# Hotel Recommendation System

**Authors:** Aditi A. Mavalankar, Ajitesh Gupta, Chetan Gandotra, Rishabh Misra  
**arXiv:** [1908.07498](https://arxiv.org/abs/1908.07498)  
**Date:** 2019-08-20  
**Summary page:** https://www.summarizepaper.com/en/paper/1908.07498v2/

## Key points

- Booking a good place to stay is one of the most crucial tasks when planning a trip
- Researchers are developing hotel recommendation systems to help users choose the right hotel
- Expedia's hotel recommendation dataset was used in this study to gain insights into personalized hotel recommendations
- Price, rating, and location were found to be useful features for predicting hotel clusters
- Various approaches including Boosting Trees, Logistic Regression, SVMRank, Random Forests along with Ensembling over these methods were studied to rank hotels using z-score normalization
- Text-based features from reviews were extracted in another study to model preferences and opinion detection using wordnet
- A combination of cluster-based model and collaborative filtering was proposed as an effective way of providing good recommendations while also addressing cold start issues
- Relations between hotels were identified based on users who booked them seasonality factors like prices etc., and filtered based on occurrence using measures like Jaccard coefficient and Simpson coefficient.
- The authors evaluated their approach by comparing MAP@5 scores obtained from different methods used for both training and test datasets.
- The results showed that their method outperformed other approaches in terms of accuracy when predicting five hotel clusters that users are more likely to book given 100 distinct clusters.
- This study contributes towards developing better hotel recommendation systems that can provide personalized recommendations based on user preferences while addressing cold-start issues.
- The findings of this study can be useful for online travel agencies like Expedia to improve their hotel recommendation systems and enhance user experience.

## Summary

Planning a trip involves several important tasks, and one of the most crucial is booking a good place to stay. With thousands of hotels available online for every destination, choosing the right one can be overwhelming. To address this challenge, researchers have been working on developing hotel recommendation systems that can predict and recommend hotel clusters to users based on their preferences. In this study, the authors used Expedia's hotel recommendation dataset which contains various features that help in understanding the process behind user choices. To gain insights into personalized hotel recommendations for Expedia, the authors reviewed relevant literature on similar problems. One such study used a different dataset and found price and rating along with location to be useful features for predicting hotel clusters. The authors also studied various approaches including Boosting Trees, Logistic Regression, SVMRank, Random Forests along with Ensembling over these methods to rank hotels using z-score normalization. Another study tackled the cold-start problem by extracting text-based features from reviews to model preferences and opinion detection using wordnet. They then combined these features to get a final review score according to which they ranked hotels. A combination of cluster-based model and collaborative filtering was proposed in another study as an effective way of providing good recommendations while also addressing cold start issues. In yet another study relations between hotels were identified based on users who booked them seasonality factors like prices etc., and filtered based on occurrence using measures like Jaccard coefficient and Simpson coefficient. This paper placed high importance on changes in user preferences for hotels by identifying transition links between them. The authors evaluated their approach by comparing MAP@5 scores obtained from different methods used for both training and test datasets. The results showed that their method outperformed other approaches in terms of accuracy when predicting five hotel clusters that users are more likely to book given 100 distinct clusters. Overall, this study contributes towards developing better hotel recommendation systems that can provide personalized recommendations based on user preferences while addressing cold-start issues. The findings of this study can be useful for online travel agencies like Expedia to improve their hotel recommendation systems and enhance user experience.