基于用户与服务特征的协同过滤推荐研究

发布时间：2018-05-01 14:09

本文选题：Web服务 + 协同过滤　；参考：《山东大学》2014年硕士论文

【摘要】：近年来,互联网技术发展日新月异,Web服务越来越得到重视。为用户选择和推荐最优的Web服务,一直是服务计算领域的核心问题。随着Web服务的种类与数量的增多,服务推荐的难度也会随之增加。在大量的功能相同或相似的Web服务中,根据用户特征与需求,考虑服务的功能性与非功能性属性,为用户推荐Web服务是本文主要研究的内容。目前主要的推荐技术包括基于内容的推荐,基于关联规则的推荐,协同过滤推荐等。协同过滤推荐方法是推荐系统中重要的方法之一,最早出现在B2C的电子商务领域,具有良好的应用和发展前景。商家可以根据用户的偏好与兴趣为用户推荐,推荐其可能喜欢或选择的产品,比如音乐、图书、影像作品等。尽管协同过滤推荐方法在Web服务领域有一定优势,但仍然存在许多问题。新用户问题、新对象问题、稀疏矩阵问题等一直是协同过滤方法研究的热点。随着Web服务数量与用户的增多,问题带来的弊端也更加明显,如何有效的解决这些问题,提高推荐系统的性能是本文研究的重点。在本文中,首先,针对协同过滤技术与Web服务QoS信息特点相结合,提出了根据用户对Web服务QoS信息偏好建立基于用户特征的用户相似度模型。原有的用户相似度计算模型仅考虑用户历史评分,考虑用户历史评分并不能完全表现用户的偏好。比如两个用户共同选择同一个Web服务,其中一个用户可能更关心Web服务的响应时间,而另一个用户看重的是Web服务的安全性。该模型在用户历史评分信息的基础上,深入挖掘用户选择与Web服务QoS信息的关系,为用户进行细分,最终基于用户特征计算用户之问的相似度,并通过实验验证了该模型的实用性。然后,对于协同过滤方法存在的新用户问题以及新对象问题,本文提出建立用户专业度模型。通过用户在某领域的涉及度,以及用户评分的准确度,衡量用户的专业度。在为新用户推荐时,由于缺少新用户的评分信息,采取结合相似用户与专业用户的评分信息为其推荐。专业用户可以较准确的评价新对象,新对象问题也得到一定解决,本文通过实验进行了研究与总结。最后,针对协同过滤方法中的稀疏矩阵问题,分析了稀疏矩阵问题会引起的相似用户数量不足,本文提出了用户相似度传递模型。根据相似用户之间的评分项集,建立了用户信任度模型,在用户信任度模型的基础上,为用户传递相似性,提高相似用户的数量,本文运用实验分析与研究了模型的可行性。
[Abstract]:In recent years, with the rapid development of Internet technology, more and more attention has been paid to Web services. Choosing and recommending the best Web service for users is always the core problem in service computing field. With the increase of Web service types and quantity, the difficulty of service recommendation will increase. In a large number of Web services with the same or similar functions, considering the functional and non-functional attributes of the services according to the characteristics and requirements of users, recommending Web services for users is the main content of this paper. At present, the main recommendation technologies include content based recommendation, association rule based recommendation, collaborative filtering recommendation and so on. Collaborative filtering recommendation method is one of the most important methods in recommendation system. It first appeared in the field of electronic commerce of B2C, and has a good application and development prospect. Merchants can recommend products they may like or choose according to their preferences and interests, such as music, books, video works and so on. Although collaborative filtering recommendation method has some advantages in the field of Web services, there are still many problems. New user problem, new object problem and sparse matrix problem have been the focus of collaborative filtering. With the increase of the number of Web services and users, the disadvantages brought by the problems are more obvious. How to effectively solve these problems and improve the performance of recommendation system is the focus of this paper. In this paper, firstly, a user similarity model based on users' preference for Web services QoS information is proposed according to the combination of collaborative filtering technology and QoS information characteristics of Web services. The original user similarity calculation model only considers the user history score, and the user history score can not completely express the user preference. For example, when two users choose the same Web service together, one user may be more concerned about the response time of the Web service, while the other user is concerned about the security of the Web service. On the basis of user history scoring information, this model deeply excavates the relationship between user selection and Web service QoS information, subdivides users, and calculates the similarity of user questions based on user characteristics. The practicability of the model is verified by experiments. Then, for the new user problem and the new object problem of collaborative filtering method, this paper proposes a user professional model. The user's degree of professionalism is measured by the user's involvement in a field and the accuracy of the user's score. When recommending for new users, because of the lack of rating information of new users, it is recommended by combining the rating information of similar users with that of professional users. Professional users can evaluate new objects more accurately, and the problem of new objects can be solved to some extent. Finally, aiming at the sparse matrix problem in collaborative filtering, the shortage of similar users caused by sparse matrix problem is analyzed, and a user similarity transfer model is proposed in this paper. According to the score item set of similar users, a user trust model is established. On the basis of user trust model, the similarity is transmitted to users and the number of similar users is increased. The feasibility of the model is analyzed and studied by experiments in this paper.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09;TP391.3

【参考文献】