An Intelligent Image Based Recommendation System for Tourism

Abstract

Tourism is a major source of revenue for many countries around the world. It provides jobs for the local residents, preserves natural resources, and uplifts culture and heritage, further economically benefiting the destination. In order to improve tourism, a system is developed which analyses the social media activity of users to understand their emotions. Sentiment analysis is generally done on text from comments or reviews, but this new system takes a different approach and tries to perform a sentiment analysis on the images that a user likes on social networking applications and detects features from these images to understand the mood of the user. These images are classified as ‘happy’ or ‘sad’, which is used to develop a context-based recommender system. To facilitate this several classification methods are implemented and compared to obtain the best method. The images are classified using Convolutional neural networks with different numbers of layers, by using Transfer learning with VGG16 and Inception Model, and by adding a novel layer of bilinear pooling to the VGG16 model to study its effect on the performance. The psychological effects of the emotion of a user on tourism recommendations are studied and relevant recommendations are made.