LDA-Based Personalized recommendation for Airbnb

106

Airbnb is one of the most successful sharing economy platforms in the hospitality industry. Although the availability of large-scale reviews can be beneficial but it is more difficult in the decision-making process, because of the huge amount of reviews which make guests confused in selecting the best possible and suitable properties. In this thesis, we propose a personalized recommender system by applying LDA to extract latent topics of textual resource of each property and use the probability of topic distribution to represent the features of each property. Further, construct guest profile based on guest’s historical records in order to realize guest preference. Finally, for each candidate property, we consider the profiles of property and guest to estimate a sorted recommend list for the guest. For the evaluation, we adopt Recall to evaluate the recommendation performance. The experimental result shows that our LDA-based model performs better than the baseline. Afterwards, we compare the performance among different textual information which shows the review and rating score are appropriate resource for the property representation and guest preference on the LDA-based personalized recommender system.