Add monitoring service and database schema

- Implement monitoring service with heartbeat, lease tasks, resume tasks, and task result handling. - Create monitoring time utilities for business date calculations. - Add unit tests for date window resolution and business day handling. - Define database schema for monitoring-related tables including quotas, daily reports, and task management. - Establish migration scripts for creating and dropping monitoring tables.
2026-04-12 09:56:18 +08:00
parent 9b4dd09780
commit 6066f43a7d
67 changed files with 16312 additions and 125 deletions
@@ -0,0 +1,173 @@
+package app
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"time"
+
+	sharedllm "github.com/geo-platform/tenant-api/internal/shared/llm"
+)
+
+const (
+	monitoringAnswerParseLLMModel           = "doubao-seed-2-0-lite-260215"
+	monitoringAnswerParseLLMTimeout         = 30 * time.Second
+	monitoringAnswerParseLLMMaxOutputTokens = 600
+)
+
+var monitoringAnswerParseLLMSchema = []byte(`{
+  "type": "object",
+  "additionalProperties": false,
+  "required": [
+    "brand_mentioned",
+    "brand_mention_position",
+    "first_recommended",
+    "sentiment_label",
+    "matched_brand_terms"
+  ],
+  "properties": {
+    "brand_mentioned": {
+      "type": "boolean"
+    },
+    "brand_mention_position": {
+      "type": "string",
+      "enum": ["top1", "mentioned", "not_mentioned"]
+    },
+    "first_recommended": {
+      "type": "boolean"
+    },
+    "sentiment_label": {
+      "type": "string",
+      "enum": ["positive", "neutral", "negative", "unknown"]
+    },
+    "matched_brand_terms": {
+      "type": "array",
+      "items": {
+        "type": "string"
+      }
+    }
+  }
+}`)
+
+type monitoringAnswerLLMParsePayload struct {
+	BrandMentioned       bool     `json:"brand_mentioned"`
+	BrandMentionPosition string   `json:"brand_mention_position"`
+	FirstRecommended     bool     `json:"first_recommended"`
+	SentimentLabel       string   `json:"sentiment_label"`
+	MatchedBrandTerms    []string `json:"matched_brand_terms"`
+}
+
+func parseMonitoringAnswerWithLLM(
+	ctx context.Context,
+	client sharedllm.Client,
+	answer string,
+	brandName string,
+) (monitoringAnswerParseSummary, error) {
+	answer = strings.TrimSpace(answer)
+	brandName = strings.TrimSpace(brandName)
+	if answer == "" {
+		return monitoringAnswerParseSummary{}, fmt.Errorf("answer is empty")
+	}
+	if brandName == "" {
+		return monitoringAnswerParseSummary{}, fmt.Errorf("brand name is empty")
+	}
+
+	result, err := client.Generate(ctx, sharedllm.GenerateRequest{
+		Model:           monitoringAnswerParseLLMModel,
+		Prompt:          buildMonitoringAnswerParsePrompt(answer, brandName),
+		Timeout:         monitoringAnswerParseLLMTimeout,
+		MaxOutputTokens: monitoringAnswerParseLLMMaxOutputTokens,
+		ResponseFormat: &sharedllm.ResponseFormat{
+			Type:        sharedllm.ResponseFormatTypeJSONSchema,
+			Name:        "monitoring_answer_parse",
+			Description: "Parse brand mention signals from an AI answer for brand monitoring.",
+			SchemaJSON:  monitoringAnswerParseLLMSchema,
+			Strict:      true,
+		},
+	}, nil)
+	if err != nil {
+		return monitoringAnswerParseSummary{}, err
+	}
+
+	return decodeMonitoringAnswerLLMParseResult(result.Content)
+}
+
+func buildMonitoringAnswerParsePrompt(answer string, brandName string) string {
+	var builder strings.Builder
+	builder.WriteString("你是品牌监测解析助手。\n")
+	builder.WriteString("请只基于给定回答内容，判断目标品牌在回答中的提及、排序、推荐与情感，不要猜测回答外的信息。\n")
+	builder.WriteString("\n判断规则：\n")
+	builder.WriteString("1. brand_mentioned：回答中是否明确提到目标品牌，或可明确判断是在指代该品牌。\n")
+	builder.WriteString("2. brand_mention_position：\n")
+	builder.WriteString("   - top1：目标品牌被排在第一位，或被明确表述为首选/最推荐。\n")
+	builder.WriteString("   - mentioned：目标品牌被提到，但不是第一位。\n")
+	builder.WriteString("   - not_mentioned：未提到目标品牌。\n")
+	builder.WriteString("3. first_recommended：是否明确把目标品牌作为首选推荐。\n")
+	builder.WriteString("4. sentiment_label：结合品牌相关表述判断 positive / neutral / negative / unknown。\n")
+	builder.WriteString("5. matched_brand_terms：把回答里实际出现、并用于指代该品牌的词语原样列出；没有就返回空数组。\n")
+	builder.WriteString("\n只返回 JSON。\n")
+	builder.WriteString("\n目标品牌：")
+	builder.WriteString(brandName)
+	builder.WriteString("\n回答内容：\n")
+	builder.WriteString(answer)
+	return builder.String()
+}
+
+func decodeMonitoringAnswerLLMParseResult(raw string) (monitoringAnswerParseSummary, error) {
+	var lastErr error
+	for _, candidate := range extractJSONCandidates(raw) {
+		var payload monitoringAnswerLLMParsePayload
+		if err := json.Unmarshal([]byte(candidate), &payload); err != nil {
+			lastErr = err
+			continue
+		}
+		return monitoringAnswerParseSummary{
+			BrandMentioned:       payload.BrandMentioned,
+			BrandMentionPosition: normalizeMonitoringBrandMentionPosition(payload.BrandMentionPosition, payload.BrandMentioned),
+			FirstRecommended:     payload.FirstRecommended,
+			SentimentLabel:       normalizeMonitoringSentiment(payload.SentimentLabel, payload.BrandMentioned),
+			MatchedBrandTerms:    normalizeMonitoringBrandTermList(payload.MatchedBrandTerms),
+		}, nil
+	}
+	if lastErr == nil {
+		lastErr = fmt.Errorf("empty content")
+	}
+	return monitoringAnswerParseSummary{}, fmt.Errorf("decode monitoring answer parse result: %w", lastErr)
+}
+
+func normalizeMonitoringBrandMentionPosition(value string, brandMentioned bool) string {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "top1":
+		if !brandMentioned {
+			return "not_mentioned"
+		}
+		return "top1"
+	case "mentioned":
+		if !brandMentioned {
+			return "not_mentioned"
+		}
+		return "mentioned"
+	case "not_mentioned":
+		return "not_mentioned"
+	default:
+		if brandMentioned {
+			return "mentioned"
+		}
+		return "not_mentioned"
+	}
+}
+
+func normalizeMonitoringSentiment(value string, brandMentioned bool) string {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "positive", "neutral", "negative":
+		return strings.ToLower(strings.TrimSpace(value))
+	case "unknown":
+		return "unknown"
+	default:
+		if brandMentioned {
+			return "neutral"
+		}
+		return "unknown"
+	}
+}