Episode

Sustainable Recommender Systems for Tourism

Podcast: Data Skeptic
Published: Oct 9, 2025
Duration seconds: 2282
Processing state: processed
Canonical source: https://dataskeptic.com/blog/episodes/2025/sustainable-recommender-systems-for-tourism
Audio: https://pscrb.fm/rss/p/mgln.ai/e/35/traffic.libsyn.com/secure/dataskeptic/Ashmi_with_Ads_V2.mp3?dest-id=201630
JSON: /v1/public/podcasts/data-skeptic/episodes/sustainable-recommender-systems-for-tourism
Markdown: /podcast/data-skeptic/sustainable-recommender-systems-for-tourism.md

Actions

POST https://stenobird.com/v1/public/podcasts/data-skeptic/episodes/sustainable-recommender-systems-for-tourism/transcription-requests
Idempotently request low-priority transcript generation for this episode.
GET https://stenobird.com/podcast/data-skeptic/sustainable-recommender-systems-for-tourism.md
Read the agent-friendly Markdown representation of this episode resource.

Summary

AI-powered recommender systems can be redesigned to combat overtourism by actively promoting less crowded, sustainable destinations. This discussion explores using LLMs and synthetic data to balance user satisfaction with environmental and social responsibility in travel.

Topics

Recommender Systems
Sustainable Tourism
Large Language Models
Synthetic Data
Popularity Bias
RAG Architecture
AI Ethics
Data Augmentation

Highlights

Main idea: Recommender systems can move beyond 'popularity bias' to distribute tourism more equitably across destinations
Practical takeaway: Large Language Models can be used to generate synthetic datasets and knowledge bases for training specialized tourism models
Failure mode: Relying solely on popularity-based algorithms leads to overtourism in hotspots like Times Square while neglecting hidden gems
Technical approach: Integrating sustainability indicators directly into RAG (Retrieval-Augmented Generation) workflows to influence LLM outputs
Future challenge: The lack of a 'gold standard' for evaluating personalized, sustainable recommendations makes benchmarking difficult

Chapters

1:10 Research Focus: Sustainable Tourism Recommendations: Introduction to Ashmi's work at the Technical University of Munich regarding AI-driven sustainable travel recommendations.
6:55 The Problem of Popularity and Position Bias: An analysis of how current algorithms favor famous landmarks, contributing to the concentration of tourists in specific areas.
9:55 Using LLMs for Synthetic Data Generation: Exploring the pros and cons of using Large Language Models to create datasets when real-world data is scarce.
17:50 Architecture of a Knowledge-Based Recommender: A deep dive into building open-source datasets and knowledge bases that power text-to-recommendation pipelines.
23:45 Implementing Sustainability Filters: How to use temporal granularity and sustainability scores to recommend destinations with lower footfall during peak months.
32:05 Evaluating System Transparency and User Acceptance: Discussing the importance of A/B testing, explainability, and user interface design in the future of AI travel tools.
34:55 Testing LLM Adherence to Sustainability Constraints: Insights into using smaller, resource-constrained LLMs and evaluating their ability to follow sustainability instructions.