Article Preview
Top1. Introduction
Sentiment Analysis (SA), also known as opinion mining or opinion analysis, is a process of detecting, extracting, analyzing, and classifying subjective and objective texts by using text mining and natural language processing techniques (Ravi and Ravi 2015). Along with the development of online social networks, especially the emergence of microblogging platforms such as Twitter and Sina-Weibo, users can easily produce blogs and distribute their feelings, emotions, and attitudes on breaking news, public events, or products. The massive amount of microblogging data is a useful and timely source that carries mass sentiments and opinions on various topics related to political, economic, and social life, among other topics. Effectively mining the information within this text data, which is full of opinions and sentiments, is of great practical value for market intelligence (Li and Li 2013), recommendations (Ren and Wu 2013), information prediction (Rui et al. 2012), and public opinion monitoring (Shi et al. 2013). Therefore, in recent years, Microblog Sentiment Analysis (MSA) has attracted an increasing number of researchers’ attention and become a hot research area.
Two main SA techniques have been intensively applied: lexicon-based approaches and machine learning. Studies tend to apply these two techniques directly to the content of microblogging text (Go et al. 2009; Bermingham and Smeaton 2010; Kiritchenko et al. 2014). However, the microblogging texts are characterized by their short length, low information content, unstructured nature, and high data noise, resulting in poor accuracy for MSA using content-based machine learning methods. Moreover, the complexity and diversity of microblogging data, as well as the characteristics of certain informal words, such as acronyms or new abbreviations, also requires the construction of a wide range of domain dictionaries for lexicon-based approaches. Thus, it is a challenge to apply traditional pure content-based methods to MSA.
In fact, microblogs are networks, with users being nodes in social network, and based on the “mutual follow” link structure in the social network, users’ sentiments are easily unconsciously influenced by other nodes in the social network. Research carried out on SA for a friendship relations network on Myspace indicated that linked users usually tend to have the same emotional tendencies (Thelwall 2010). Meanwhile, some recent studies have pointed out that certain sociological theories, such as sentiment consistency (Abelson 1983) and emotional contagion (Hatfreld et al. 1993), can be usefully applied to social networks. Moreover, based on emotional contagion between friends, researchers have reported the phenomenon of sentiment diffusion (Miller et al. 2011) in the “follower/followee” relations (also called friend relations) in social networks. Therefore, researchers have increasingly realized that MSA is no longer confined to traditional content-based analysis, and that incorporating the users’ friend relationship information into the analysis process may be useful to improve the accuracy of MSA (Tan et al. 2011; Hu et al. 2013).
With the arrival of Web 2.0, interactive behavior has become a basic characteristic of users. In addition to the user relationship of “follower/followee”, users’ interactive behaviors, such as retweeting, pressing a “like” button, or commenting on other people’s posts, may also reflect emotional contagion (Hatfreld et al. 1993) to a certain extent. Based on the enormous volume of data on Twitter, a study of the homogeneity of happiness in social networks indicates that users are more likely to choose people with the same index of happiness to interact with (Bollen et al. 2011). Although some researchers have been aware of the importance of user interaction information, only a small number of researchers incorporate this information into MSA (Deng et al. 2014; So and Were 2016). Moreover, when mining social network data, research usually considers the similarity of the nodes. According to the sociological theory that “birds of a feather flock together” (Mcpherson et al. 2003), users who are similar to others in some way, such as by education or location, tend to hold similar attitudes or express the same sentiment on a particular topic. However, research incorporating user similarity information in MSA is lacking.