照片事件浏览中基于文本信息的个性化照片排序

发布时间：2018-01-24 09:06

本文关键词： 基于文本的图像搜索个性化排序照片事件排序局部匹配算法全局排序算法　出处：《上海交通大学》2013年硕士论文　论文类型：学位论文

【摘要】：在许多流行的图像分享网站（如Flickr，Google+）中，用户们可以上传照片并且把同一事件的照片归到一类，作为一个照片事件（event）。浏览者可以通过搜索引擎搜到这些事件，并且这些事件以其中某几张照片的缩略图的形式表现给浏览者，浏览者也通过这为数不多的缩略图来判断是不是这个时间是他们所希望得到的，然后可以更具这一判断选择是否继续浏览剩下的没有被选作缩略图的事件中的别的照片。这个让我们产生了一个想法，那就是我们应该把那些让用户感兴趣的、在事件中比较重要的、和用户搜索关键词、个性化档案关系比较大的照片来作为这个事件的缩略图，这样浏览者就可以很好地把握这个照片事件是否真的符合他们兴趣所在了。在这篇论文里，我们将提出一个基于文本信息的个性化排序算法，来给照片事件中的所有照片进行打分排序，使得a）在照片事件中占有重要地位的照片，b）符合用户个性化喜好文档的照片，，c）符合用户搜索关键词的照片将被给予更高的分数。为了做到这点，我们提出了一个局部匹配模型和一个全局排序模型。局部匹配模型用来根据照片的文字信息给照片打一个初始分数，这个分数将大致的反应照片与用户喜好、搜索关键词的匹配程度以及照片在所属的照片事件中的重要性。全局排序模型将使用局部匹配模型所设定的初始分数，通过迭代得到一个相对于整个照片事件更加平滑的分数。
[Abstract]:In many popular image-sharing sites, such as Flickrn Google, users can upload photos and group photos of the same event. As a photo event, the viewer can search the event through the search engine, and the events are presented to the viewer in the form of a thumbnail of some of the photos. Visitors also use these few thumbnails to determine whether this time is what they want. It is then possible to decide whether or not to continue to browse other photos of the remaining events that were not selected as thumbnails. This has led us to the idea that we should take those that interest users. Photos that are more important in an event, and users search for keywords and personalize files have a large relationship to use as a thumbnail for this event. This gives visitors a good grasp of whether the photo event really suits their interests. In this paper, we propose a personalized sorting algorithm based on text information to sort all photos in a photo event, so that a) plays an important role in the photo event. B) photos that conform to the user's personalized preferences will be given a higher score for photos that match the user's search keywords. In order to achieve this, we propose a local matching model and a global sorting model. The local matching model is used to give an initial score to the photo based on the text information of the photo. This score will roughly reflect the photo and user preferences, match the search keyword and the importance of the photo in the photo event to which it belongs. The global ranking model will use the initial score set by the local matching model. Iterate to get a smoother score relative to the entire photo event.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【相似文献】