基于兴趣漂移协同过滤算法的推荐系统研究与实现

发布时间：2018-05-19 07:17

本文选题：显式兴趣漂移模型 + 隐式兴趣遗忘模型　；参考：《西北农林科技大学》2017年硕士论文

【摘要】：随着科技的快速发展,互联网中的数据正在进爆炸性的增长。当下如何在庞大的数据中,准确并迅速的找到用户所需要的有用信息,成为研究推荐领域的焦点。用户获得有价值信息的两个基本的途径是用户的手动的搜索和系统的个性化推荐,在对于某些领域,需要比较智能性方式为用户推荐想要的信息。本文主要研究内容如下:(1)提出基于显式与隐式兴趣漂移检测模型,并进行兴趣漂移。由于用户的兴趣会随着时间推移而改变,因此,通过显式兴趣漂移检测模型对其检测,并利用隐式进行兴趣修正。本文提出五种兴趣模型,利用其相关属性值及关联密集度增量对其进行模式匹配,得到用户兴趣序列。对该序列利用指数遗忘函数进行隐式兴趣修正,进而得到用户真正的兴趣。实验结果表明,显式兴趣漂模型对其用户兴趣的判别准确率及召回率达到了88.23%~90.96%和87.43%~89.54%,能够准确地捕获到用户当前的兴趣。(2)设计了基于显式与隐式兴趣漂移检测模型的协同过滤算法。针对协同过滤算法推荐准确度低的问题,采用属性相似度与评分相似度加权得到新的相似度,利用基于项目的协同过滤算法,通过显隐式兴趣漂移检测模型得到用户当前兴趣,并进行相关项目的推荐。实验结果表明,本文算法与传统的协同过滤算法相比较,平均绝对偏差MAE值降低了3.41%,说明了与传统协同过滤算法相比,对用户兴趣的检测以及混合相似度的计算能够提高算法对其用户的推荐准确率。(3)电影推荐系统的实现。通过利用本文提出的关键算法对该系统进行模块化分析设计与实现。系统利用SSH框架进行后台开发,该平台提供了用户电影的相关推荐,以及用户对浏览项目进行评分等功能。
[Abstract]:With the rapid development of science and technology, the data in the Internet is increasing explosively. How to accurately and quickly find the useful information needed by users in the huge data has become the focus of research recommendation field. The two basic ways for users to obtain valuable information are the manual search of users and the personalized recommendation of the system. For some fields, it is necessary to recommend the desired information in a more intelligent way for users. The main contents of this paper are as follows: (1) an interest drift detection model based on explicit and implicit interest drift detection is proposed. Because the user's interest will change with time, the explicit interest drift detection model is used to detect the user's interest, and the implicit interest correction is used. In this paper, we propose five interest models, and use their correlation attribute value and correlation intensity increment to match their patterns and get the user interest sequence. The sequence is modified by implicit interest using exponential forgetting function, and then the real interest of the user is obtained. The experimental results show that, The accuracy and recall rate of explicit interest drift model for user interest is 88.230.96% and 87.43% respectively, which can accurately capture the user's current interest. A collaborative filtering algorithm based on explicit and implicit interest drift detection model is designed. Aiming at the problem of low recommendation accuracy of collaborative filtering algorithm, the new similarity is obtained by weighted attribute similarity and score similarity, and the current interest of users is obtained by explicit implicit interest drift detection model using item-based collaborative filtering algorithm. And carry on the related project recommendation. The experimental results show that compared with the traditional collaborative filtering algorithm, the average absolute deviation MAE value of this algorithm is 3.41% lower than that of the traditional collaborative filtering algorithm. The detection of user interest and the calculation of mixed similarity can improve the accuracy of the proposed algorithm. By using the key algorithm proposed in this paper, the modularization analysis and implementation of the system are carried out. The system uses the SSH framework for background development, the platform provides users with the relevant recommendations of movies, as well as users to browse the project score and other functions.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】