基于LM-BP神经网络的推荐算法的研究与应用

发布时间：2018-03-03 09:56

本文选题：LM-BP神经网络　切入点：信息熵　出处：《北京交通大学》2016年硕士论文　论文类型：学位论文

【摘要】：近年来,随着互联网技术的发展,网络数据越来越庞大,用户怎样从众多的数据中更快速的捕捉到自己感兴趣的数据信息成为网络技术发展的研究热点。学术界和业界对这种信息过载问题开展了大量的研究和实践工作,提出了多种形式的信息个性化解决方案。推荐系统作为一种智能个性化信息服务系统,具有用户需求驱动、主动服务和信息个性化程度高等优点,在电子商务、在线学习和数字图书馆等领域得到了广泛应用,并已成为公认最有前途的信息个性化技术发展方向。协同过滤推荐算法是推荐系统中最成功的技术之一,虽然已经被应用超过了十年,但是由于商品量越来越多,用户评价和购买的数量有限,数据稀疏性问题越来越严重,另外还存在冷启动等问题,导致协同过滤推荐算法的准确性还有待提高。针对数据稀疏性问题,本文设计了新的协同过滤算法,根据用户评分交集的大小选择用户最近邻居集,采用LM-BP神经网络对用户一项目的评分矩阵进行估值填充,提高评分矩阵的密度。这种方法避免了传统降维法导致信息缺失的缺点,能提高预测值的准确度,从而提高协同过滤推荐系统的推荐质量。相似度的计算是协同过滤的一个重要步骤,传统的计算方法容易夸大或缩小相似性,从而影响推荐质量。本文采用信息熵的方法来计算相似度,通过计算评分差值的信息熵,并将用户评分差异和交叠程度加权到公式中,提高相似度计算的准确度,从而提高协同过滤推荐算法的推荐效果。最后,本文采用了Movielens的真实数据集进行实践研究,用Matlab对本文提出的基于LM-BP神经网络的推荐算法进行测试。从准确度、平均绝对误差、召回率、F1指标四个方面,相似度求法、和估值填充两个角度,对本文提出的推荐算法的效果进行了研究。数据结果表明基于LM-BP神经网络的协同过滤推荐算法比传统的协同过滤推荐算法有更好的推荐效果。
[Abstract]:In recent years, with the development of Internet technology, the network data become more and more huge. How users can quickly capture the information of their interest from a large number of data has become a research hotspot in the development of network technology. Academic circles and industry have carried out a lot of research and practice on this kind of information overload. As an intelligent personalized information service system, recommendation system has the advantages of user demand driven, active service and high degree of information personalization. Online learning and digital library have been widely used, and have become the most promising development direction of personalized information technology. Collaborative filtering recommendation algorithm is one of the most successful technologies in recommendation system. Although it has been used for more than a decade, due to the increasing volume of commodities, the limited number of user evaluations and purchases, the problem of data sparsity is becoming more and more serious, and there are also problems such as cold startup. The accuracy of collaborative filtering recommendation algorithm still needs to be improved. Aiming at the problem of data sparsity, this paper designs a new collaborative filtering algorithm, which selects the nearest neighbor set according to the size of the user score intersection. The LM-BP neural network is used to populate the evaluation matrix of a user item to increase the density of the scoring matrix. This method avoids the shortcoming of the traditional dimensionality reduction method which leads to the lack of information, and can improve the accuracy of the prediction value. In order to improve the recommendation quality of collaborative filtering recommendation system, the calculation of similarity is an important step in collaborative filtering, and the traditional calculation method is easy to exaggerate or reduce the similarity. In this paper, the information entropy method is used to calculate the similarity, and the information entropy of the score difference is calculated, and the user score difference and overlapping degree are weighted to the formula to improve the accuracy of the similarity calculation. In order to improve the recommendation effect of collaborative filtering recommendation algorithm. Finally, this paper uses the real data set of Movielens for practical research, and uses Matlab to test the recommendation algorithm based on LM-BP neural network. Recall rate and F1 index four aspects, similarity method, and valuation filling two angles, The results show that the collaborative filtering recommendation algorithm based on LM-BP neural network has better recommendation effect than the traditional collaborative filtering recommendation algorithm.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3;TP183

【相似文献】