位置感知的协同过滤式Web服务推荐方法研究

发布时间：2018-10-24 06:34

【摘要】：随着Web服务数量的迅速增长,面对海量的Web服务,构建高效的Web服务推荐系统很有必要。为了向用户推荐高质量的服务,关键问题是如何获得Web服务的Qo S值。尽管用户可以通过亲自调用Web服务来评估它的Qo S,但是由于服务的用户并不是评价服务的专家,要在短时间对大量候选服务的Qo S进行准确评估是不太现实的。考虑到Web服务的Qo S值是与具体用户相关的,近年来不少工作利用协同过滤推荐技术来进行个性化的Qo S预测和服务推荐,取得了一定的成效。然而,传统的协同过滤技术在应用时受数据稀疏性的影响较大,且存在冷启动以及可扩展性差等问题。此外,考虑到网络延迟和网络条件,同一个地区的用户有较大可能在相同Web服务上观察到相似的响应时间。针对以往基于协同过滤的Web服务推荐方法的不足,本文提出了一种新的Web服务Qo S预测及推荐方法。本文的主要贡献如下:(1)提出了一种基于位置聚类的协同式Web服务推荐方法,该方法首先利用服务Qo S与用户位置的相关性,将用户根据自治系统(国家)进行聚类,并根据聚类结果对空缺Qo S值进行填充;然后再对空缺Qo S值预先进行填充和计算活动用户与各个用户相似度的基础上,利用To P-K算法,求得最相似来为活动用户预测未知服务的Qo S值,完成推荐。我们的方法能够有效解决Web服务数据稀疏性问题和冷启动问题,同时,在精度和覆盖率之间获得一个更好的平衡。为了更好的验证我们所提出的方法的准确性,我们将该方法在真实的Web服务数据集上进行了一系列全面的实验,结果显示了所提方法的优越性。(2)提出了一种基于因子分解机的质量感知Web服务推荐方法,本文利用Web服务的特点,将用户和服务的网络位置信息和因子分解机相结合,提出了一种位置感知的因子分解机模型及相应的Web服务推荐方法。该方法根据位置信息确定用户和服务的相似邻居集合,然后显式地利用相似用户和相似服务信息改进因子分解机模型,以准确预测未知Web服务的质量和推荐高质量的Web服务。该方法使用了在真实数据集上的实验表明该算法在预测精度上优于其它协同过滤式推荐算法。同时该算法具有较高的运行效率,预测服务质量的时间复杂度与数据规模的大小呈线性相关,可以较好地解决大规模推荐系统的数据稀疏性与可扩展性问题。
[Abstract]:With the rapid growth of Web services, it is necessary to build an efficient Web services recommendation system in the face of massive Web services. In order to recommend high quality service to users, the key problem is how to get the Qo S value of Web service. Although the user can evaluate the Web service by calling it himself, it is not realistic to evaluate the Qo S of a large number of candidate services in a short time because the user of the service is not an expert in evaluating the service. Considering that the Qo S value of Web services is related to specific users, in recent years, a lot of work has made use of collaborative filtering recommendation technology to carry out personalized Qo S prediction and service recommendation, and achieved certain results. However, the traditional collaborative filtering technology is greatly affected by data sparsity in application, and there are some problems such as cold start and poor scalability. In addition, considering network latency and network conditions, users in the same area are more likely to observe similar response times on the same Web service. In view of the shortcomings of the previous Web service recommendation methods based on collaborative filtering, a new Web service Qo S prediction and recommendation method is proposed in this paper. The main contributions of this paper are as follows: (1) A collaborative Web service recommendation method based on location clustering is proposed. Firstly, by using the correlation between service Qo S and user location, users are clustered according to autonomous system (state). According to the clustering result, the vacant Qo S value is filled, and then the vacant Qo S value is filled in beforehand and the similarity between the active user and each user is calculated, then the To P-K algorithm is used. Obtain the most similar to predict the unknown service Qo S value for the active user, complete the recommendation. Our method can effectively solve the problem of Web service data sparsity and cold start, and achieve a better balance between precision and coverage. In order to better verify the accuracy of the proposed method, we conducted a series of comprehensive experiments on the real Web services data set. The results show the superiority of the proposed method. (2) A quality-aware Web service recommendation method based on factorizer is proposed. This paper combines the network location information of user and service with the factoring machine by using the characteristics of Web service. This paper presents a location-aware factoring machine model and a corresponding Web service recommendation method. This method determines the set of similar neighbors of users and services according to location information, and then explicitly uses similar users and similar service information to improve the factoring machine model to accurately predict the quality of unknown Web services and recommend high-quality Web services. Experiments on real data sets show that the proposed algorithm is superior to other collaborative filtering recommendation algorithms in prediction accuracy. At the same time, the algorithm has high running efficiency, and the time complexity of prediction quality of service is linearly related to the size of data, which can solve the problem of data sparsity and scalability in large-scale recommendation systems.
【学位授予单位】：湖南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3;TP393.09

【参考文献】