基于社交网络影响力的推荐算法的研究与实现

发布时间：2018-08-03 15:33

【摘要】：近年来，互联网不断发展与普及，尤其是手持设备的智能化与移动互联网的普及化，使得用户数量增加，在线时间加长。在这样一个互联网浪潮中，参与的用户在不断地将线下社交人脉往线上迁移，由此形成了社交网络。在社交网络中，志趣相投的用户聚集到了一起，并且由于用户可以选择关注自己感兴趣的内容，使得社交网络拥有高访问量和高聚集度的特点。从阿里巴巴与新浪微博的合作推广，到Facebook个性化广告系统的繁荣发展，都是社交网络的商业化进程的体现。而其中对用户兴趣爱好进行挖掘分析，并向用户推荐其最可能感兴趣和接受的商品或广告的核心，便是基于社交网络的推荐系统。本文首先介绍了国内外在社交网络和推荐系统两个领域的研究现状，总结了当前对基于社交网络的推荐系统的已有理论成果。然后，本文针对社交网络进行建模分析，并详细阐述了用户关系网、用户和物品特征、用户好感度的概念与数学公式化计算方法，，以用于处理社交网络关系图。接着，在此基础上提出社交网络查询方法，并结合随机森林分类器框架，形成了基于社交网络影响力模型的推荐算法。本文在提出了推荐算法的基础上，对其进行系统设计与分析，开发了一套基于B/S模型，并拥有可移植性的，模块化的推荐系统。文中首先对系统的整体结构设计进行总述，以及针对其中的主要功能模块进行详细解析。然后，说明开发环境与相关技术，并对系统的具体实现进行展示。最后，本文采用数据集针对提出的推荐算法进行参数实验、冷启动实验和算法耗时实验。通过实验结果分析，本文实现的基于社交网络影响力的推荐算法能较好地解决冷启动问题，并能够通过缓存策略而在算法效率方面有较好的表现。
[Abstract]:In recent years, the continuous development and popularization of the Internet, especially the intellectualization of handheld devices and the popularization of mobile Internet, make the number of users increase and the time of online increase. In such an Internet wave, participating users constantly migrate offline social networks to form social networks. In social networks, like-minded users gather together, and because users can choose to focus on the content they are interested in, the social network has the characteristics of high traffic and high concentration. From the cooperation between Alibaba and Sina Weibo, to the prosperity of Facebook personalized advertising system, it is the embodiment of the commercialization process of social network. The social network-based recommendation system is the core of the analysis of the user's interests and hobbies and the recommendation of the products or advertisements that they are most likely to be interested in and accept. This paper first introduces the current research situation of social network and recommendation system at home and abroad, and summarizes the existing theoretical achievements of the recommendation system based on social network. Then, this paper analyzes the modeling of social network, and describes the concept of user relationship network, user and object characteristics, user sensitivity and mathematical formulaic calculation method to deal with social network relationship diagram in detail. Then, a social network query method is proposed, and a recommendation algorithm based on the influence model of social network is formed by combining with the framework of stochastic forest classifier. On the basis of proposing the recommendation algorithm, this paper designs and analyzes the system, develops a set of recommendation system based on the B / S model, and has the portability and modularization of the recommendation system. Firstly, the whole structure of the system is described, and the main function modules are analyzed in detail. Then, the development environment and related technologies are explained, and the implementation of the system is demonstrated. Finally, the data set is used to carry on the parameter experiment, the cold start experiment and the algorithm time-consuming experiment. The experimental results show that the proposed recommendation algorithm based on the influence of social networks can solve the cold start problem and has a good performance in the efficiency of the algorithm by caching strategy.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP393.09;TP391.3

【共引文献】