基于协同过滤家装方案推荐算法的研究与应用

发布时间：2018-05-05 19:11

本文选题：家装方案 + 用户行为　；参考：《上海交通大学》2015年硕士论文

【摘要】：近年来互联网技术的高速发展以及人们消费水平的提高,我国的家装市场电子商务化将越来越全面,推荐引擎类似于我们实际生活中的产品推荐员的作用,为人们在进行家庭装修过程中选择合适的家装方案提供了极大的便利,同时也为家装网站带来更多的价值。推荐引擎的使用明显加快协同过滤算法的研究和应用,然而在大多数有关协同过滤技术的研究只注重理论的算法层面,忽略了在实际应用中的局限性。因此有必要在深入了解家装网站用户行为和具体实际情况,针对家装方案设计一种适合的推荐算法,为有装修需求的用户提供家装方案推荐服务。本文不仅结合了多种协同过滤算法的思想来设计家装方案推荐算法,而且还对于家装网站家装方案推荐引擎的数据收集和处理过程进行详细阐述。论文的主要工作有以下几个方面:(1)本文通过对家装网站用户行为进行了解之后,研究和分析了协同过滤算法以及Slope One算法的优缺点,并对国内外现有的Slope One算法改进方式进行比较和分析,得出在大数据量和数据相对稀疏的情况下,现有的Slope One算法改进方式将难以获得较高的家装方案推荐服务质量的结论。由此提出一种结合用户相似性和项目相似性对Slope One算法进行改进的家装方案推荐算法。算法主要采用协同过滤中相似性的概念,同时在项目相似性的计算中,结合了家装用户行为评分矩阵中项目相似性以及项目本身相似性两方面因素。(2)通过对实习公司icolor家装网站一个月的用户行为日志进行收集和处理之后,将数据分为训练和测试两个数据集,将新算法与Slope One算法进行比较,实验通过比较算法的MAE值来证明本文设计的家装方案推荐算法具有更好的推荐精准度。考虑到不同相似性度量的影响,实验也对新算法采取不同相似性度量进行比较验证,得出本文采取余弦定理相似度度量更好的改进策略。(3)将本文设计的家装方案推荐算法应用到icolor家装网站中,并对家装网站推荐引擎进行需求分析和架构设计,从用户行为日志收集和运输、离线数据预处理和家装方案算法推荐几部分进行设计和实现,为了避免大量数据计算对网站造成的影响,在数据进行处理和计算的过程中,结合Hadoop、Hive等大数据相关技术以及Mahout技术采取离线的方式进行计算,对涉及的相关数据进行数据库设计,最后对家装网站推荐引擎进行功能测试及界面展示,完成本课题最初期望结果。
[Abstract]:In recent years, with the rapid development of Internet technology and the improvement of people's consumption level, the E-commerce of our home decoration market will be more and more comprehensive, and the recommendation engine will be similar to the role of the product recommender in our real life. It provides great convenience for people to choose suitable home improvement plan in the process of family decoration, and also brings more value to home decoration website. The use of recommendation engine obviously speeds up the research and application of collaborative filtering algorithms. However, most of the researches on collaborative filtering only focus on the theoretical level of algorithms, ignoring the limitations in practical applications. Therefore, it is necessary to deeply understand the user behavior and the actual situation of the home improvement website, and design a suitable recommendation algorithm for the home improvement scheme to provide the home improvement scheme recommendation service for the users with decoration needs. This paper not only combines the idea of various collaborative filtering algorithms to design the home improvement scheme recommendation algorithm, but also describes the data collection and processing process of home improvement website recommendation engine in detail. The main work of this paper is as follows: 1) after understanding the user behavior of home improvement website, this paper studies and analyzes the advantages and disadvantages of collaborative filtering algorithm and Slope One algorithm. By comparing and analyzing the existing improved methods of Slope One algorithm at home and abroad, it is concluded that under the condition of large amount of data and relatively sparse data, it is difficult for the existing improved Slope One algorithm to obtain high quality of service (QoS) of home improvement scheme. This paper proposes a home improvement scheme recommendation algorithm which combines user similarity and item similarity to improve the Slope One algorithm. The algorithm mainly adopts the concept of similarity in collaborative filtering, and at the same time, in the calculation of item similarity, Combining the two factors of item similarity and item itself similarity in the home improvement user behavior score matrix, we collected and processed the user behavior log of icolor Home improvement website for one month. The data is divided into two data sets: training and testing. The new algorithm is compared with the Slope One algorithm, and the MAE value of the algorithm is compared to prove that the recommended algorithm of home improvement scheme designed in this paper has better recommendation accuracy. Considering the influence of different similarity measures, the experiment also compares and verifies that the new algorithm adopts different similarity measures. It is concluded that this paper uses the improved strategy of cosine theorem similarity measure to apply the home improvement scheme recommendation algorithm designed in this paper to the icolor home improvement website, and carries on the requirement analysis and the architecture design to the home improvement website recommendation engine. From the user behavior log collection and transportation, off-line data preprocessing and home improvement scheme algorithm recommended several parts of the design and implementation, in order to avoid the impact of a large amount of data calculation on the website, in the process of data processing and calculation, Combined with big data technology and Mahout technology, the related data are designed, and the function test and interface display of recommendation engine of home decoration website are carried out. Complete the initial expected results of this project.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.3

【相似文献】