Research On Semantic Learning And Content Recognition Of Cross-media Big Data On Tourism

At present,the rapid development of Internet has brought opportunities and challenges to Intelligent Tourism.The opportunity is that we produce a wealth of cross-media data.And the challenge is the"semantic gap" between the different modes of cross-media data and"semantic gap" hinders the development of Intelligent Tourism.We research on the following three problems in this paper.The first one is Semantic analysis and modeling on the big data of cross-media on tourism.The second one is the automatic image semantic annotation based on the tourism domain ontological knowledge base reasoning.The last one is the crowded tourist scene recognition with the method fusing GIST and micro behavior features.The main work in this paper as follows:(1)In this paper,a symmetrical modeling method based on PLSA theme model was proposed.With this method,we build a latent semantic association model between different modes of cross-media data.By the modeling methods herein,we can build data model contains potential mapping between text words and visual words.(2)We proposed an automatic image semantic annotation algorithm based on tourism domain ontology knowledge base reasoning mechanism.First to annotate images with the annotation algorithm fusing the topic model of text words and visual words based on the model we build above.Then,recognize the content in the image more accurate by reasoning according to the tourism domain ontological knowledge base,so that the words we recognized are more concrete and easier to understand.By using the tourism domain ontology knowledge base,the annotate results are associated with specific attractions,but not the independent and common image content elements.Compared with the traditional asymmetric algorithm,the accuracy of symmetric image annotation algorithm fusing semantic topic has improved greatly.(3)We proposed a crowded tourist scene recognition algorithm that based on fusion of GIST and micro behavior features.First,split out every spot from the video and extract the background image of each shot.Then we can recognize the scene by the scene recognition algorithm based on GIST features.Finally,extract the micro behavior according to the regular pattern of the crowd moving.And then we can recognize the scene of one spot according to micro behavior.With the method,we achieve the purpose to recognize the crowed tourist scene effectively.Compared with the traditional scene recognition algorithm based on the GIST features,the accuracy and recall rate of results have further improved by fusing crowd micro behavior features.(4)We designed and implemented the system of semantic learning and content recognition of cross-media on tourism,including semantic analysis and modeling on the big data of cross-media on tourism,the automatic image semantic annotation based on the tourism domain ontological knowledge base reasoning,the crowded tourist scene recognition with the method fusing GIST and micro behavior features.We can verify the experimental results of each part above by this system.As shown in the result of the experiments,all the three algorithms have better results in semantic learning and content recognition of cross-media data on tourism.