基于攻击用户识别和贝叶斯概率矩阵分解的鲁棒推荐算法

发布时间：2018-08-14 12:40

【摘要】：随着大数据时代的到来和电子商务的发展,协同过滤推荐系统以其个性化的推荐优势正逐渐渗透人们的生活。但是由于推荐系统的开放性,容易招致恶意用户向推荐系统中注入攻击概貌以改变推荐结果,这严重影响了推荐系统的安全性。本文针对这个问题提出了一种鲁棒推荐算法,用于提高推荐系统的鲁棒性,同时保证推荐精准度。具体研究内容如下。首先,针对推荐系统中恶意攻击用户的存在影响推荐系统鲁棒性的问题,提出了一种嫌疑用户聚类算法,该算法从项目流行度入手,引入了用户平均评分流行度的概念,并基于此概念重新定义了用户之间距离的计算公式。该算法目的是将嫌疑用户聚集到一类,方便攻击用户的识别。其次,针对嫌疑用户聚类算法中存在的真实用户误判问题,由于真实用户误判会影响推荐系统的推荐精准度,所以提出了一种基于嫌疑用户聚类和目标项目识别的攻击用户识别方法,在嫌疑攻击类中进一步准确识别并标记攻击用户。该算法首先识别目标项目,然后在嫌疑攻击类中识别并标记攻击用户,目的是减小真实用户误判率,保障推荐系统的推荐精准度。然后,针对推荐算法低鲁棒性的问题,将攻击用户识别标记结果与贝叶斯概率矩阵分解模型结合,形成基于攻击用户识别和贝叶斯概率矩阵分解的鲁棒推荐算法。该算法在模型的学习过程中屏蔽被标记的攻击用户对目标项目的评分,目的是在保证推荐精准度的同时提高推荐系统的鲁棒性。最后利用MovieLens 100K数据集在Mat Lab实验平台上模拟仿真实验,并与一些经典鲁棒推荐算法作对比分析。实验结果表明,本文算法可以在提高推荐鲁棒性的同时保证推荐精准度。
[Abstract]:With the arrival of big data era and the development of electronic commerce, collaborative filtering recommendation system is gradually infiltrating people's life with its personalized recommendation advantage. However, due to the openness of the recommendation system, it is easy to cause malicious users to inject an attack profile into the recommendation system to change the recommendation result, which seriously affects the security of the recommendation system. In this paper, a robust recommendation algorithm is proposed to improve the robustness of recommendation system and ensure the accuracy of recommendation. The specific contents of the study are as follows. First of all, aiming at the problem that malicious attacks on users in recommendation system affect the robustness of recommendation system, a clustering algorithm for suspected users is proposed. The concept of average user rating popularity is introduced in this algorithm based on item popularity. Based on this concept, the formula for calculating the distance between users is redefined. The purpose of this algorithm is to cluster suspected users into a class and to facilitate the identification of attacking users. Secondly, aiming at the real user misjudgment problem in the suspect user clustering algorithm, the recommendation accuracy of the recommendation system will be affected by the real user misjudgment. Therefore, an attack user identification method based on suspect user clustering and target item identification is proposed, which can further accurately identify and mark the attack user in the suspect attack class. The algorithm first identifies the target item and then identifies and marks the target user in the suspect attack class. The purpose of the algorithm is to reduce the false judgment rate of the real user and to ensure the recommendation accuracy of the recommendation system. Then, aiming at the problem of low robustness of recommendation algorithm, a robust recommendation algorithm based on attack user identification and Bayesian probability matrix decomposition is formed by combining the result of attack user identification with Bayesian probability matrix decomposition model. In order to improve the robustness of the recommendation system, the algorithm blocks the target item score of the target user during the learning process of the model. The goal of the algorithm is to ensure the accuracy of the recommendation and improve the robustness of the recommendation system. Finally, the MovieLens 100K data set is used to simulate the simulation experiment on the Mat Lab platform, and it is compared with some classical robust recommendation algorithms. The experimental results show that the proposed algorithm can improve the robustness of recommendation and ensure the accuracy of recommendation.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】