Ranking Roughly Tourist Destinations Using BERT-Based Semantic Search

With the development of the Internet, extensive online reviews have been generated in the tourism field. To utilize big data and avoid the shortcomings of traditional techniques, we designed a tourist destination ranking system by introducing a semantic search that extracts data related to the input query. For this system, we reviewed tourist spot reviews and pre-processed text reviews. Then, we embed the corpus and query using SBERT, measure their similarity, and leave data similar to the query above the threshold. By implementing a count-based ranking algorithm with the data within the boundary, the tourist destinations are derived in a semantically similar order to the query. We entered three queries to obtain the top 5 relevant tourist destinations. Although there are problems with optimal thresholds and imbalanced data, semantic search derives information of desired conditions and may be referenced in policymaking and recommendation systems.