Smart Booking Without Looking: Providing Hotel Recommendations in the TripRebel Portal

Hotel booking should be simple, but in practice, it’s anything but. The avalanche of choices overwhelms travelers, and generic lists just don’t cut it. TripRebel set out to actually improve this experience—by not just offering hotels, but useful recommendations tailored both to the user and the specific hotel a traveler is viewing.

The messiness comes from having to balance two key use cases: surfacing high-quality hotels that match a user’s implicit preferences (gleaned from their clicks, views, bookings), and showing context-aware alternatives—think: you’re looking at Hotel A, what other, actually viable options are nearby, with better rates or relevant amenities? That meant supporting stringent requirements. Filtering needed to be flexible (location, attributes), recommendations had to be personalized or context-sensitive, new data should impact results instantly, and, most importantly, everything had to be real-time. Most off-the-shelf recommender frameworks (MyMediaLite, Mahout) fell short—great with explicit feedback, but not handling content/location well and too slow for live consumer interfaces.

So, we built a hotel recommender system around Apache Solr. Why Solr? Its geospatial support, built-in similarity (MoreLikeThis), and horizontal scalability let us handle both location-based and user-based relevance at scale, and incorporate new user-hotel data the moment it arrives. Under the hood, we indexed all of Expedia Affiliate Network’s hotels and rooms, logged real user interactions, and exposed everything via a REST API. Algorithms include content-based filtering (leveraging hotel metadata), user-based collaborative filtering (with a smart context-sensitive variant: only recommend based on users who interacted with the hotel currently being viewed), and a hybrid approach where the TripRebel portal specifies how much each strategy contributes to the final list.

Does context awareness actually translate into better engagement? Early results say yes. Recommendations using our context-aware collaborative filtering drove a 4.3% conversion rate—almost double the baseline—by adapting instantly to both the person and the page. Swapping hotels while searching becomes an easy, value-adding step, not another round of second-guessing.

Looking back, tight integration between recommender logic, storage, and live data flow is what made this possible. Solr turned out to be more than just a database; it let us treat hotel search as a rich, evolving conversation. There’s more left to do—especially dynamic weighting in the hybrid setup—but the core insight holds: real-time, context-rich filtering beats static lists, both for users and for business impact.

@inproceedings{traub2015smart,
  title = {Smart booking without looking: providing hotel recommendations in the TripRebel portal},
  author = {Traub, M. and Kowald, D. and Laci\'{c}, E. and Schoen, P. and Supp, G. and Lex, E.},
  year = {2015},
  booktitle = {Proceedings of the 15th International Conference on Knowledge Technologies and Data-driven Business},
  pages = {50},
  publisher = {ACM},
}