社会网络中的情感影响模型建立及分析

发布时间：2018-04-13 10:00

本文选题：Twitter数据 + 情感影响研究　；参考：《北京邮电大学》2014年博士论文

【摘要】：Web2.0时代的技术发展正在不断影响和改变着人们的生活。各种各样的社会网络服务,给人们的在线互动交流带来了前所未有的便利与快捷,同时也开启了大规模真实数据的时代,为基于社会网络的用户行为研究提供了机会与挑战。一方面,传统研究中不易获得的大量用户数据及关系信息,现在可以由大型社会网络服务平台较为轻松地获得；另一方面,处理这些大规模数据并从中抽取出有用的信息应用于实际任务,也为研究者们提出了新的难题。用户的情感信息被认为是能够影响用户行为决策的主要因素之一。随着社会网络服务的日益壮大和发展,如何利用从社会网络数据中获取到的相关情感信息来对网络中的用户行为和主观观点进行分析和预测,是非常具有实际研究意义的工作。本文以真实社会网络中出现的用户和话题为对象,着重研究了在不同的用户行为预测任务下情感影响模型的建立和分析,并通过实验给出了一系列有参考价值的结果。这三个研究任务分别为：用户关系预测,个性化话题推荐,以及用户对话题的情感推测。各个任务中的实验结果均表明,有针对性地利用情感影响因素而建立的预测模型,能够得到比已有方法更好的效果。论文的主要工作和贡献如下： (1)基于近年来最受欢迎的社会网络服务之一的Twitter,爬取了一段时间内的真实数据,建立了一个可用于各项实验的Twitter数据库。该数据库不仅包括了Twitter用户的基本信息,还可以提取到用户间的关系网络,以及每个用户在该时段内的发布的消息文本。文中应用了一个快速有效的情感分析工具对数据库中的用户消息文本进行了情感类别标记。 (2)对用户关系预测中的情感影响进行了研究。这部分工作中考虑的情感影响因素是用户在社会网络中带有情感倾向的影响力。首先定义和计算了用户情感影响力,并基于计算出的结果对用户进行了属性划分。本文将用户情感影响力属性作为新的特征,针对两个不同的用户关系预测子任务分别建立了情感影响模型SA-UFP和SA-RFP。对比实验的结果分析显示,SA-UFP和SA-RFP模型能够有效提高预测正确率。 (3)对个性化话题推荐中的情感影响进行了研究。这部分工作中考虑的情感影响因素是社会网络话题下用户情感观点分布的影响。文中提出了关于话题的情感分布特征,并在真实数据上对它们进行了观察分析,而后基于话题情感分布对用户兴趣的影响建立了SDA-TR话题推荐模型。通过与已有推荐模型进行对比实验分析,证明了SDA-TR模型能够更好地为用户进行个性化话题推荐。 (4)对用户对话题的情感推测这一应用任务中的情感影响进行了研究。这部分工作中考虑的情感影响因素是朋友用户间的相互情感影响。在分析了朋友用户间的情感影响并验证了相关假设的基础上,本文建立了SFMF推测模型。用户对话题情感推测任务上的对比实验分析表明,考虑了情感影响的SFMF模型更为准确有效。
[Abstract]:Web 2.0 technology development is continuously influencing and changing people ' s life . Various kinds of social networking services bring unprecedented convenience and shortcut to people ' s online interactive exchange , meanwhile , it also opens up the era of large - scale real data , and provides the opportunity and challenge for the research of user behavior based on social network . On the one hand , the large number of user data and relationship information which are not easily obtained in the traditional research can be easily obtained by the large social network service platform ;
On the other hand , processing these large - scale data and extracting useful information from them is a new challenge for researchers . The user ' s emotional information is thought to be one of the main factors that can influence the user ' s behavior decision - making . With the growth and development of social networking services , how to analyze and predict the user ' s behavior and subjective viewpoint from the social network data is very meaningful .

Based on the users and topics appearing in the real social network , this paper focuses on the establishment and analysis of the emotion influence model under different user behavior prediction tasks , and gives a series of valuable results through experiments . The three research tasks are : user relation prediction , personalized topic recommendation and user ' s emotion speculation about the topic .

The main work and contribution of the thesis are as follows :

( 1 ) Based on Twitter , one of the most popular social networking services in recent years , the real data over a period of time has been crawled , and a Twitter database that can be used in various experiments has been established . The database not only includes the basic information of Twitter users , but also the relationship network between users , as well as the text of messages published by each user during that period . A quick and effective emotion analysis tool is used to mark the user message text in the database .

( 2 ) The influence of emotion on user ' s relationship is studied . The affective factors considered in this part are the influence of user ' s emotional tendency in the social network . Firstly , we define and calculate the influence of user ' s emotion , and divide the attribute of the user based on the result of the calculation . This paper sets up the emotion influence model SA - UFP and SA - RFP respectively for two different user relationship prediction sub - tasks . The results show that SA - UFP and SA - RFP model can improve the prediction accuracy effectively .

( 3 ) The emotion influence in the recommendation of personalized topic is studied . The emotion influencing factor considered in this part is the influence of the user ' s emotional view distribution under the topic of social network . The article puts forward the emotion distribution characteristic of the topic , and then sets up the SDA - TR topic recommendation model based on the influence of the topic emotion distribution on the user ' s interest . Through the comparison experiment analysis with the existing recommendation model , it is proved that the SDA - TR model can make personalized topic recommendation better for the user .

( 4 ) The emotional influence of the user on the subject ' s emotion is studied . The affective factors considered in this part are mutual affection between friends and users . Based on the analysis of the emotional impact between friends and users and the related assumptions , a SFMF speculation model is established . The comparison between the user ' s emotion estimation task shows that the SFMF model considering the influence of emotion is more accurate and effective .

【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】