An Images-Textual Hybrid Recommender System for Vacation Rental

To look for the specific vacation rental that suits ones personal preferences can be time consuming due to the wealth of information available on the Internet. Often collaborative filtering is used to help people narrow down their search results. However, most of the methods are solely based on the textual data which might insufficient to capture comprehensive details about the accommodation that suits individuals' preferences. The visual effects of the images, on the other hands, might reveal hidden users' preferences which cannot be told through the text. In this paper, an images-textual hybrid recommender system is proposed to enhance the preferable vacation accommodation prediction by leveraging the strength of both data collected from users' traveling histories. The proposed recommender system is demonstrated on the Airbnb dataset for all the advertised accommodation offers in Hong Kong. Around 1 million images features are extracted from a total of 110572 accommodations' images for similarity calculation. Rooms description and review scores are collected through a custom built web crawler program, the review scores are used as a reference to filter out the low quality accommodation prior to the implementation of the proposed recommender system. The proposed hybrid recommender system achieves better recommendations with an average precision score of 36.23%, which shows a 26.44% improvement compared to the baseline, which has a mean precision score of 9.79%.