Flatlogic Bot 6bd90312de Initial import

2026-03-04 18:25:09 +00:00

12 KiB

Raw Permalink Blame History

Fallback Recommendation System

Overview

The AI recommendation system now implements a tiered fallback strategy that ensures meaningful service recommendations for almost all trips, instead of simply returning recommend: false.

Philosophy

Old Approach (Rejected):

Binary decision: recommend tour OR reject
Strict criteria: 2+ days, 3+ places, qualified activities
Result: Many viable trips got no recommendations

New Approach (Implemented):

Tiered recommendations: Best match → Fallback services → Only reject trivial trips
Flexible criteria: Consider trip characteristics holistically
Result: Almost all trips get helpful service suggestions

Recommendation Tiers

Tier 1: Matched Daily Tours (Confidence: 0.70-0.95)

Trigger: Place types match existing tour routes with ≥50% overlap

Services:

red_tour - Museums, valleys, Göreme area
green_tour - Underground cities, Ihlara Valley, nature
blue_tour - Off-beaten path, quiet villages
balloon_day - Balloon flight + light tour

Example:

Trip: 3 days, 12 places (museums, valleys, underground cities)
Density: 42 (HIGH)
→ Recommendation: red_tour (daily_tour type)
→ Confidence: 0.85
→ Reason: "Your plan matches Red Tour route with 75% overlap"

Tier 2: Fallback Services (Confidence: 0.55-0.70)

Trigger: No perfect tour match, but trip has characteristics that benefit from professional services

Fallback 2A: Short Dense Trips

Criteria: 1 day + density ≥30

Logic:

If travelers ≥4 → private_guide (confidence: 0.65)
If travelers <4 → driver_car (confidence: 0.60)

Example:

Trip: 1 day, 5 places, density: 35
Travelers: 2
→ Recommendation: driver_car
→ Confidence: 0.60
→ Reason: "A driver service would help you maximize your limited time"

Fallback 2B: Long Distance Trips

Criteria: Total distance ≥50km

Service: driver_car (confidence: 0.65)

Example:

Trip: 2 days, 4 places, distance: 85km
→ Recommendation: driver_car
→ Confidence: 0.65
→ Reason: "The distances between your destinations make a driver service valuable"

Fallback 2C: Multiple Destinations

Criteria: 3+ places (no tour match)

Service: private_guide (confidence: 0.55)

Example:

Trip: 2 days, 4 places, density: 25
→ Recommendation: private_guide
→ Confidence: 0.55
→ Reason: "A private guide could enhance your experience across multiple sites"

Fallback 2D: Large Groups

Criteria: 4+ travelers

Service: private_guide (confidence: 0.60)

Example:

Trip: 2 days, 3 places
Travelers: 5
→ Recommendation: private_guide
→ Confidence: 0.60
→ Reason: "Your group size makes a private guide service worthwhile"

Tier 3: AI Fallback (Confidence: 0.35-0.55)

Trigger: Rule-based fallbacks don't match, but AI finds value

Services: Any service type based on AI analysis

Example:

Trip: 1 day, 2 places, density: 18
But: Historical sites requiring expert knowledge
→ AI Recommendation: private_guide
→ Confidence: 0.45
→ Reason: "Historical context would significantly enhance your experience"

Tier 4: No Recommendation (Confidence: <0.35)

Trigger: Trip is truly trivial

Criteria:

1 place OR
<5km total distance AND <2 hours total time

Example:

Trip: 1 place, 2km, 1 hour
→ Recommendation: None (recommend: false)
→ Reason: "Your trip is simple enough to manage independently"

Decision Flow

┌─────────────────────────────────────────────────────────────┐
│                    TRIP ANALYSIS                             │
│  Calculate: density, distance, time, place count            │
└────────────────────────┬────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────┐
│              TIER 1: MATCHED DAILY TOUR?                     │
│  Query database for tours matching place types              │
└────────────┬────────────────────────────┬───────────────────┘
             │ YES (confidence ≥0.5)      │ NO
             ▼                            ▼
    ┌────────────────────┐    ┌──────────────────────────────┐
    │ Return Tour        │    │  TIER 2: FALLBACK SERVICES?  │
    │ (red/green/blue)   │    │  Check trip characteristics  │
    └────────────────────┘    └──────┬───────────────────┬───┘
                                     │ YES               │ NO
                                     ▼                   ▼
                          ┌──────────────────┐  ┌────────────────┐
                          │ Return Fallback  │  │ TIER 3: AI     │
                          │ (private_guide/  │  │ Analysis       │
                          │  driver_car)     │  └────┬───────┬───┘
                          └──────────────────┘       │ YES   │ NO
                                                     ▼       ▼
                                          ┌──────────────┐ ┌────────┐
                                          │ Return AI    │ │ TIER 4 │
                                          │ Suggestion   │ │ Reject │
                                          └──────────────┘ └────────┘

Fallback Logic Implementation

Rule-Based Fallbacks (Lines 534-710)

// Check if trip is truly trivial
const isTrivialTrip = totalPlaces <= 1 && totalDistanceKm < 5 && totalTimeHours < 2;

if (isTrivialTrip) {
  return { recommend: false, ... };
}

// Fallback 1: Short but dense trips
if (totalDays === 1 && maxDensityScore >= 30) {
  if (travelers >= 4) {
    return { recommend: true, slug: 'private_guide', confidence: 0.65, ... };
  } else {
    return { recommend: true, slug: 'driver_car', confidence: 0.60, ... };
  }
}

// Fallback 2: Long distances
if (totalDistanceKm >= 50) {
  return { recommend: true, slug: 'driver_car', confidence: 0.65, ... };
}

// Fallback 3: Multiple places
if (totalPlaces >= 3) {
  return { recommend: true, slug: 'private_guide', confidence: 0.55, ... };
}

// Fallback 4: Large groups
if (travelers >= 4) {
  return { recommend: true, slug: 'private_guide', confidence: 0.60, ... };
}

AI Fallback (Lines 774-940)

Updated AI prompt with fallback instructions:

FALLBACK STRATEGY: Even if no perfect tour match, consider:
- Short but dense trips (1 day, density ≥30) → private_guide or driver_car
- Long distances (≥50km) → driver_car
- Large groups (≥4 people) → private_guide
- Multiple places (≥3) → private_guide

ONLY return recommend:false if trip is truly trivial (1 place, <5km, <2 hours).

Adjusted confidence threshold:

// Old: if (analysis.confidence < 0.6) analysis.recommend = false;
// New: if (analysis.confidence < 0.35) analysis.recommend = false;

Benefits of Fallback System

1. Better User Experience

Users get helpful suggestions even for non-standard trips
No frustrating "no recommendations" messages
More opportunities for service providers

2. Increased Conversion

More trips trigger recommendations
Lower confidence recommendations still provide value
Users can make informed decisions

3. Flexible Service Matching

Not limited to predefined tour routes
Can recommend services based on trip characteristics
Adapts to various trip types

4. Transparent Confidence Levels

Users see confidence scores
Can judge recommendation quality
Debug info explains reasoning

Confidence Level Interpretation

Confidence	Meaning	User Action
0.85-1.00	Highly Recommended	Strong match, definitely consider
0.70-0.84	Recommended	Good match, worth exploring
0.55-0.69	Suggested	Helpful but optional
0.40-0.54	Optional	Consider if interested
0.35-0.39	Marginal	Minimal benefit
<0.35	Not Recommended	Self-planning sufficient

Examples

Example 1: Short Dense Trip (Fallback 2A)

{
  "trip": {
    "days": 1,
    "places": 5,
    "density": 35,
    "travelers": 2
  },
  "recommendation": {
    "recommend": true,
    "recommended_type": "driver_car",
    "daily_tour_slug": "driver_car",
    "confidence": 0.60,
    "reason": "A driver service would help you maximize your limited time",
    "why_better_than_self": [
      "Comfortable transportation between sites",
      "No parking hassles",
      "More time at attractions",
      "Local driver knows best routes"
    ]
  }
}

Example 2: Long Distance Trip (Fallback 2B)

{
  "trip": {
    "days": 2,
    "places": 4,
    "distance": 85,
    "travelers": 3
  },
  "recommendation": {
    "recommend": true,
    "recommended_type": "driver_car",
    "daily_tour_slug": "driver_car",
    "confidence": 0.65,
    "reason": "The distances between your destinations make a driver service valuable",
    "why_better_than_self": [
      "Comfortable long-distance travel",
      "No navigation stress",
      "Flexible stops along the way",
      "Arrive refreshed at each destination"
    ]
  }
}

Example 3: Large Group (Fallback 2D)

{
  "trip": {
    "days": 2,
    "places": 3,
    "travelers": 5
  },
  "recommendation": {
    "recommend": true,
    "recommended_type": "private_guide",
    "daily_tour_slug": "private_guide",
    "confidence": 0.60,
    "reason": "Your group size makes a private guide service worthwhile",
    "why_better_than_self": [
      "Keep everyone together",
      "Customized to group interests",
      "Better group coordination",
      "Shared cost makes it economical"
    ]
  }
}

Example 4: Trivial Trip (Tier 4 - Rejected)

{
  "trip": {
    "days": 1,
    "places": 1,
    "distance": 2,
    "time": 1
  },
  "recommendation": {
    "recommend": false,
    "reason": "Your trip is simple enough to manage independently",
    "confidence": 0
  }
}

Testing Scenarios

Scenario 1: One Day, High Density

Input: 1 day, 6 places, density: 40, 2 travelers
Expected: driver_car, confidence: 0.60
Reason: Short but dense trip fallback

Scenario 2: Two Days, Long Distance

Input: 2 days, 3 places, 95km, 3 travelers
Expected: driver_car, confidence: 0.65
Reason: Long distance fallback

Scenario 3: Small Trip, Large Group

Input: 1 day, 2 places, 4 travelers
Expected: private_guide, confidence: 0.60
Reason: Large group fallback

Scenario 4: Multiple Places, No Match

Input: 2 days, 4 places, no tour match
Expected: private_guide, confidence: 0.55
Reason: Multiple destinations fallback

Scenario 5: Truly Trivial

Input: 1 place, 2km, 1 hour
Expected: recommend: false
Reason: Trip is trivial

Migration Notes

Breaking Changes

None - API response format unchanged
Confidence thresholds adjusted (0.6 → 0.35)

Behavioral Changes

More trips now receive recommendations
Lower confidence recommendations are now valid
recommend: false is much rarer

UI Impact

No changes required
Confidence badges already display properly
Lower confidence recommendations show appropriately

Future Enhancements

Dynamic Confidence Thresholds
- Adjust based on user feedback
- A/B test different thresholds
More Fallback Types
- Photography tours for scenic trips
- Culinary tours for food-focused trips
- Adventure tours for active trips
Personalized Fallbacks
- Consider user history
- Learn from past bookings
- Adapt to user preferences
Seasonal Adjustments
- Higher confidence for peak season
- Different services for off-season
- Weather-based recommendations

12 KiB Raw Permalink Blame History

Fallback Recommendation System

Overview

Philosophy

Recommendation Tiers

Tier 1: Matched Daily Tours (Confidence: 0.70-0.95)

Tier 2: Fallback Services (Confidence: 0.55-0.70)

Fallback 2A: Short Dense Trips

Fallback 2B: Long Distance Trips

Fallback 2C: Multiple Destinations

Fallback 2D: Large Groups

Tier 3: AI Fallback (Confidence: 0.35-0.55)

Tier 4: No Recommendation (Confidence: <0.35)

Decision Flow

Fallback Logic Implementation

Rule-Based Fallbacks (Lines 534-710)

AI Fallback (Lines 774-940)

Benefits of Fallback System

1. Better User Experience

2. Increased Conversion

3. Flexible Service Matching

4. Transparent Confidence Levels

Confidence Level Interpretation

Examples

Example 1: Short Dense Trip (Fallback 2A)

Example 2: Long Distance Trip (Fallback 2B)

Example 3: Large Group (Fallback 2D)

Example 4: Trivial Trip (Tier 4 - Rejected)

Testing Scenarios

Scenario 1: One Day, High Density

Scenario 2: Two Days, Long Distance

Scenario 3: Small Trip, Large Group

Scenario 4: Multiple Places, No Match

Scenario 5: Truly Trivial

Migration Notes

Breaking Changes

Behavioral Changes

UI Impact

Future Enhancements

12 KiB

Raw Permalink Blame History