移动互联网中基于上下文信息的用户偏好提取研究

发布时间：2019-02-12 23:44

【摘要】：迅猛发展的移动互联网逐渐渗透到了人们生活的每一个角落,给传统的产业带来了前所未有的改造、重构和颠覆。凭借着与生俱来的情景感知能力,移动互联网逐渐打开了个性化服务的新局面,这也使普适的计算日渐乏力,给传统的用户偏好提取技术带来了新的挑战。传统的用户偏好提取技术,大多致力于研究用户与项目之间的关系,而较少考虑上下文信息对用户偏好的影响。尽管目前也有部分结合少量上下文因素进行个性化服务的研究,然而这些用户偏好提取算法考虑的上下文因素较为单一,很难扩展到多维的上下文的场景,具有一定的局限性。本课题在这一背景下,重点研究了多维上下文下的用户偏好提取技术,主要内容包括:首先,本课题首先对用户偏好的研究现状进行了调研,总结了用户偏好的建模方法以及用户偏好的表示方式,并评估了多维上下文下的用户偏好提取的解决方案。其次,基于多维上下文因素的场景,提出了多维上下文的用户偏好模型,该模型可扩展,能自适应上下文因素的变化,根据用户在多维上下文场景进行用户偏好的计算。在模型的构造上,本研究首先从用户心理层面出发,将用户偏好分解为长期偏好和短期偏好,针对多维上下文因素,则提出将多维上下文分类为用户上下文、项目上下文以及环境上下文。在此基础上,分析了用户偏好与各类上下文因素的关系,构建了 CB-MF(Context Bias Matrix Factorization)模型,继而提出了与之相适应的CB-MF算法。本文随后将提出的模型和算法在LDOS-CoMoDa数据集上进行了仿真,实验结果表明,该模型能相比传统的MF(MatrixFactorization)算法有较好的性能提升,其均方误差(Root Mean Square Error,RMSE)相比 MF 算法降低了 10.38%,平均绝对误差(MeanAbsolute Error,MAE)降低了 11.51%。最后,立足于理论的研究成果,本课题设计实现了基于多维上下文用户偏好提取的视频播放平台。除了播放视频,该WEB平台还具有上下文信息采集、用户偏好提取、推荐生成的功能,同时用户偏好提取引擎使用了本研究提出的CB-MF算法,充分验证了 CB-MF算法的可行性。本文同时给出了该平台的系统结构以及各个模块的实现细节。该平台对于未来多维上下文与用户偏好关系的研究有重要的实践意义。
[Abstract]:The rapid development of mobile Internet has gradually penetrated into every corner of people's lives, bringing unprecedented transformation, reconstruction and subversion to traditional industries. With the inherent ability of situational perception, the mobile Internet has gradually opened up a new situation of personalized services, which makes the pervasive computing increasingly weak, and brings new challenges to the traditional technology of user preference extraction. Most of the traditional user preference extraction techniques focus on the relationship between the user and the project without considering the influence of context information on the user preference. Although there are some researches on personalized service based on a few contextual factors these user preference extraction algorithms take into account a single contextual factor which is difficult to be extended to multi-dimensional context scenarios and has some limitations. Under this background, this paper focuses on the technology of user preference extraction in multidimensional context. The main contents are as follows: firstly, this paper investigates the current situation of user preference. The modeling method of user preference and the representation of user preference are summarized, and the solution of user preference extraction under multidimensional context is evaluated. Secondly, based on the scenario of multidimensional context, a user preference model of multidimensional context is proposed. The model is extensible and can adapt to the change of contextual factors. The user preference can be calculated according to the user's preference in multidimensional context. In the construction of the model, the user preference is decomposed into long-term preference and short-term preference from the perspective of user psychology, and the multidimensional context is classified as user context. Project context and environment context. On this basis, the relationship between user preference and various contextual factors is analyzed, the CB-MF (Context Bias Matrix Factorization) model is constructed, and the corresponding CB-MF algorithm is proposed. The proposed model and algorithm are then simulated on the LDOS-CoMoDa dataset. The experimental results show that the proposed model can improve the performance better than the traditional MF (MatrixFactorization) algorithm, and the mean square error (Root Mean Square Error,) of the model is better than that of the traditional MF (MatrixFactorization) algorithm. Compared with the MF algorithm, RMSE reduces 10.38 points and the average absolute error (MeanAbsolute Error,MAE) decreases 11.51%. Finally, based on the theoretical research results, this paper designs and implements a video playback platform based on multidimensional context user preference extraction. In addition to playing video, the WEB platform also has the functions of context information collection, user preference extraction and recommendation generation. Meanwhile, the user preference extraction engine uses the CB-MF algorithm proposed in this study, which fully verifies the feasibility of the CB-MF algorithm. This paper also gives the system structure of the platform and the implementation details of each module. The platform has important practical significance for the future study of multidimensional context and user preference.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】