轻量级服务推荐算法研究

发布时间：2018-01-10 19:20

本文关键词：轻量级服务推荐算法研究　出处：《北京邮电大学》2015年博士论文　论文类型：学位论文

【摘要】：信息技术的发展给人们的生活带来了巨大的便利。随着网络中信息的大量增加,信息出现了过载现象。为了使用户准确地获取所需的信息,推荐系统应运而生。因为能带来巨大的商业价值和利益,无论是在学术界还是工业界,推荐系统都受到了极大的关注。在学术界,出现了许多高效的推荐方法,在工业界,推荐系统被广泛应用在各种场合。服务推荐是推荐系统的应用实例之一。目前,基于协同过滤的推荐是推荐系统中一种广泛使用的算法。在基于评分的协同过滤推荐过程中,有两个重要的科学问题：一个是用户或者物品(item)之间的相似度计算,另一个是当前用户对当前物品的评分值预测。现有相似度计算方法及评分值预测方法在准确度和效率等方面性能有待进一步提高。本文主要研究基于协同过滤的推荐问题。从相似度计算、评分值预测以及自组织网络中的推荐问题等方面,本文主要做了如下的研究工作： (1)：针对评分的推荐中相似度计算问题,提出了一种基于比值的相似度计算方法。通过比较用户对共同评价过的物品的评分,即可得出用户的相似度。通过比较相同的用户对不同物品的评分即可得出物品的相似度。避免了目前许多相似度计算方法中复杂的运算。实验结果表明,本文提出的相似度计算方法比文中的对比方法更为有效。 (2)：针对未知评分值预测问题,本文在提出的基于比值的相似度计算方法的基础上,提出了一种新的未知评分值预测的方法。该方法只需简单的运算并比较用户评分的大小以及统计数量即可得到预测值。为了评价本文提出的方法的有效性,本文以真实的大规模web服务数据集为基础,与现有的几种主要预测方法进行了对比。实验结果显示,本文提出的方法在误差较小的预测值数量、平均绝对误差(MAE)以及预测时间等方面优于对比方法。 (3)：为了研究移动自组织网络环境中非评分的服务推荐问题,本文提出了一种自组织网络中非评分的服务推荐模型,提出了一种在自组织网络中进行服务推荐的节点之间的相似度计算方法。本文认为节点之间的相似度包含两方面的因素,一方面是移动终端的客观性因素,另一方面是节点所在的用户的主观性因素。本文根据节点的上下文信息,提出了一种节点客观性部分的相似度计算方法；同时,根据用户的行为信息,提出了一种非评分的节点主观性部分的相似度计算方法。最后通过实验研究了自组织网络中各种因素对服务推荐成功率的影响。
[Abstract]:The development of information technology has brought great convenience to people's life. With the increase of network information, information overload. In order to accurately obtain the required information, the recommendation system came into being. Because it can bring huge commercial value and interest, whether in academia or industry, recommended the system has attracted great attention in academic circles, there are many efficient methods recommended, in the industrial sector, the recommendation system is widely used in various occasions. Service recommendation is one of the application examples of the recommendation system.
At present, recommendation based on collaborative filtering is a widely used algorithm in recommendation system. Based on the scores of the collaborative filtering recommendation process, there are two important scientific problems: one is the user or item (item) between similarity computation, another is the current user rating items on the current value of the forecast. The existing similarity calculation method and score prediction method in terms of accuracy and efficiency need to be further improved.
In this paper, we mainly study collaborative filtering recommendation problem. From the aspects of similarity computation, score prediction and recommendation in self organizing network, we have done the following research work.
(1): according to the similarity score of the recommendation of the computational problems, put forward a calculation method based on the similarity ratio. Through the comparison of common user items evaluation score, can be obtained by comparing the similarity of users. The same user of different objects can be obtained. The similarity score items to avoid many similarity in the method of complex computation. Experimental results show that the similarity calculation method than in contrast method is more effective.
(2): for unknown score prediction problem, this paper puts forward the calculation method of the similarity ratio based on the proposed prediction method, a new unknown score. This method only needs simple operation and compared the size and number of users score statistics can get the forecast value. Effective method for this paper presents the evaluation, based on the large-scale web services real data sets as the basis, compared with several existing main prediction methods. Experimental results show that the proposed method value quantity in small prediction error, mean absolute error (MAE) method is better than the contrast and prediction time.
(3): in order to study the mobile ad hoc network environment in the non scoring service recommendation problem, this paper proposes a self organizing network in non score service recommendation model, proposes a calculation method of similarity between node service recommendation in self-organizing network. This paper argues that the similarity between nodes containing factors in two aspects, one is the objective factor of the mobile terminal, on the other hand is a node where the user's subjective factors. Based on the context information of nodes, this paper proposes a calculation method of node objectivity part similarity; at the same time, according to the information of user behavior, proposed a non similarity score subjective part of the node calculation method. Finally, through the experimental study of self organization network in various factors to influence the success rate of service recommendation.

【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TP391.3

【参考文献】