基于网站式的图片过滤推荐系统研究

发布时间：2018-03-07 11:52

本文选题：图片推荐　切入点：标签　出处：《天津大学》2016年硕士论文　论文类型：学位论文

【摘要】：在网络时代的今天,以互联网为传播媒介的信息如井喷式增长,怎么从这样大数量级的数据中搜索出有效信息的难点随之产生,个性化推荐的出现就是为了解决这一难点。不容易直接提的内容主要为多媒体资源,它主要包括图片,音视频等,应用标签可以很好解决这一问题,标签系统作为个性化推荐系统的基础应运而生。在我们搜索有效信息的同时,常常需要主观去输入关键词和联想描述,但是用户有时很难精确的把握到自己目标词汇的关键词,常常搜索出来的结果与自己的预想相差甚远,而个性化推荐不需要用户主动去输入什么,便可以达到智能推荐的目的。标签系统可以根据不同用户所提供的标签互相关联,从而达到推荐的目的。简单来说,标签就是组成多媒体资源特点的不同元素,一个信息有不同的元素,而这些元素都是用户赋予的,所以由算法计算这些元素之间的关联性,从而推断资源乃至用户的爱好,由此便可实现智能化推荐的目标。为了是用户能迅速找到对自己有用的有效图片,本文对以协同过滤及标签为基础的图片推荐系统进行讨论。一开始要介绍的是推荐标签的方法,对相关图片进行标注,这时候标签就成为了该图片的基础元素,对图片的所有标签元素出现频率进行排列,以此作为标签推荐的依据,称之为关联度。在此基础上,对图片的所有标签元素进行求和计算,以此来完成图片的推荐。而以系统过滤为基础的推荐方法,计算的则是另一个相关元素,用户评分,对评分进行矩阵计算,以此构建用户相似度模型。
[Abstract]:In the age of network, how to search for effective information from such a large order of magnitude data with the Internet as the medium of communication, such as blowout growth, comes into being with it. The appearance of personalized recommendation is to solve this difficulty. The main content that is not easy to mention directly is multimedia resources. It mainly includes pictures, audio and video, etc. The application of tags can solve this problem very well. Tag system as the basis of personalized recommendation system arises at the historic moment. When we search for effective information, we often need to input keyword and associative description subjectively, but sometimes it is difficult for users to accurately grasp the keywords of their target words. The results are often far from their expectations, and personalized recommendation does not require the user to enter anything to achieve the purpose of intelligent recommendation. The tag system can be based on the tags provided by different users to relate to each other. In simple terms, tags are different elements that make up the characteristics of multimedia resources, and a message has different elements, and these elements are all given by the user, so the algorithm calculates the correlation between these elements. The goal of intelligent recommendation can be achieved by inferring resources and even users' hobbies, so that users can quickly find valid pictures that are useful to them. In this paper, we discuss the image recommendation system based on collaborative filtering and label. At the beginning, we want to introduce the method of recommending label, and label the relevant picture, then label becomes the basic element of the picture. The frequency of all tag elements in the picture is arranged as the basis of the label recommendation, which is called the correlation degree. On this basis, the sum of all the tag elements of the picture is calculated. The recommendation method based on system filtering is used to calculate another related element, the user score, the matrix calculation of the score, so as to construct the user similarity model.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】