Travel Package Recommendation System Using Topic Modeling Approach

Marianne  P. Vitug,

Travel Package Recommendation System Using Topic Modeling Approach

Abstract:

The goal of the study is to build a recommender system that will incorporate important factors in designing a travel package – the inherent attributes of tourist destinations, and their distance from other tourist destinations. This project sourced its primary data from reviews from Tripadvisor collected using the package rvest from R. 37 current tourist destinations were included in the data gathering with 50 reviews gathered for each location. Distance matrices using Geodesic distance calculation and driving distance via Google API were gathered for this project. These were used as a penalizing factor to produce the hierarchy of the final set of recommendations. The Geodesic distances between the various tourist destinations were gathered using R ggmap and lmap libraries. For the Google driving distance, Google Cloud platform’s Distance Matrix API registration had been necessary to get the key that was used for the R code using gmapdistance library. The Google Distance Matrix API gave travel distance based on the recommended route for a supplied matrix of origins (start) and destinations (end point). For the modeling technique, since the data had characteristics that were unsuitable for the algorithms commonly used in recommender systems, topic modeling was used as an alternative method of extracting the intrinsic features of both the tourists and the locations. Latent Dirichlet Allocation (LDA) and Bidirectional Encoder Representations from Transformers (BERT) were used for topic modeling. The results were evaluated using a mix of eyeballing on top N words and intrinsic evaluation metrics through topics interpretability. Aside from this, the final list of recommendations was sent to Ark Travel’s President and the head of local tour operations for evaluation and successfully fitted their requirements. The model was able to produce recommendations which are deemed acceptable based on these criteria. The solution made using this recommender system can help not only the main stakeholders— the travelers and the travel agency, but also the business owners on less popular or just-emerging tourist destinations since they can also be recommended as long as they are part of the dataset.