Travel Destination Recommendation System

Overview

This project implements a scalable, hybrid travel recommendation system natively designed for PySpark and Databricks. It combines Content-Based Filtering (via user-driven UI widgets) and Collaborative Filtering (using the Alternating Least Squares or ALS algorithm) to deliver highly personalized, top-15 travel destination rankings.

The system dynamically handles strict filtering requirements and features intelligent fallback protocols if initial user constraints yield zero results, ensuring continued relevance and functionality without breaking.

Features

• Hybrid Recommendation Architecture: Fuses content-based pre-filtering with matrix factorization.
• Interactive UI Widgets: Built-in Databricks widgets allow users to define preferences:
  • Max Budget (USD/Day)
  • Climate Type
  • Travel Type
  • Preferred Season
• Intelligent Fallback Logic: Automatically relaxes strict "Climate" and "Season" constraints sequentially if multi-factor limits cannot be met, protecting essential budget and travel-type requirements.
• Synthetic Interaction Generation: Bootstraps the ALS model with dynamically generated baseline user interactions (15,000 synthetic ratings) across 1,000 simulated user parameters.
• Evaluation Metrics: Validates predictive performance objectively using Root Mean Square Error (RMSE).
• Rich Output Format: Joins predictions back to the core dataset to display 17 dimensions of relevant destination data (e.g., predicted ratings, safety indexes, average temperature, visa requirements, currency).

Prerequisites

• Platform: Apache Spark / Databricks Platform.
• Runtime: Recommended DBR (Databricks Runtime) 10.4 LTS ML or higher to guarantee standard pyspark.ml and dbutils widget support.

Dataset

The primary recommendation model relies on a destination details dataset named ds1_grok. Ensure the dataset features the following core dimensions to function properly:

• country
• destination
• avg_cost_usd_per_day (Numeric)
• travel_type
• best_season
• climate
• And other dimensional columns required in final output (top_attraction, currency, safety_rating, etc.)

Note: The code first attempts to load the dataset as a native Databricks table (spark.table("ds1_grok")). If unavailable, it gracefully falls back to reading the raw CSV direct from DBFS (/FileStore/tables/ds1_grok.csv).

Usage Instructions

1. Import the Code: Create a new PySpark Notebook in your Databricks workspace and paste the script.
2. Mount/Upload Data: Register ds1_grok.csv as a Spark table or upload the file to DBFS at /FileStore/tables/.
3. Configure Dashboard Widgets: Run the first block to instantiate the UI. Interactive dropdown widgets will appear at the top of your notebook. Select your active preferences.
4. Execute to Train & Predict: Run the remainder of the notebook.
   • Pre-filtering isolates valid destinations.
   • ALS collaborative filtering trains on the remaining valid index.
   • An evaluation (RMSE) prints to standard out.
   • Ranked top-15 destinations render seamlessly via Databricks' display() function.

Pipeline Architecture

1. Widget Ingestion: Captures stateful user preferences.
2. Normalization: Concatenates and cleans destination strings, enforcing uniqueness.
3. Content Filtering: Narrow downs destinations via strict threshold logic or fallback cascades.
4. Collaborative Filtering (ALS):
   • Implements string indexing.
   • Evaluates train/test split via Spark ML.
   • Trains matrix parameters.
5. Prediction & Window Ranking: Extrapolates predicted user affinity across active filtered locations, maximizing performance overhead.
6. Data Presentation: Structures and explicitly types all final outputs for robust end-user display.