Arsitektur Sistem Rekomendasi

Content-Based Filtering dengan SBERT dan Cosine Similarity

📊 Diagram Alur Sistem

File GPX
Input
Preprocessing Cleaning, Lowercase
Data Processing
SBERT Embedding 384 dimensi
Feature Extraction
Cosine Similarity Q·D / (||Q||×||D||)
Similarity
Top-N Ranking Rekomendasi
Output

1. Data Collection & Fusion

  • ✓ Ekstraksi fitur dari GPX (jarak, elevasi, grade, durasi Naismith)
  • ✓ Deskripsi manual (vegetasi, sumber air, panorama)
  • ✓ Koordinat rute untuk visualisasi peta

2. Text Preprocessing

  • ✓ Data Cleaning (regex: URL, karakter non-ASCII)
  • ✓ Case Folding (lowercase)
  • ✓ Stopword Removal Selektif (pertahankan negasi & kata sifat)
  • ✗ No Stemming (SBERT sensitif konteks)

3. SBERT Embedding

  • ✓ Model: paraphrase-multilingual-MiniLM-L12-v2
  • ✓ Dimensi embedding: 384
  • ✓ Support bahasa Indonesia

4. Cosine Similarity

Sim(Q, D) = (Q · D) / (||Q|| × ||D||)

  • Q = Query embedding (user search)
  • D = Document embedding (route)
  • Range = -1 to 1 (1 = identical)

🛠️ Technology Stack

🐘
Laravel 11
Backend Framework
🐍
Python 3.11
ML Processing
🤖
SBERT
Sentence Transformers
🗺️
Leaflet.js
Map Visualization