Trek Recommendation System

Trekking Destination Recommender

Built as a freelance project in March 2022 — a recommendation system for trekking destinations in India. Put in a trek you know and it returns 10 similar ones, ranked by cosine similarity across description vectors. 208+ treks in the database.

Background

How the Recommendation Engine Works

1. 1
   Feature Extraction — Trek descriptions are vectorised using CountVectorizer, converting text into numerical feature vectors.
2. 2
   Similarity Calculation — Cosine similarity is computed between all trek vectors, capturing how similar any two treks are based on description content.
3. 3
   KNN Search — A NearestNeighbors model (scikit-learn) returns the k most similar treks to the query, using the cosine similarity matrix.
4. 4
   Ranking — Results are sorted by similarity distance and returned with trek images, descriptions, type labels, and percentage match scores.
5. 5
   Model Caching — The trained model is serialised to disk (minnn_rf.pkl) and loaded at startup, keeping response times under 1 second.

Core Algorithm

from sklearn.neighbors import NearestNeighbors
from sklearn.feature_extraction.text import CountVectorizer
import pandas as pd
import pickle

# Load trek dataset
df = pd.read_csv('b1.csv')

# Vectorise trek descriptions
vectorizer = CountVectorizer()
vectors = vectorizer.fit_transform(df['description'])

# Fit NearestNeighbors with cosine similarity
model = NearestNeighbors(metric='cosine', algorithm='brute')
model.fit(vectors)

# Get recommendations for a given trek
def recommend(trek_name, n=10):
    idx = df[df['title'] == trek_name].index[0]
    distances, indices = model.kneighbors(
        vectors[idx], n_neighbors=n + 1
    )
    results = []
    for i, dist in zip(indices[0][1:], distances[0][1:]):
        results.append({
            'title': df.iloc[i]['title'],
            'similarity': round((1 - dist) * 100, 2),
            'image': df.iloc[i]['image'],
            'type': df.iloc[i]['treck type']
        })
    return results

Screenshots

Technical Stack

Features