CrowdTravel: Leveraging Cross-Modal CrowdSourced Data for Fine-Grained and Context-Based Travel Route Recommendation

Travel route planning is generally a very time-consuming task due to massive travel information and diverse travel needs. In this paper, we propose CrowdTravel to automatically extract context-related scenic route recommendation through cross-modal mining and crowd intelligence extraction.First, we leverage the hybrid CNN-RNN model to learn the relationship between the photos and descriptive texts in travelogues. Second, we propose the CrowdRank algorithm to select diverse and representative photos for each scenic spot. Finally, according to users requests and particular contexts, we leverage sequential pattern mining and context filtering to generate visual and context-based scenic routes. We conduct experiments over a dataset of 11,542 travelogues and 11,228 travel albums of eight popular scenic spots in China. Extensive experiments show the effectiveness of the proposed framework.