社交网站中基于用户社会活动和好友网络的推荐技术研究

发布时间：2018-07-05 06:00

本文选题：社会化推荐 + 社交网站　；参考：《复旦大学》2014年硕士论文

【摘要】：随着互联网的飞速发展,全球网民数量急剧增长。互联网世界中,人们在获取信息的同时也创造着信息,如何为用户挖掘有用的信息,避免信息过载带来的不良体验,成为学术界和业界关注的热点问题。个性化推荐系统便为解决此类问题应运而生,它旨在分析并挖掘用户兴趣,帮助用户在大量信息中快速作出决策；或为用户推荐潜在感兴趣的内容,从而提升用户体验。近年来的社交网络站点大多已实现了推荐系统的雏形,可为用户推荐好友或感兴趣的内容。随着社交网站用户规模和好友网络的不断扩大,用户生成内容急剧增多,社交网站中用户面临两个普遍问题：(1)由于信息过载,导致用户错过感兴趣的话题；(2)由于好友众多,话题的分享者难以筛选出待分享的目标用户。协同过滤技术(Collaborative Filtering, CF)是迄今为止最成功的个性化推荐技术之一。由于社交网站自身特性,基于协同过滤技术的传统推荐方法在用于社交网站的推荐时,存在一定的局限性。近些年,基于社交网站的个性化推荐的研究越来越多,大部分的文献关注于将社交网站中社会上下文信息建模集成到协同过滤模型中以改进推荐效果。本文从上述两个问题出发,基于协同过滤的基本思想,从聚集相关用户的角度将可能错过的话题推荐给用户,并为分享者推荐好友列表辅助筛选目标分享用户,主要工作包括以下几个方面：·提出一个基于用户社会活动和好友网络的推荐算法SoSAN,它结合用户之间的关注度和兴趣相似度构建用户之间影响度。SoSAN推荐算法在计算用户相似度时采用本文基于Jaccard改进的相似性方法,该方法扩大了用户共同评论行为的权重。基于真实社交网络的实验分析表明,基于影响度的推荐可提高推荐质量,基于Jaccard改进的相似度方法比标准Jaccard表现出更佳效果；·提出一个ComL线性模型,用于为分享者推荐一个好友列表辅助筛选目标好友,它基于分享者的分享习惯和候选好友对分享话题的兴趣度构建。基于真实社交网络的实验分析表明,ComL可表现出最优的命中率；·提出一个可应用于典型社交网站的具有良好通用性的推荐系统框架一AOPUT,它包含两个核心功能：(a)基于SoSAN算法将特定话题推荐给对话题感兴趣的用户：(b)基于ComL模型为分享者推荐待分享的目标用户。对该框架的主要组件、工作流程、数据模型及算法设计进行了详细介绍,并分析了框架的通用性和响应性能。
[Abstract]:With the rapid development of the Internet, the number of Internet users in the world has increased dramatically. In the Internet world, people not only obtain information but also create information. How to mine useful information for users and avoid the bad experience caused by information overload has become a hot issue in academia and industry. In order to solve this kind of problems, personalized recommendation system arises at the historic moment. It aims to analyze and excavate users' interests, help users to make decisions quickly in a large amount of information, or recommend content of potential interest to users so as to enhance user experience. In recent years, most social network sites have realized the prototype of recommendation system, which can recommend friends or content of interest to users. With the continuous expansion of user size and friend network of social networking sites, user-generated content has increased dramatically. Users in social networking sites face two common problems: (1) users miss topics of interest due to information overload; (2) because of the number of friends, users miss topics of interest. Topic sharers find it difficult to screen target users to share. Collaborative filtering (CF) is one of the most successful personalized recommendation technologies. Because of its own characteristics, the traditional recommendation method based on collaborative filtering technology has some limitations when it is used for social networking site recommendation. In recent years, there are more and more researches on personalized recommendation based on social networking site. Most of the literatures focus on integrating social context information modeling into collaborative filtering model to improve the recommendation effect. Starting from the above two problems and based on the basic idea of collaborative filtering, this paper recommends the topic that may be missed to the user from the angle of gathering related users, and assists in filtering the target sharing user for the list of recommended friends. The main work includes the following aspects: a recommendation algorithm SoSAN based on user social activities and friend network is proposed, which combines the attention and interest similarity between users to construct the influence degree between users. The similarity method based on Jaccard is used to calculate user similarity. This method expands the weight of user's common comment behavior. The experimental analysis based on real social network shows that the recommendation based on influence degree can improve the quality of recommendation, and the improved similarity method based on Jaccard is more effective than the standard Jaccard, and a ComL linear model is proposed. It is used to recommend a friend list to assist the selection of target friends. It is based on the sharing habits of the sharer and the interest of the candidate friends in the sharing topic. Experimental analysis based on real social network shows that ComL can show the best hit ratio. In this paper, a general recommendation system framework, AOPUTT, which can be applied to typical social networking sites, is proposed. It contains two core functions: (a) based on SoSan algorithm to recommend specific topics to users interested in topics: (b) based on ComL-based The model recommends the target user to be shared for the sharer. The main components, workflow, data model and algorithm design of the framework are introduced in detail, and the generality and response performance of the framework are analyzed.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092

【相似文献】