NER for Extracting Tourist Attractions from Textual Information: A BERT-BiLSTM-CRF Approach
Shankhanil Borthakur, Jignesh N. Sarvaiya · 2023
Planning a trip typically involves searching for tourist destinations, and travel blogs are a rich source of information for such destinations. However, these blogs typically contain an extensive amount of text and descriptions of multiple tourist attractions, which can be time-consuming to read through. To solve this challenge, Named Entity Recognition (NER) can be used to help users in extracting tourist destinations from travel narratives. In this work, a BERT-BiLSTM-CRF NER model is proposed that combines contextualized word embeddings from BERT, BiLSTM for capturing sequential information, and CRF for assigning labels to sequential data. The model was trained on a public tourism NER dataset to identify tourist attractions present in the textual data and categorize them into natural attractions, heritage sites, and purposefully built attractions. Experimental tests and comparisons with other models were conducted. When performing NER on tourism corpus, the proposed model outperforms the compared models with respect to precision, recall, and F1-score, achieving 77.2%, 83.04%, and 80.01% values respectively.