基于二部网络分析的推荐算法研究及其应用

发布时间：2018-07-24 13:15

【摘要】：随着网络的高速发展与普及,人们的生活与网络密切相关,网络上的购房租房信息过载成为购置或租赁房屋的用户的一大难题。随着房屋网站和中介网站中的房屋出租、二手房出售等信息越来越多,部分信息得不到更新,使得用户很难高效地找到适合自己的房屋信息。相关网站面临着的重要问题就是怎样实时并且准确地向用户提供其感兴趣的房屋,而解决这一问题的有效方法便是个性化推荐系统。本文作者参加了 365淘房网房屋购买与租赁推荐系统的部分开发工作,根据项目的需要,从二部网络分析的角度研究了对房屋的个性化推荐方法,并设计与开发了相应的房屋推荐系统,主要研究工作和成果如下:(1)根据房屋推荐系统的数据的特点,本文提出了基于二部网络社区挖掘的推荐算法。用带权的二部图来表达用户-项目的评分矩阵。我们提出了二部网络社区挖掘的标号传递的算法,先将用户划分社区,再将需要推荐的用户划分到最相关的社区,然后利用用户间的相似性进行推荐。算法综合考虑了用户与社区的关系,以及用户之间、项目之间的相似性,找出用户有潜在兴趣的项目。实验结果说明该算法的推荐结果比较其他相类似的推荐算法具有较高的精度。(2)提出了将二部网络链接预测中的Jaccard指标与奇异值分解相结合的混合推荐算法。算法先用Jaccard算法求出其推荐房源的相似度指标矩阵,然后我们用奇异值分解算法补齐Jaccard指标矩阵的零元素,从而得到一张完整的推荐列表。我们提出了对Jaccard指标与奇异值分解动态增量更新的算法,以适应评分表的动态变化。实验结果说明,该算法比较其他相关的推荐算法具有较高的推荐准确率和召回率。(3)设计与开发了 365淘房网房屋购买与租赁的推荐系统,在该系统中应用了本文所提出的基于社区挖掘的推荐算法和Jaccard指标与奇异值分解相结合的混合推荐算法。对房屋购买与租赁推荐系统进行了需求分析,提出系统框架的总体设计方案,介绍了系统的模块结构,以及各个模块的实现。
[Abstract]:With the rapid development and popularization of the network, people's life is closely related to the network. The overloading of housing information over the network has become a big problem for the users of the purchase or rental housing. With the housing site and the housing rental in the intermediary websites, the second-hand housing is more and more information and some information can not be updated, making it difficult for the users. The important problem facing the relevant websites is how to provide the real and accurate housing to the users in real time and accurately, and the effective way to solve this problem is the personalized recommendation system. The author participated in the part of the development workers of the 365 house network housing purchase and rental recommendation system. According to the needs of the project, this paper studies the personalized recommendation method of the house from the angle of two network analysis, and designs and develops the corresponding house recommendation system. The main research work and results are as follows: (1) according to the characteristics of the data of the house recommendation system, this paper proposes a recommendation algorithm based on the two network community mining. The two graph is used to express the rating matrix of user project. We put forward two network community mining labeling algorithms. First, we divide the users into the community, then divide the recommended users into the most relevant communities, and then use the similarity between users to recommend them. The algorithm fully considers the relationship between the users and the community, and the users. Between the similarity between the projects and identifying the potential interests of the user, the experimental results show that the proposed results of the algorithm have a higher accuracy compared with other similar recommendation algorithms. (2) a hybrid recommendation algorithm which combines the Jaccard index with the singular value decomposition in the two network link prediction is proposed. The algorithm first uses the algorithm to calculate the algorithm. The similarity index matrix of the recommended house is obtained by the method. Then we use the singular value decomposition algorithm to complement the zero element of the Jaccard index matrix and get a complete recommendation list. We propose an algorithm to update the dynamic increment of the Jaccard index and the singular value decomposition to adapt to the dynamic change of the score table. Compared with other relevant recommendation algorithms, it has high recommendation accuracy and recall rate. (3) a recommendation system for the purchase and lease of 365 house net houses is designed and developed. In this system, a recommendation algorithm based on community mining and a hybrid recommendation algorithm combined with Jaccard index and singular value decomposition are applied. The demand analysis of the purchase and leasing recommendation system is carried out, and the overall design scheme of the system framework is presented. The module structure of the system and the realization of each module are introduced.
【学位授予单位】：扬州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】