​

Glood’s Similar Products: Smart Alternative Discovery

​

Algorithm Architecture

​

AI Model Specializations

​

OpenAI GPT-4: Semantic Understanding

​

Natural Language Processing

• Description Embedding: Converts product descriptions into high-dimensional vectors
• Feature Extraction: Identifies key product attributes from text
• Intent Matching: Understands the core purpose and use case
• Specification Parsing: Extracts technical details for comparison

​

Semantic Similarity Scoring

# Conceptual representation
similarity = cosine_similarity(
    gpt4_embed(product_A_description),
    gpt4_embed(product_B_description)
)

​

Cross-lingual Capabilities

• Matches products across different languages
• Understands regional terminology variations
• Handles technical jargon and colloquialisms

​

Google Gemini Pro: Visual Intelligence

​

Image Understanding

• Style Detection: Identifies design patterns, aesthetics, and visual themes
• Color Analysis: Extracts dominant colors and palettes
• Shape Recognition: Understands product form factors
• Texture Identification: Recognizes materials and finishes

​

Multi-modal Fusion

• Combines image and text understanding
• Validates text claims with visual evidence
• Identifies discrepancies between descriptions and images

​

Fashion & Design Expertise

• Pattern matching in clothing and textiles
• Furniture style classification
• Artistic and decorative element recognition

​

Glood Proprietary Models: E-commerce Optimization

​

Behavioral Similarity

• Co-view Patterns: Products frequently viewed together
• Switch Rates: How often customers switch between products
• Comparison Behavior: Items compared in the same session
• Purchase Substitution: Actual replacement patterns

​

Commercial Factors

• Price Band Matching: Keeps recommendations in similar price ranges
• Brand Affinity: Considers brand preferences and loyalty
• Quality Tier Alignment: Matches premium with premium
• Availability Scoring: Prioritizes in-stock alternatives

​

Similarity Dimensions

​

1. Functional Similarity

• Running shoes � Other running shoes
• Coffee makers � Espresso machines
• Phone cases � Protective covers

​

2. Visual Similarity

• Similar color schemes
• Matching design aesthetics
• Comparable styles or patterns

​

3. Contextual Similarity

• Beach towels � Beach umbrellas
• Yoga mats � Yoga blocks
• Camping tents � Sleeping bags

​

4. Attribute Similarity

• Same size or dimensions
• Compatible features
• Similar performance metrics

​

Advanced Techniques

​

Hierarchical Similarity

​

Dynamic Weight Adjustment

• Product Category: Fashion emphasizes visual, electronics emphasizes specs
• Price Point: Luxury items weight brand higher
• Customer Segment: Tech-savvy users get spec-heavy matches
• Session Context: Search queries influence weight distribution

​

Negative Sampling

• Products with high return rates when bought together
• Items frequently removed from comparison
• Negative review mentions of alternatives

​

Real-time Processing

​

Feature Extraction Pipeline

1. Image Processing (50ms)
   • Resize and normalize
   • Extract visual features
   • Generate embeddings
2. Text Processing (30ms)
   • Tokenize descriptions
   • Generate semantic embeddings
   • Extract entities and attributes
3. Similarity Computation (20ms)
   • Calculate multi-modal distances
   • Apply category-specific weights
   • Generate ranked list

​

Caching Strategy

• Embedding Cache: Pre-computed embeddings for all products
• Similarity Matrix: Pre-calculated top-N similar items
• Real-time Adjustment: Dynamic re-ranking based on context

​

Quality Metrics

​

Relevance Metrics

• Click-through Rate: 45% on similar product carousels
• Dwell Time: 3.2x longer on recommended products
• Conversion Rate: 22% purchase similar items
• Return Rate: 15% lower for AI-recommended alternatives

​

Diversity Metrics

• Category Coverage: Recommendations span appropriate subcategories
• Price Range: �20% of original product price
• Brand Mix: Balance between same-brand and alternatives
• Visual Variety: Different colors/styles when appropriate

​

Use Case Examples

​

Fashion Retail

Viewing: Blue Denim Jacket
Similar Products:
  - Light Wash Denim Jacket (color variation)
  - Canvas Jacket (material alternative)
  - Bomber Jacket (style alternative)
  - Designer Denim Jacket (premium option)

​

Electronics

Viewing: 65" 4K Smart TV
Similar Products:
  - 55" 4K Smart TV (size variation)
  - 65" OLED TV (technology upgrade)
  - 65" QLED TV (brand alternative)
  - 70" 4K Smart TV (size upgrade)

​

Home Decor

Viewing: Modern Floor Lamp
Similar Products:
  - Arc Floor Lamp (style variation)
  - Tripod Floor Lamp (design alternative)
  - Smart Floor Lamp (feature upgrade)
  - Industrial Floor Lamp (aesthetic alternative)

​

Continuous Improvement

​

Learning Mechanisms

1. Click Feedback: Learn from which similarities users explore
2. Purchase Analysis: Understand actual substitution patterns
3. Return Analysis: Identify poor similarity matches
4. Session Analysis: Learn from comparison behavior

​

Model Updates

• Weekly Retraining: Behavioral models updated weekly
• Daily Embeddings: New products get embeddings daily
• Real-time Adjustments: Weights adjust based on performance
• Quarterly Reviews: Major algorithm improvements

​

Glood Dashboard Configuration

​

Setting Up Similar Products

1. Quick Setup
   Section Name: "You May Also Like"
   Algorithm: Similar Products
   Number of Products: 6
   Layout: Grid (2x3)
   Similarity Type: Balanced
   
2. Similarity Controls
   • Visual Weight (0-100%): How much appearance matters
   • Price Weight (0-100%): Keep alternatives in budget
   • Brand Weight (0-100%): Same brand vs competitors
   • Category Strictness: Same category or explore related
3. Manual Similarity Mapping
   product_id,similar_1,similar_2,similar_3,priority
   SHOE001,SHOE002,SHOE003,SHOE004,high
   BAG001,BAG002,BAG003,BAG004,medium
   

​

Glood’s Similarity Dashboard

// Glood similarity configuration
{
  "similarity_config": {
    "algorithm": "hybrid",
    "weights": {
      "visual": 0.4,
      "semantic": 0.3,
      "behavioral": 0.2,
      "price": 0.1
    },
    "constraints": {
      "price_variance": 0.3,  // ±30% price
      "min_score": 0.6,       // Minimum similarity
      "max_results": 12,      // Pool size
      "diversity": 0.3        // Result variety
    }
  }
}

​

Glood-Specific Features

​

Visual Similarity Toggle

• Fashion: High visual weight (70%)
• Electronics: Spec-based (visual 20%)
• Furniture: Balanced (visual 50%)
• Consumables: Feature-based (visual 10%)

​

Smart Filtering Rules

1. Stock Status: Hide out-of-stock alternatives
2. Price Bands: Define acceptable price ranges
3. Brand Rules: Same-brand only or mix
4. Exclusions: Never show certain products
5. Promotions: Boost sale items

​

A/B Testing Framework

Test Variations:
  Control: AI-only similarity
  Variant A: Manual overrides + AI
  Variant B: Visual-heavy matching
  Variant C: Price-conscious alternatives
  
Metrics Tracked:
  - Click-through rate
  - Conversion rate
  - Time to purchase
  - Customer satisfaction

​

Glood Analytics for Similar Products

​

Performance Metrics

• Alternative Click Rate: 45% average
• Cross-sell Success: 23% buy alternatives
• Out-of-stock Recovery: 67% find substitute
• Session Extension: +4.2 pages per visit

​

Popular Use Cases

​

Out-of-Stock Management

Scenario: Product unavailable
Glood Action: Auto-show similar in-stock items
Result: 72% cart recovery rate

​

Price Sensitivity

Scenario: Customer views premium product
Glood Action: Include budget alternatives
Result: 34% convert to lower price point

​

Style Discovery

Scenario: Fashion browsing
Glood Action: Show different colors/patterns
Result: 2.8x items viewed per session

​

Implementation Best Practices

​

With Glood Dashboard

1. Initial Setup (5 minutes)
   • Install Glood from Shopify App Store
   • Create “Similar Products” section
   • Choose display locations
   • Customize appearance
2. Optimization (Ongoing)
   • Review weekly analytics
   • Adjust similarity weights
   • Add manual mappings for key products
   • Test different layouts
3. Advanced Configuration
   • Set up category-specific rules
   • Create seasonal similarity adjustments
   • Configure mobile vs desktop differences
   • Implement personalized similarity

​

Common Glood Configurations

​

Fashion Store

Visual Weight: 70%
Brand Weight: 20%
Price Weight: 10%
Show: Different colors, similar styles

​

Electronics Store

Spec Weight: 60%
Price Weight: 25%
Brand Weight: 15%
Show: Compatible alternatives

​

Home Decor

Visual Weight: 50%
Style Weight: 30%
Price Weight: 20%
Show: Complementary pieces

​

Success with Glood Similar Products

​

Case Studies

• Increased page views by 156%
• Reduced bounce rate by 42%
• Generated $2.3M additional revenue

• 89% out-of-stock recovery rate
• 34% higher AOV
• 4.5/5 customer satisfaction

• 67% explored similar styles
• 23% bought multiple items
• 3x longer session duration