基于冲突选择模型的用户偏好挖掘及活动参与预测研究

发布时间：2018-07-28 13:27

【摘要】：近年来,基于位置的社交网络(LBSNs)的蓬勃发展,特别是随着线下社交活动组织平台的建立,使得互联网上的陌生人可以在现实世界中相见。这种全新的商业模式为线下活动的组织者提出了新的挑战,例如设计活动方案以及预测活动参加人数。直觉上,这些服务依赖于对于用户偏好的准确评估。然而,因为用户参加活动可能出于多种动机,即用户参加了活动并不一定体现了用户的真实偏好,所以传统的基于分析用户历史参加活动记录的推荐技术可能不能准确地挖掘出用户的真实偏好。与此同时,我们发现用户有时会面临"冲突-选择"的现象,即用户同时收到多个线下活动的邀请,但是由于兴趣、时间、经费等原因,用户只能选择参加部分活动,并拒绝其他活动的邀请。我们认为这种"冲突-选择"现象可能会更好地反应出用户的真实偏好。因此,准确地建模这种"冲突-选择"现象将更加有利于对用户行为的理解。本文设计了一种新颖的基于"冲突-选择"现象的模型,重建用户面临多个活动邀请时的决策过程。具体而言,从选择模型中的"效用"的视角,本文公式化了用户的"冲突-选择"效用,并且整合了用户基于活动内容的效用、基于社交关系的效用和基于活动消费的效用,这样,本文提出的模型可以同时挖掘用户对于活动内容的偏好,对于社交关系的偏好以及对于活动消费的偏好。此外,本文将"冲突-选择"三元组中的选择问题转化为成对的排序问题,并提出一个基于排序学习的优化模型框架的解决方案。本文在真实的数据集上进行了系统的实验,结果表明,"冲突-选择"框架的准确率明显优于其他基线算法。此外,本文进行了基于网络结构指标的社交网络剪枝,进一步地保证了"冲突-选择"框架的计算效率,验证了本文的"冲突-选择"框架可以更好地解释用户的真实偏好的假设,并且同时表明了"冲突-选择"框架在线下社交活动参与预测应用中的潜力。
[Abstract]:In recent years, with the rapid development of location-based social network (LBSNs), especially with the establishment of offline social activity organization platform, strangers on the Internet can meet in the real world. This new business model presents new challenges for offline event organizers, such as designing program and predicting the number of participants. Intuitively, these services rely on an accurate assessment of user preferences. However, because users can participate in activities for a variety of motives, that is, the user's participation in the activity does not necessarily reflect the user's true preference, Therefore, the traditional recommendation technology based on analyzing the history of users' participation in activities may not be able to accurately excavate users' true preferences. At the same time, we find that users sometimes face the phenomenon of "opt-to-choose", that is, users receive invitations for more than one offline activity at the same time, but because of interest, time, funding and other reasons, users can only choose to participate in some activities. And reject invitations to other events. We believe that this conflict-choice phenomenon may better reflect users' true preferences. Therefore, accurate modeling of this conflict-choice phenomenon will be more conducive to understanding user behavior. In this paper, a novel model based on the phenomenon of "conflict and selection" is designed to reconstruct the decision-making process of users facing multiple activity invitations. Specifically, from the perspective of "utility" in the selection model, this paper formulates the "conflict-choice" utility of the user, and integrates the utility of the user based on the activity content, the utility based on the social relationship and the utility based on the activity consumption. In this way, the proposed model can simultaneously mine users' preferences for active content, social relationships and activity consumption. In addition, in this paper, the selection problem in the "conflict-selection" triple is transformed into a pair ordering problem, and a solution to the optimization model framework based on ranking learning is proposed. In this paper, a systematic experiment is carried out on a real data set. The results show that the accuracy of the "conflict selection" framework is obviously better than that of other baseline algorithms. In addition, the network pruning based on the network structure index is carried out in this paper to further ensure the computational efficiency of the "conflict selection" framework. This paper verifies the hypothesis that the "conflict-choice" framework can better explain users' real preferences, and also shows the potential of the "conflict-choice" framework for off-line social activities to participate in prediction applications.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】