Research On Personalized Retrieval And Cloud Service System For Cross-media Big Data In Tourism Domain

With the rapid development of social network, there is a huge amount of tourism data on the Internet, which leads to the information overload problem. Users need to spend a lot of effort to obtain desired information from huge tourist information, which makes the users'demand for efficient search getting higher and higher. Study on the personalized retrieval and cloud service system for large cross-media tourism data has important theoretical and practical significance. The main work of this thesis is as follows:(1) Aiming at the characteristics of users-sharing tourism photo resources, this thesis presents a multi-feature indexing method for tourism images based on hypergraph. This method uses the hypergraph to construct the relationship between the photos and their additional information (such as shooting time, user tags, etc.) and fuses different features of the photo during the indexing phase, while using the traditional visual vocabulary model to query. This method makes use of the different characteristics of tourism pictures, and avoids the computing time and storage space consumption caused by the fusion in the online query and sorting stage.(2) A hypergraph and random walk based personalized search method for tourism is proposed. Combined with the various characteristics of tourism images, the low-level image features as well as additional information such as text labels and geographical locations are used. The relationship between these feature information is constructed using hypergraph model,and a random walk method is used to query and rank the result. The method allows users providing multiple types of cross-media information such as text labels and images as query examples,and can provides personalized search results based on personalized information provided by the user. Experiments on Internet datasets show that the quality of search result is improved compared with the other image search methods.(3) A distributed visual vocabulary tree training method based on cloud computing and an image search cloud service method based on distributed visual vocabulary tree are proposed. Distributed visual vocabulary tree training method uses distributed K-means algorithm based on MapReduce model, and it can train and query images in parallel.This distributed visual vocabulary tree training method supports the training and querying of a large number of images in distributed memory.Experiments of proposed method show that the training time and memory consumption of each node decrease linearly with the increase of computing units, and the proposed method accelerates the indexing and searching process of tourism image data.(4) Designed and developed a personalized search cloud service system for large-scale cross-media tourism data. The system is divided into multi-feature index module, personalized search module and search cloud service module. It provides users reliable personalized tourism information search cloud service.