Location based Recommendation System Using Transformer Based Sentiment Analysis and Similarity Measuring Metrics

The location recommendation system is designed to assist travelers or tourists to select destinations based on their personal preferences. Massive amount of data is being generated through various travel websites and social media platforms. Therefore, getting preferable suggestions is overwhelming. Additionally, research's has failed to build viable model considering both textual and numerical data simultaneously, as user preferences may be generated through either of these means. Therefore, we propose a collaborative filtering-based model that incorporates both textual and numerical data for better extraction of user preferences. Our approach involves using graded sentiment analysis and embedding techniques such as BERT, RoBERTa, DistilBERT, and MPNet as well as similarity measuring metrics such as cosine similarity, euclidean distance, Manhattan distance, Kendall rank, and Pearson correlation. Our proposed model, MPNet with Euclidean distance, has demonstrated promising results with a precision rate of 96% in recommending top 5 locations. The models performance make it a viable alternative for assisting individual by recommending personalized locations.