Influential user weighted sentiment analysis on topic based microblogging community

In recent years, the number of social microblogging services has grown rapidly. They have become a popular expression platform of what people think. People use these platforms to produce content on different topics in real time. With the proliferation of social microblogging sites, the massive amount of opinion texts have become available in digital forms, thus enabling research on sentiment analysis to both deepen and broaden in different sociological fields. Sentiment analysis research field, also named as opinion mining, is a sub discipline of computational linguistics and data mining. The aim of this research field is to determine a person's mood and opinion from the textual documents created by him. Nowadays, because of the proliferation of social microblogging services, sentiment analysis has become a popular research area in computer science. The researchers have made several important sentiment analysis studies in different sociological topics such as politics, economy and especially in the finance field on microblogging service. The research area of influence is derived from sociology and psychology. A number of studies are conducted to identify influence. In an important study, influence means change in a person's attitude or behavior, which has its origin in another person or group of people. In another definition in the literature, influence is defined as affect other people in their immediate environment. In another study, researcher describes influence as the behavior of an individual which can directly or indirectly impress the thoughts, opinions, and actions of other people in a population. Social Microblogging services have built an audience in nearly every topic. Today, it is a common practice for people to follow each other to get relevant information about their topics of interest. Thereby, common interest communities are formed in social networks. By the agency of these communities, information about the topic is spread very quickly and influences the community members very effectively. Thus, microblogging users who have common interests influence each other in an active way. Some users in a microblogging community have a higher degree of influence on the opinions of community members. These people called as influential users. When studies in the literature about influential users in social network are examined, it is seen that trusted users in a community are actually influential users. Because, many studies in the literature indicate that influential users are generally more involved and updated on the topic in which they are interested. Other community members see them as experts and trustworthy about the topic and are closely influenced by them. For this reason, it is easily to say that their postings reflect opinions more realistic than the opinions of other community members. Therefore, calculating users's influence values in a community is highly important in determining a community's opinion about the topic. In this study, existing sentiment analysis studies on microblogging services are examined in details. After a comprehensive analysis of previous research it is seen that they generally focused only on the text contents of microblogging sites and ignored the users in a social community. The researchers regarded microblogging users as equals in their studies. They ignored the users trust as well as their postings reliability in their analysis. Some researchers considered the user's importance on the community as well. In their studies, they only used sentiment analysis methods, which incorporate social network analysis methods to assess a user posting's sentiment polarity or a user's sentiment mood. They did not use social network analysis methods to evaluate the user's trust for sentiment analysis in a topic based community. On the other hand, some researchers define influential users and communities in order to assign sentiment of the communities towards selected topics but they did not use his influence value on the social community to measure the reliability of his postings. In this study, it is proposed that the effect of influential users should also be addressed while analyzing and determining a community's opinion about a topic. Hence, this thesis research proposes a novel sentiment analysis approach that takes into account the social network information as well. The study concentrates on the effect of influential users on the sentiment polarity of a topic based microblogging community. The method, which is proposed in this study, extends the classical sentiment analysis methods, which only consider text content, by adding a novel PageRank based influential user-finding method. In this thesis in order to suggest an influential user weighted setniment analysis and prove the research's suggestions, firstly, a detailed research is carried out on how people influence each other in social communities and why some social sharing information is more influential than the others. With the information obtained as a result of these researches, the necessary features of influential user in microblogging services has been defined. In the next step, these features are considered in order to identify the new influential user metrics to solve deficiencies of previous influential user-finding approaches. After examination previous influential user-finding methods, the PageRank based approaches which are the best solution but has some deficiencies for finding influential user in social network, has been chosen to bu used in the proposed novel method. Next the PageRank method was extended with the new defined influential user's metrics to eliminate the deficiencies. So that new influential user-finding method is defined. In the second phase of this thesis study, in order to adress the effect of influential users while analyzing and determining a community's sentiment about any topic, the level values of influential user were determined by the proposed PageRank based method and microblogging community's users sharings were weighted according to these values. Thus, previous sentiment analysis studies which incorporate social network analysis methods have been extended. As a result, this study defined a new sentiment analysis method that includes the influential users in topic-based microblogging communities. Finally, in the experiment section, The comprehensive empirical study of two real world Twitter datasets are carried out to analyze the correlation between the sentiment polarity of the financial social community and the behavior of the stock exchange of Turkey, namely BIST100, using Pearson correlation coefficient method. The proposed method is compared with previous classical sentiment analysis methods which use only text content in sentiment polarity calculation, and also with other sentiment analysis methods that are extended with different PageRank based approaches. Experimental results validate the research's assumptions and show that the proposed PageRank based influential user weighted sentiment analysis method is more effective than previous studies in finding topic based microblogging community's sentiment polarity.