【摘要】：隨著社交網(wǎng)絡(luò)的快速發(fā)展,互聯(lián)網(wǎng)上產(chǎn)生了海量的旅游數(shù)據(jù),導(dǎo)致信息過載問題的出現(xiàn)。用戶從中獲取有效信息需要花費(fèi)很多精力,這使得用戶對旅游信息高效搜索的需求越來越高。研究面向跨媒體旅游大數(shù)據(jù)的個(gè)性化搜索與云服務(wù)系統(tǒng)具有重要的理論和應(yīng)用意義。本文完成的主要工作如下:(1)針對旅游領(lǐng)域中用戶分享照片資源的特點(diǎn),提出了一種基于超圖的隨機(jī)游走旅游圖片索引方法。這個(gè)方法利用超圖來建立旅游圖片和其附加信息(例如拍攝時(shí)間、用戶標(biāo)簽等)間的關(guān)系,并在圖片索引階段對圖片的不同特征進(jìn)行融合,而在查詢時(shí)使用傳統(tǒng)的視覺詞匯模型進(jìn)行搜索。這個(gè)方法綜合利用了旅游圖片的不同特征,并且避免了在查詢階段和排序階段進(jìn)行融合所帶來的計(jì)算時(shí)間和存儲空間消耗,提供了一種更加全面且高效的圖片索引方法。(2)提出了基于超圖隨機(jī)游走的個(gè)性化旅游信息搜索方法,結(jié)合旅游圖片的特征,綜合利用了圖片本身的底層圖像特征以及圖片的標(biāo)簽、地理位置等附加信息,使用超圖的方法構(gòu)造這些特征信息之間的關(guān)系,使用隨機(jī)游走的方法在超圖模型上進(jìn)行搜索并排序。本方法允許用戶提供文本標(biāo)簽和圖像等多種類型的跨媒體信息作為搜索樣例,并能根據(jù)用戶提供的個(gè)性化信息為用戶提供個(gè)性化的搜索結(jié)果。在互聯(lián)網(wǎng)數(shù)據(jù)集上的實(shí)驗(yàn)表明,與使用單一特征的通用圖片搜索方法相比,本方法的搜索結(jié)果質(zhì)量有所提升。(3)針對旅游圖片大數(shù)據(jù)搜索時(shí)數(shù)據(jù)龐大且實(shí)時(shí)更新的問題,提出了基于云計(jì)算的分布式視覺詞匯樹訓(xùn)練方法和基于分布式視覺詞匯樹的圖像搜索云服務(wù)方法。分布式視覺詞匯樹訓(xùn)練方法基于MapReduce模型的分布式K-means算法,用于并行地訓(xùn)練圖像并進(jìn)行檢索,這種分布式視覺詞匯樹訓(xùn)練方法可以支持在內(nèi)存中訓(xùn)練大量的圖像。實(shí)驗(yàn)結(jié)果表明當(dāng)計(jì)算單元增加時(shí)每個(gè)節(jié)點(diǎn)的訓(xùn)練時(shí)間和內(nèi)存消耗呈線性減少趨勢,加快了跨媒體索引的建立和搜索過程。(4)設(shè)計(jì)和開發(fā)了面向跨媒體旅游大數(shù)據(jù)的個(gè)性化搜索云服務(wù)系統(tǒng)。該系統(tǒng)分為多特征索引模塊、個(gè)性化搜索模塊與搜索云服務(wù)模塊,可為用戶提供可靠的個(gè)性化旅游數(shù)據(jù)搜索云服務(wù)。
[Abstract]:With the rapid development of social networks, mass travel data are produced on the Internet, which leads to the problem of information overload. It takes a lot of effort for users to obtain effective information, which makes the demand for efficient search of tourism information more and more high. It is of great theoretical and practical significance to study the personalized search and cloud service system for cross-media tourism big data. The main work of this paper is as follows: (1) according to the characteristics of users sharing photo resources in the field of tourism, a hypergraph-based random walk travel image index method is proposed. This method uses hypergraphs to establish the relationship between tourist pictures and their additional information (such as shooting time, user tags, etc.), and fuses the different features of the pictures in the image index stage. The traditional visual lexical model is used to search the query. This method synthesizes the different features of tourist pictures, and avoids the computational time and storage space consumption caused by fusion in query stage and sorting stage. A more comprehensive and efficient method of image indexing is provided. (2) A method of personalized travel information search based on hypergraph random walk is proposed, which combines the features of tourism images. Using the underlying image features of the image itself and the additional information of the image label, geographical location, etc., using the method of hypergraph to construct the relationship between these feature information, A random walk method is used to search and sort hypergraph models. The method allows users to provide multiple types of cross-media information, such as text labels and images, as search samples, and can provide personalized search results for users according to personalized information provided by users. Experiments on Internet dataset show that compared with the general image search method with a single feature, the quality of search results of this method is improved. (3) aiming at the problem of huge data and real-time updating when big data searches tourist images, The training method of distributed visual vocabulary tree based on cloud computing and the method of image searching cloud service based on distributed visual vocabulary tree are proposed. The distributed visual vocabulary tree training method is based on the distributed K-means algorithm of MapReduce model, which is used to train and retrieve images in parallel. This distributed visual vocabulary tree training method can support the training of a large number of images in memory. The experimental results show that the training time and memory consumption of each node decrease linearly when the computing unit increases. It speeds up the establishment and search process of cross-media index. (4) A personalized search cloud service system for cross-media tourism big data is designed and developed. The system is divided into multi-feature index module, personalized search module and search cloud service module, which can provide users with reliable personalized travel data search cloud services.
【學(xué)位授予單位】：北京郵電大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.3;TP393.09

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 左欣;沈繼鋒;于化龍;高尚;徐丹;胡春龍;;基于哈希編碼學(xué)習(xí)的圖像檢索方法[J];江蘇科技大學(xué)學(xué)報(bào)(自然科學(xué)版);2015年06期

2 宋天勇;趙輝;鄭山紅;王國春;;基于查詢-概念的用戶興趣模型構(gòu)建[J];吉林大學(xué)學(xué)報(bào)(信息科學(xué)版);2015年03期

3 李武軍;周志華;;大數(shù)據(jù)哈希學(xué)習(xí):現(xiàn)狀與趨勢[J];科學(xué)通報(bào);2015年Z1期

4 任樹懷;;LUCENE搜索算法剖析及優(yōu)化研究[J];圖書館雜志;2014年12期

5 楊昭;高雋;謝昭;吳克偉;;局部Gist特征匹配核的場景分類[J];中國圖象圖形學(xué)報(bào);2013年03期

6 徐磊;;基于內(nèi)容的大規(guī)模圖像檢索基本方法[J];科技信息;2013年08期

相關(guān)博士學(xué)位論文 前3條

1 蔣鍇;含地理位置信息的社交媒體挖掘及應(yīng)用[D];中國科學(xué)技術(shù)大學(xué);2014年

2 戴金波;基于視覺信息的圖像特征提取算法研究[D];吉林大學(xué);2013年

3 尹華罡;基于海量時(shí)空數(shù)據(jù)的路線挖掘與檢索[D];中國科學(xué)技術(shù)大學(xué);2012年

相關(guān)碩士學(xué)位論文 前3條

1 郭劍飛;基于LDA多模型中文短文本主題分類體系構(gòu)建與分類[D];哈爾濱工業(yè)大學(xué);2014年

2 葉君峰;基于圖像的多樣化景點(diǎn)搜索[D];上海交通大學(xué);2013年

3 李雪;旅游個(gè)性化搜索系統(tǒng)的研究與實(shí)現(xiàn)[D];北京郵電大學(xué);2013年

，

本文編號：2429626