perf: rebuild ai classify prompt which will have more better performance

Author: moonrailgun

src/server/model/prompt/survey.ts

@@ -92,23 +92,62 @@ export function buildSurveyClassifyPrompt(
   }[],
   suggestionCategory: string[],
   language: string = 'en'
-): string {
-  return `
-You are a content data analysis and classification expert. You need to make a simple classification based on the information collected from users in multiple languages ​​around the world, and return the classified json directly to me.
+) {
+  return {
+    prompt: `
+## Role
+
+You are a content analysis expert. You will receive a list of content from various countries in JSON format. Your job is to analyze these items and provide a classification for each record.
+
+## Skills
+
+- **Multilingual Proficiency**: Able to understand content in different languages.
+- **Semantic Understanding**: Capable of interpreting the meaning behind the text.
+- **Content Summarization & Organization**: Skilled in extracting key points and structuring them.
+- **Structured Output Production**: Able to present findings in a clear, organized format.
+
+## Action
+
+1. Read the user's input, which is a JSON array of objects. Each object contains:
+   - \`id\`: a unique identifier
+   - \`content\`: text in one or more languages
+2. Determine the best-fit category for each \`content\` from the list below.
+3. Return a JSON object where each key is the \`id\` and the value is the chosen category.
 
-The data is as follows:
-${data.map((obj) => `- ${JSON.stringify(obj)}`).join('\n')}
+## Example
 
-The classification results of the example are as follows:
-{"id1": "some category which summary", "id2": "another category which summary"}
+### Input Example
+\`\`\`
+- {"id": "id1", "content": "This option dont appear i cant use it"}
+- {"id": "id2", "content": "Toda vez que tento fazer uma persona acaba que o aplicativo saí, impossibilitado de fazer uma persona"}
+\`\`\`
+
+### Output Example
+
+\`\`\`json
+{
+  "id1": "Application malfunction",
+  "id2": "Application crash"
+}
+\`\`\`
+
+## Reference Categories
 
-The existing categories are as follows. Please refer to the existing categories as much as possible:
 ${JSON.stringify(suggestionCategory)}
 
-No explanation is required.
 
-${language !== 'en' ? `And response result should use \`${language}\` as response language.` : ''}
-`.trim();
+## Constraints
+
+1. **Output Format**: All responses must be returned in JSON format.
+2. **Do not create new categories**: If you introduce a new category, ensure it is strictly necessary to capture the main intent of the content. If content seems to fall between two categories, pick the one that is the closest match rather than creating a new category.
+3. **Key-Value Structure**: The output must be a JSON object where each \`id\` serves as the key and its corresponding classification is the value.
+4. **Consistency in Data Count**: The number of entries in your output must match the number of entries in the input.
+5. **Use Existing Categories**: Only create new categories if absolutely necessary.
+6. **Return user perfer language**: response result should use **${language}** as response language.
+
+`.trim(),
+    question: data.map((obj) => `- ${JSON.stringify(obj)}`).join('\n'),
+  };
 }
 
 export function buildSurveyTranslationPrompt(

src/server/mq/worker.ts

@@ -198,7 +198,7 @@ async function runSurveyAIClassifyWorker(msg: string) {
 
     let inc = 0;
     for (const group of groups) {
-      const prompt = buildSurveyClassifyPrompt(
+      const { prompt, question } = buildSurveyClassifyPrompt(
         group,
         currentSuggestionCategory,
         languageStrategy === 'user' ? language : 'en'
@@ -208,14 +208,9 @@ async function runSurveyAIClassifyWorker(msg: string) {
         `Process run survey AI classify, group batch ${++inc} group size: ${group.length}`
       );
 
-      const res = await requestOpenAI(
-        workspaceId,
-        prompt,
-        'Please help me generate data.',
-        {
-          response_format: { type: 'json_object' },
-        }
-      );
+      const res = await requestOpenAI(workspaceId, prompt, question, {
+        response_format: { type: 'json_object' },
+      });
 
       const json = JSON.parse(res);