基于用户兴趣向量的混合推荐算法

发布时间：2018-04-02 04:15

本文选题：推荐系统　切入点：数据稀疏性　出处：《山东大学》2015年硕士论文

【摘要】：随着网络信息化新格局的出现,人们在互联网中的角色逐渐发生了变化。方面,作为信息浏览者,可以利用更加丰富的网络资源满足自己的需求。另一方面,作为信息制造者,人们正在习惯将生活中的点点滴滴上传到互联网,同时以史无前例的速度继续生产内容。这种海量信息的呈现使得用户无所适从,想要从中挑出真正吻合用户兴趣的内容非常困难,这就出现了信息过载现象。所以,当下信息过载问题的解决变得日益迫切。推荐系统是解决信息过载问题的关键技术之一,成为了无数学者追逐研究的热点。推荐系统通过获取服务器中用户的行为日志,得到可以描述用户兴趣的原始数据,进而构建用户的兴趣模型,通过相似度分析计算,为用户呈现更加个性化的浏览页面,从而提高用户的浏览效率和使用感受。推荐系统不仅仅是一个热门的理论研究方向,而且作为一种有效的营销手段已经广泛应用于互联网。然而,面对越来越复杂多样的应用场景,推荐系统暴露出了若干问题,如：数据稀疏性问题、用户兴趣迁移问题等。本文针对现有技术存在的问题,研究了电影推荐中的推荐算法,同时研究了基于推荐算法的医疗冷柜存储策略,提出有效的解决方案。主要内容如下：(1)以电影推荐为应用背景,提出了一种基于用户兴趣向量的混合电影推荐算法。众所周知,基于协同过滤的推荐算法对于用户的兴趣变化不敏感,同时数据稀疏性问题也制约了该算法的发展。针对这两个问题,提出了一种新型的基于用户兴趣向量的混合电影推荐算法。①为了解决数据稀疏性问题,本文引入了用户混合兴趣向量。从电影特征向量入手,借助用户的评分矩阵以迭代的方式处理得到用户的兴趣特征向量,根据得到的用户兴趣向量和用户的评分信息组成用户混合兴趣向量,进而构建用户相似矩阵,最终根据传统的协同过滤评分方式完成推荐。②针对用户兴趣变化的情况,在构建用户兴趣向量过程中融入时间因子,使得越接近当前时间的评分行为权重越大,越能反映出用户的当前兴趣。本文在Movielens数据集上进行了实验,并与现有的相关算法进行了性能比较。实验结果表明本文算法在预测评分准确性和收敛性上都有明显的提高。(2)以医疗冷柜为应用背景,提出一种基于用户行为的智能医疗冷柜系统中样品的智能存取策略。该策略在智能医疗冷柜的自动化提取过程中加入样品推荐模块,增强用户与冷柜系统的交互能力,提升用户的工作效率。具体的,该策略重点解决以下技术问题：①如何利用丰富的样品内容信息辅助存储。②如何根据用户的存取行为构建有效的提取策略。通过实时收集用户的存储和提取行为建立用户的行为数据,结合样品本身特征属性和用户行为的数据分析,建立样品之间的关联度矩阵,从而针对待存储样品给出合理的存储位置(问题①)。同时,在用户提取样品阶段提出相应的推荐,从而提升用户的使用体验(问题②)。
[Abstract]:With the advent of the new pattern of the information network, on their role in the Internet has gradually changed. As information, visitors can use more cyber source to meet their own needs. On the other hand, as information producers, people are accustomed to the life of the little Didi uploaded to the Internet, at the same time the There was no parallel in history. continues to produce content. This information allows users to pick out the real situation, consistent with the user interested in the content is very difficult, it appeared the phenomenon of information overload. Therefore, when solving the problem of information overload has become increasingly urgent. Recommendation system is one of the key technologies to solve the problem of information overload, has become a focus of countless the scholars chase research. The recommendation system get the server user behavior log, the original data can be obtained to describe user interest, Then construct the user interest model, through the analysis of similarity calculation, the user presents a more personalized browsing page, so as to improve the user's browsing efficiency and experience. The recommendation system is not only a theoretical study on the hot direction, but also as an effective marketing tool has been widely used in the Internet. However, in the face of more and more application scenarios more complex, the recommendation system exposes some problems, such as: data sparsity, user interest migration problems. Aiming at the problems of existing technology, on the recommendation of the movie recommendation algorithm, and studies the medical freezer storage strategy based on recommendation algorithm, propose effective solutions. The main contents are as follows: (1) the movie recommendation for the application background, proposes a hybrid recommendation algorithm based on user interest vector movie. As everyone knows, based on collaborative filtering The filter recommendation algorithm is not sensitive to the user's interest, while the data sparsity problem also restricts the development of the algorithm. To solve these two problems, proposes a new hybrid film recommendation algorithm based on user interest vector. In order to solve the problem of data sparsity, this paper introduces the user interest vector of mixed. From the movie feature vector, get the user interest feature vectors in an iterative manner using score matrix of the user, based on user interest vector mixed user interest vector and users get the score information, then construct the user similarity matrix, according to the final score of traditional collaborative filtering recommendation. For the completion of the change of user interest. Into the time factor in the process of constructing the user interest vector, the score closer to the current time behavior weight is bigger, more can reflect the user's The current interest. We carried out experiments on Movielens data sets, and comparisons with existing algorithms. The experimental results show that this algorithm has obvious improvement in predicting accuracy and convergence. (2) to the medical refrigerator as the application background, the proposed intelligent access strategy of a sample of intelligent medical refrigerator system based on the user behavior in the process of adding samples. The strategy recommendation module in automatic intelligent medical freezer extraction, enhance the interaction ability of user and the refrigerator system, to enhance the user's work efficiency. In particular, the strategy to solve the following technical problems: how to use the sample content information auxiliary storage rich. How to construct extraction effective strategies according to the user's access behavior. The establishment of user behavior data collected by users in real time storage and extraction behavior, combined with the characteristics of the sample itself Data analysis of sex and user behavior, and establishment of correlation matrix between samples, so as to provide reasonable storage location for storage samples (problem 1). At the same time, recommend corresponding recommendation in user sampling stage, so as to enhance user experience.

【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【参考文献】