基于群智数据的情境关联旅游路线推荐

Cross-modal Crowd Sourced Data for Context-based Scenic Route Recommendation

针对基于不同出行需求的景区内路线规划问题,首先运用卷积-循环神经网络(CNN-RNN)对游记中图像与文本进行联合嵌入,将数据按照景点进行分类识别,然后使用基于图模型的PhotoRank算法优选出具有多样性、代表性的图片,最后采用关联规则挖掘得到针对不同出行人群的特定需求情境的推荐路线。以8个热门景点为例,对马蜂窝中采集的游记数据进行实验,结果表明提出的基于群智数据的跨模态分析和情境关联旅游路线推荐方法能够从多角度真实地刻画景点,并且所推荐的情境关联路线可满足不同人群的特定游玩需求。

How to plan a route within the scenic spot to satisfy different travel preferences was studied.Firstly,convolution-recurrent neural network(CNN-RNN)was leveraged to classify embed images and texts in travelogues and identify which landscape data were describing.Next,graph-based PhotoRank algorithm was used to select pictures with diversity and representativeness within each landscape.Finally,the association rules were employed to find the recommended routes for different needs of different travel groups.An experiment on seven popular scenic was conducted.The travel data were collected in Mafengwo.The results showed that the cross-modal analysis and context-related travel route recommendation method based on group intelligence data could truly depict the scenic spots from multiple angles,and the recommended context-related route could meet the specific needs of different groups.