestionOSSDto[]>(taskInfo.Id.ToString(), postMessages, "提取试题");
                         //var resData = await chatClient.ChatAsync<VideoQuestionOSSDto[]>(taskInfo.Id.ToString(), postMessages, "提取试题");
                         if (resData is null || resData.Count() == 0)
                             break;
@@ -435,7 +438,7 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                     }
                     catch (Exception ex)
                     {
-                        await redisManager.AddTaskLog(taskInfo.Id, DateTime.Now + $"=>{taskInfo.Id} 提取{knowInfoArr.StartTime}秒试题出现错误 {ex.Message}");
+                        await redisManager.AddTaskLog(taskInfo.Id, $"=>{taskInfo.Id} 提取{knowInfoArr.StartTime}秒试题出现错误 {ex.Message}");
                     }
                 }
             }
@@ -520,6 +523,9 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                 var insertData = await GetVideoKnow(questionRes, taskInfo, sections, knowledgeInfos);//ChatGPT
                 //校验结果质量
                 var checkRes = await VerifySpanQuality(questionRes, taskInfo, captions, sections, Course_Id);
+                
+                await redisManager.AddTaskLog(taskInfo.Id,  $"=>课堂内容AI分析结果 得分=>{checkRes.Score}  ");
+                await redisManager.AddTaskLog(taskInfo.Id,   checkRes.Evaluation);
 
                 if (checkRes != null && checkRes.Score >= 80)
                 {
@@ -527,10 +533,11 @@ namespace VideoAnalysisCore.AICore.GPT.DeepSeek
                     await videoKonwPointDB.DeleteAsync(s => s.VideoTaskId == taskInfo.Id);
                     await videoKonwPointDB.InsertRangeAsync(insertData);
                     break;
-                }
+                }else
+                    await redisManager.AddTaskLog(taskInfo.Id, $"=>课堂内容AI分析结果不合格!即将重试 剩余次数{tryCount}");
                 if (questionRes.Any(s => s.KeepTime < 30))
                 {
-                    await redisManager.AddTaskLog(taskInfo.Id, DateTime.Now + "=>视频分段过短!! 重新进行AI分析");
+                    await redisManager.AddTaskLog(taskInfo.Id, "=>视频分段过短!! 重新进行AI分析");
                     continue;
                 }
             }