基于改进的三维马尔可夫模型推荐系统的研究与实现

发布时间：2018-01-19 20:16

本文关键词： 临时兴趣三维马尔科夫模型混合推荐兴趣阈值　出处：《天津财经大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着大数据时代的到来,网络数据正在以爆炸式的速度出现在人们生活当中。丰富的数据资源在给人们带来便利的同时也造成了很大的困扰,如何从庞大的数据集中找到期望的数据成为了亟待解决的热门问题。因此,数据挖掘技术逐渐被人们重视起来。推荐算法是数据挖掘技术的一个非常重要的应用领域。一个高效并精准的个性化推荐方法不仅能够减少用户筛选无用信息的麻烦,同时也避免了大量的数据传输造成的网络成本消耗。具有很高的商业价值。目前,个性化推荐系统主要在新闻浏览,娱乐信息浏览,电子商务等领域得到了广泛的应用。基于此,本文以用户能够更高效更精准的浏览新闻信息的目的出发,提出了一种以引入了兴趣阈值参数的三维马尔科夫模型推荐算法进行精确预测并以基于加权用户行为数据归一化的推荐算法进行辅助预测的推荐策略。在算法的研究过程中,首先分析了新闻信息区别其他信息所具有的时效性强、数量大的特性。在数据整合的过程中,在排除了特殊情况的前提下,将信息的属性进行了模糊替换以便提取出符合该特性的用户偏好特征。其次分析了用户临时兴趣现象和用户兴趣陡峭式变化的现象带来的预测不准的问题,并提出在用户偏好特征中引入兴趣阈值参数的方式将用户的兴趣变化及时反馈给推荐系统的解决办法。再次依据该特性和问题对要实现的新闻推荐算法进行需求分析。为了满足对算法的需求分析,将Apriori算法以及马尔科夫模型等算法的特性以及适用性进行分析,最终确定了本文的推荐模型,并通过在模型中引入试探性推送重大主流新闻以及用户主动反馈的机制调整兴趣阈值的取值,进而改善了用户产生临时兴趣现象和用户兴趣陡峭式变化的现象带来的预测不准的问题。然后通过对不同参与度用户分组实验,对比了未引入兴趣阈值的传统推荐模型与本文提出的推荐模型在推荐结果有效性上的差别。验证了本文提出的推荐策略的有效性。最终应用本文提出的推荐策略实现了一款具有个性化推荐功能的iOS端的手机应用。
[Abstract]:With the arrival of big data era, network data is appearing in people's lives at an explosive speed. Rich data resources not only bring convenience to people, but also cause a lot of trouble. How to find the expected data from the huge data set has become a hot problem to be solved. Data mining technology has been paid more and more attention. Recommendation algorithm is a very important application field of data mining technology. A highly efficient and accurate personalized recommendation method can not only reduce user filtering useless information. Trouble. At the same time, it also avoids the network cost consumption caused by a large number of data transmission. It has high commercial value. At present, personalized recommendation system mainly in news browsing, entertainment information browsing. Electronic commerce and other fields have been widely used. Based on this, the purpose of this paper is to enable users to browse news information more efficiently and accurately. In this paper, we propose a recommendation strategy based on 3D Markov model recommendation algorithm with interest threshold parameters and a recommendation algorithm based on weighted user behavior data normalization. In the process of research. First of all, the paper analyzes the characteristics of news information in distinguishing other information with strong timeliness and large quantity. In the process of data integration, the special circumstances are excluded. The attribute of information is replaced by fuzzy to extract the characteristic of user preference. Secondly, the problem of uncertain prediction caused by the phenomenon of temporary interest and steep change of interest is analyzed. The method of introducing the threshold parameter of interest into the feature of user preference is put forward to feedback the change of user's interest to the recommendation system in time. According to this feature and problem, the news recommendation algorithm to be implemented is required again. Analysis. In order to meet the needs of the algorithm analysis. The characteristics and applicability of Apriori algorithm and Markov model are analyzed, and the recommendation model of this paper is finally determined. And through the introduction of exploratory push major mainstream news and user active feedback mechanism to adjust the threshold of interest in the model. Then it improves the problem of uncertain prediction caused by the phenomenon of user's temporary interest and the steep change of user's interest. Then the user grouping experiment with different participation degree is carried out. The difference between the traditional recommendation model without interest threshold and the recommendation model proposed in this paper is compared in this paper. The validity of the recommendation strategy proposed in this paper is verified. Finally, the recommendation strategy proposed in this paper is applied. Realized. A personalized recommendation of the iOS end of the mobile application.
【学位授予单位】：天津财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】