UCLA

This thesis aims to identify key topics from travel blog journals created by two bloggers Jones and handluggageonly and create interest directed recommendations to tourists. By applying Text Mining, sentiment analysis, the two writers exhibited different styles of travel guide recommendation but shares common traits in using sentimental words to lead their audiences through a trustful, exciting journey. Jones focus on providing insights in hostel rental versus handluggageonly shares his adventures on a personal level. Both Latent Dirichlet Allocation (LDA) and BERTopic successfully categorized the journals into meaningful topics and identifies popular themes such as beach/island, music festival, historical sites, airbnb rental allocations. However, BERTopic provides additional interaction feature, which enables a powerful travel guide recommendation system that links user input to relevant documents. Whether it is historical sites, or music festivals, by associating country with the relevant documents, tourists can determine their next trip destination by referencing the firsthand travel experience provided by professional travel bloggers. This model can efficiently capture important content with only one keyword input. It not only save time compared to manual search but also can capture all associated themes without having to input all the keywords. By inputting ”dessert”, you get various topic groups identified in the documents such as ”pizza”, ”dessert, pudding”, ”restaurants”. The topics also reflect multiple interpretations of the word. By applying the model for both travelers and travel agents, the model can align travelers interest with travel website contents and provide feedback for a more user-focused experience. Whether the ”catch” is beach, or museums, travel agents can incorporate those key attractions and create more personalized tour paths based on user preference.