Classifying Sightseeing Tweets using Convolutional Neural Networks with Multi-Channel Distributed Representation

Contents posted on social media have attracted attention as a means to enhance the development of tourism, because there are huge amounts of information that are for the organizers of tourist spots and events. In this study, we focus on tweets posted by tourists including descriptions related to tourist spots and events. These kinds of tweets are called sightseeing tweets. To extract useful opinions from the sightseeing tweets, we define classes into which they are classified of them. These classified sightseeing tweets are used for analyzing opinions and polarities related to tourist spots and events. In addition, we propose a new model with convolutional neural networks to classify tweets. Distributed representation is one of the most well-known approaches to convert text data into numeric vectors so that text data can be input to neural networks. Our model utilizes multi-channel distributed representation, which is a hybrid representation of the word representation for text data. To evaluate the proposed model, we conducted experiments using actual sightseeing tweets. The experimental results show that the proposed model outperforms other deep-learning-based models.