Novault-backend/internal/service/classification_service.go

package service

import (
	"errors"
	"fmt"
	"strings"

	"accounting-app/internal/models"
	"accounting-app/internal/repository"
)

// Classification service errors
var (
	ErrClassificationRuleNotFound = errors.New("classification rule not found")
	ErrInvalidKeyword             = errors.New("keyword cannot be empty")
	ErrInvalidCategoryID          = errors.New("invalid category ID")
	ErrInvalidAmountRange         = errors.New("min amount cannot be greater than max amount")
	ErrRuleAlreadyExists          = errors.New("a rule with this keyword and category already exists")
)

// ClassificationRuleInput represents the input data for creating or updating a classification rule
type ClassificationRuleInput struct {
	UserID     uint     `json:"user_id"`
	Keyword    string   `json:"keyword" binding:"required"`
	CategoryID uint     `json:"category_id" binding:"required"`
	MinAmount  *float64 `json:"min_amount,omitempty"`
	MaxAmount  *float64 `json:"max_amount,omitempty"`
}

// ClassificationSuggestion represents a suggested category with confidence score
type ClassificationSuggestion struct {
	CategoryID  uint                       `json:"category_id"`
	Category    *models.Category           `json:"category,omitempty"`
	Confidence  float64                    `json:"confidence"` // 0.0 to 1.0
	MatchedRule *models.ClassificationRule `json:"matched_rule,omitempty"`
	MatchReason string                     `json:"match_reason"`
}

// ClassificationService handles business logic for smart classification
type ClassificationService struct {
	classificationRepo *repository.ClassificationRepository
	categoryRepo       *repository.CategoryRepository
}

// NewClassificationService creates a new ClassificationService instance
func NewClassificationService(
	classificationRepo *repository.ClassificationRepository,
	categoryRepo *repository.CategoryRepository,
) *ClassificationService {
	return &ClassificationService{
		classificationRepo: classificationRepo,
		categoryRepo:       categoryRepo,
	}
}

// CreateRule creates a new classification rule with business logic validation
func (s *ClassificationService) CreateRule(input ClassificationRuleInput) (*models.ClassificationRule, error) {
	// Validate keyword
	keyword := strings.TrimSpace(input.Keyword)
	if keyword == "" {
		return nil, ErrInvalidKeyword
	}

	// Validate category exists
	exists, err := s.categoryRepo.ExistsByID(input.UserID, input.CategoryID)
	if err != nil {
		return nil, fmt.Errorf("failed to validate category: %w", err)
	}
	if !exists {
		return nil, ErrInvalidCategoryID
	}

	// Validate amount range
	if input.MinAmount != nil && input.MaxAmount != nil && *input.MinAmount > *input.MaxAmount {
		return nil, ErrInvalidAmountRange
	}

	// Check if rule already exists
	exists, err = s.classificationRepo.ExistsByKeywordAndCategory(input.UserID, keyword, input.CategoryID)
	if err != nil {
		return nil, fmt.Errorf("failed to check rule existence: %w", err)
	}
	if exists {
		return nil, ErrRuleAlreadyExists
	}

	// Create the rule
	rule := &models.ClassificationRule{
		Keyword:    keyword,
		CategoryID: input.CategoryID,
		MinAmount:  input.MinAmount,
		MaxAmount:  input.MaxAmount,
		HitCount:   0,
	}

	if err := s.classificationRepo.Create(rule); err != nil {
		return nil, fmt.Errorf("failed to create classification rule: %w", err)
	}

	// Load the category relationship
	rule, err = s.classificationRepo.GetByID(input.UserID, rule.ID)
	if err != nil {
		return nil, fmt.Errorf("failed to load created rule: %w", err)
	}

	return rule, nil
}

// GetRule retrieves a classification rule by ID
func (s *ClassificationService) GetRule(userID, id uint) (*models.ClassificationRule, error) {
	rule, err := s.classificationRepo.GetByID(userID, id)
	if err != nil {
		if errors.Is(err, repository.ErrClassificationRuleNotFound) {
			return nil, ErrClassificationRuleNotFound
		}
		return nil, fmt.Errorf("failed to get classification rule: %w", err)
	}
	return rule, nil
}

// GetAllRules retrieves all classification rules
func (s *ClassificationService) GetAllRules(userID uint) ([]models.ClassificationRule, error) {
	rules, err := s.classificationRepo.GetAll(userID)
	if err != nil {
		return nil, fmt.Errorf("failed to get classification rules: %w", err)
	}
	return rules, nil
}

// GetRulesByCategory retrieves all classification rules for a specific category
func (s *ClassificationService) GetRulesByCategory(userID, categoryID uint) ([]models.ClassificationRule, error) {
	// Validate category exists
	exists, err := s.categoryRepo.ExistsByID(userID, categoryID)
	if err != nil {
		return nil, fmt.Errorf("failed to validate category: %w", err)
	}
	if !exists {
		return nil, ErrInvalidCategoryID
	}

	rules, err := s.classificationRepo.GetByCategoryID(userID, categoryID)
	if err != nil {
		return nil, fmt.Errorf("failed to get classification rules: %w", err)
	}
	return rules, nil
}

// UpdateRule updates an existing classification rule
func (s *ClassificationService) UpdateRule(userID, id uint, input ClassificationRuleInput) (*models.ClassificationRule, error) {
	// Get existing rule
	rule, err := s.classificationRepo.GetByID(userID, id)
	if err != nil {
		if errors.Is(err, repository.ErrClassificationRuleNotFound) {
			return nil, ErrClassificationRuleNotFound
		}
		return nil, fmt.Errorf("failed to get classification rule: %w", err)
	}

	// Validate keyword
	keyword := strings.TrimSpace(input.Keyword)
	if keyword == "" {
		return nil, ErrInvalidKeyword
	}

	// Validate category exists
	exists, err := s.categoryRepo.ExistsByID(userID, input.CategoryID)
	if err != nil {
		return nil, fmt.Errorf("failed to validate category: %w", err)
	}
	if !exists {
		return nil, ErrInvalidCategoryID
	}

	// Validate amount range
	if input.MinAmount != nil && input.MaxAmount != nil && *input.MinAmount > *input.MaxAmount {
		return nil, ErrInvalidAmountRange
	}

	// Update fields
	rule.Keyword = keyword
	rule.CategoryID = input.CategoryID
	rule.MinAmount = input.MinAmount
	rule.MaxAmount = input.MaxAmount

	if err := s.classificationRepo.Update(rule); err != nil {
		return nil, fmt.Errorf("failed to update classification rule: %w", err)
	}

	// Reload to get updated category
	rule, err = s.classificationRepo.GetByID(userID, id)
	if err != nil {
		return nil, fmt.Errorf("failed to reload classification rule: %w", err)
	}

	return rule, nil
}

// DeleteRule deletes a classification rule by ID
func (s *ClassificationService) DeleteRule(userID, id uint) error {
	err := s.classificationRepo.Delete(userID, id)
	if err != nil {
		if errors.Is(err, repository.ErrClassificationRuleNotFound) {
			return ErrClassificationRuleNotFound
		}
		return fmt.Errorf("failed to delete classification rule: %w", err)
	}
	return nil
}

// SuggestCategory suggests categories based on transaction note and amount
// This is the core smart classification algorithm that runs entirely locally
// Requirement 2.1.1: Recommend most likely category based on historical data
// Requirement 2.1.2: Runs completely locally, no external data transmission
// Requirement 2.1.4: Match based on note keywords and amount range
func (s *ClassificationService) SuggestCategory(userID uint, note string, amount float64) ([]ClassificationSuggestion, error) {
	if strings.TrimSpace(note) == "" {
		return []ClassificationSuggestion{}, nil
	}

	// Get all matching rules based on keyword and amount
	matchingRules, err := s.classificationRepo.GetMatchingRules(userID, note, amount)
	if err != nil {
		return nil, fmt.Errorf("failed to get matching rules: %w", err)
	}

	if len(matchingRules) == 0 {
		return []ClassificationSuggestion{}, nil
	}

	// Calculate confidence scores and build suggestions
	suggestions := make([]ClassificationSuggestion, 0, len(matchingRules))

	// Find the maximum hit count for normalization
	maxHitCount := 0
	for _, rule := range matchingRules {
		if rule.HitCount > maxHitCount {
			maxHitCount = rule.HitCount
		}
	}

	for i := range matchingRules {
		rule := &matchingRules[i]
		confidence := s.calculateConfidence(note, amount, rule, maxHitCount)

		matchReason := s.buildMatchReason(note, amount, rule)

		suggestion := ClassificationSuggestion{
			CategoryID:  rule.CategoryID,
			Category:    &rule.Category,
			Confidence:  confidence,
			MatchedRule: rule,
			MatchReason: matchReason,
		}
		suggestions = append(suggestions, suggestion)
	}

	// Sort by confidence (highest first) - already partially sorted by hit_count from DB
	s.sortSuggestionsByConfidence(suggestions)

	// Deduplicate by category ID, keeping highest confidence
	suggestions = s.deduplicateSuggestions(suggestions)

	return suggestions, nil
}

// calculateConfidence calculates a confidence score for a rule match
// The score is based on:
// 1. Keyword match quality (exact match vs partial match)
// 2. Amount range match (if specified)
// 3. Historical hit count (popularity)
func (s *ClassificationService) calculateConfidence(note string, amount float64, rule *models.ClassificationRule, maxHitCount int) float64 {
	var confidence float64 = 0.0

	// Base score for keyword match (0.3 - 0.5)
	noteLower := strings.ToLower(note)
	keywordLower := strings.ToLower(rule.Keyword)

	if noteLower == keywordLower {
		// Exact match
		confidence += 0.5
	} else if strings.Contains(noteLower, keywordLower) {
		// Partial match - score based on keyword length relative to note
		keywordRatio := float64(len(rule.Keyword)) / float64(len(note))
		confidence += 0.3 + (0.2 * keywordRatio)
	}

	// Amount range match bonus (0.0 - 0.3)
	amountBonus := 0.0
	hasAmountConstraint := rule.MinAmount != nil || rule.MaxAmount != nil

	if hasAmountConstraint {
		inRange := true
		if rule.MinAmount != nil && amount < *rule.MinAmount {
			inRange = false
		}
		if rule.MaxAmount != nil && amount > *rule.MaxAmount {
			inRange = false
		}

		if inRange {
			// Calculate how well the amount fits in the range
			if rule.MinAmount != nil && rule.MaxAmount != nil {
				rangeSize := *rule.MaxAmount - *rule.MinAmount
				if rangeSize > 0 {
					// Closer to the middle of the range = higher score
					midPoint := (*rule.MinAmount + *rule.MaxAmount) / 2
					distanceFromMid := abs(amount - midPoint)
					normalizedDistance := distanceFromMid / (rangeSize / 2)
					amountBonus = 0.3 * (1 - normalizedDistance)
				} else {
					amountBonus = 0.3 // Exact amount match
				}
			} else {
				amountBonus = 0.2 // Only one bound specified
			}
		}
	}
	confidence += amountBonus

	// Historical popularity bonus (0.0 - 0.2)
	if maxHitCount > 0 {
		popularityRatio := float64(rule.HitCount) / float64(maxHitCount)
		confidence += 0.2 * popularityRatio
	}

	// Cap confidence at 1.0
	if confidence > 1.0 {
		confidence = 1.0
	}

	return confidence
}

// buildMatchReason builds a human-readable explanation for why this category was suggested
func (s *ClassificationService) buildMatchReason(note string, amount float64, rule *models.ClassificationRule) string {
	reasons := []string{}

	// Keyword match reason
	noteLower := strings.ToLower(note)
	keywordLower := strings.ToLower(rule.Keyword)

	if noteLower == keywordLower {
		reasons = append(reasons, fmt.Sprintf("备注完全匹配关键词'%s'", rule.Keyword))
	} else {
		reasons = append(reasons, fmt.Sprintf("备注包含关键词'%s'", rule.Keyword))
	}

	// Amount range reason
	if rule.MinAmount != nil && rule.MaxAmount != nil {
		reasons = append(reasons, fmt.Sprintf("金额 %.2f 在范围 %.2f-%.2f 内", amount, *rule.MinAmount, *rule.MaxAmount))
	} else if rule.MinAmount != nil {
		reasons = append(reasons, fmt.Sprintf("金额 %.2f >= %.2f", amount, *rule.MinAmount))
	} else if rule.MaxAmount != nil {
		reasons = append(reasons, fmt.Sprintf("金额 %.2f <= %.2f", amount, *rule.MaxAmount))
	}

	// Hit count reason
	if rule.HitCount > 0 {
		reasons = append(reasons, fmt.Sprintf("历史匹配 %d 次", rule.HitCount))
	}

	return strings.Join(reasons, "; ")
}

// sortSuggestionsByConfidence sorts suggestions by confidence in descending order
func (s *ClassificationService) sortSuggestionsByConfidence(suggestions []ClassificationSuggestion) {
	// Simple bubble sort for small arrays (typically < 10 items)
	n := len(suggestions)
	for i := 0; i < n-1; i++ {
		for j := 0; j < n-i-1; j++ {
			if suggestions[j].Confidence < suggestions[j+1].Confidence {
				suggestions[j], suggestions[j+1] = suggestions[j+1], suggestions[j]
			}
		}
	}
}

// deduplicateSuggestions removes duplicate category suggestions, keeping the highest confidence
func (s *ClassificationService) deduplicateSuggestions(suggestions []ClassificationSuggestion) []ClassificationSuggestion {
	seen := make(map[uint]bool)
	result := make([]ClassificationSuggestion, 0, len(suggestions))

	for _, suggestion := range suggestions {
		if !seen[suggestion.CategoryID] {
			seen[suggestion.CategoryID] = true
			result = append(result, suggestion)
		}
	}

	return result
}

// ConfirmSuggestion confirms a classification suggestion, incrementing the hit count
// This is called when the user accepts a suggested category
// Requirement 2.1.3: Update local classification model when user confirms/modifies
func (s *ClassificationService) ConfirmSuggestion(userID, ruleID uint) error {
	err := s.classificationRepo.IncrementHitCount(userID, ruleID)
	if err != nil {
		if errors.Is(err, repository.ErrClassificationRuleNotFound) {
			return ErrClassificationRuleNotFound
		}
		return fmt.Errorf("failed to confirm suggestion: %w", err)
	}
	return nil
}

// LearnFromTransaction creates or updates a classification rule based on a confirmed transaction
// This allows the system to learn from user behavior
func (s *ClassificationService) LearnFromTransaction(userID uint, note string, amount float64, categoryID uint) error {
	if strings.TrimSpace(note) == "" {
		return nil // Nothing to learn from empty notes
	}

	// Extract keywords from the note (simple approach: use the whole note as keyword)
	// In a more sophisticated implementation, we could use NLP to extract key phrases
	keyword := strings.TrimSpace(note)

	// Check if a rule already exists for this keyword and category
	existingRule, err := s.classificationRepo.GetByExactKeyword(userID, keyword)
	if err != nil && !errors.Is(err, repository.ErrClassificationRuleNotFound) {
		return fmt.Errorf("failed to check existing rule: %w", err)
	}

	if existingRule != nil {
		// Rule exists - increment hit count if same category
		if existingRule.CategoryID == categoryID {
			return s.classificationRepo.IncrementHitCount(userID, existingRule.ID)
		}
		// Different category - could create a new rule or update existing
		// For now, we'll create a new rule for the new category
	}

	// Create a new rule
	input := ClassificationRuleInput{
		UserID:     userID,
		Keyword:    keyword,
		CategoryID: categoryID,
	}

	// Set amount range based on the transaction amount (±20% range)
	minAmount := amount * 0.8
	maxAmount := amount * 1.2
	input.MinAmount = &minAmount
	input.MaxAmount = &maxAmount

	_, err = s.CreateRule(input)
	if err != nil {
		// If rule already exists (same keyword, same category), just increment hit count
		if errors.Is(err, ErrRuleAlreadyExists) {
			existingRule, err := s.classificationRepo.GetByExactKeyword(userID, keyword)
			if err == nil && existingRule.CategoryID == categoryID {
				return s.classificationRepo.IncrementHitCount(userID, existingRule.ID)
			}
		}
		return fmt.Errorf("failed to learn from transaction: %w", err)
	}

	return nil
}

// RuleExists checks if a classification rule exists by ID
func (s *ClassificationService) RuleExists(userID, id uint) (bool, error) {
	exists, err := s.classificationRepo.ExistsByID(userID, id)
	if err != nil {
		return false, fmt.Errorf("failed to check rule existence: %w", err)
	}
	return exists, nil
}

// abs returns the absolute value of a float64
func abs(x float64) float64 {
	if x < 0 {
		return -x
	}
	return x
}