Prediction Sequence Patterns of Tourist from the Tourism Website by Hybrid Deep Learning Techniques

Abstract

Tourism is an important industry that generates incomes and jobs in the country where this industry contributes considerably to GDP. Before traveling, tourists usually need to plan an itinerary listing a sequence of where to visit and what to do. To help plan, tourists usually gather information by reading blogs and boards where visitors who have previously traveled posted about traveling places and activities. Text from traveling posts can infer travel itinerary and sequences of places to visit and activities to experience. This research aims to analyze text postings using 21 deep learning techniques to learn sequential patterns of places and activities. The three main techniques are Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU) and a combination of these techniques including their adaptation with batch normalization. The output is sequential patterns for predicting places or activities that tourists are likely to go and plan to do. The results are evaluated using mean absolute error (MAE) and mean squared error (MSE) loss metrics. Moreover, the predicted sequences of places and activities are further assessed using a sequence alignment method called the Needleman–Wunsch algorithm (NW), which is a popular method to estimate sequence matching between two sequences.