Opinion Dynamics: A Multidisciplinary Review and Perspective on Future Rearch
Haoxiang Xia, Huili Wang, Zhaoguo Xuan
西番莲果Dalian University of Technology, China
ABSTRACT
As a key sub-field of social dynamics and sociophysics, opinion dynamics utilizes mathematical and physical models and the agent-bad computational modeling tools, to investigate the spreading of opinions in a collection of human beings. This rearch field stems from various disciplines in social sciences, especially the social influence models developed in social psychology and sociology. A multidisciplinary review is given in this paper, attempting to keep track of the historical development of the field and to shed light on its future directions. In the review, we firstly discuss the disciplinary origins of opinion dynamics, showing that the combination of the social process, which are conventionally studied in social sciences, and the analytical and computational tools, which are developed in mathematics, physics and complex system studies, gives birth to the interdisciplinary field of opinion dynamics. The current state of the art of opinion dynamics is then overviewed, with the rearch progress on the typical models like the voter model, the Sznajd model, the culture dismination mo
空落del, and the bounded confidence model being highlighted. Correspondingly, the future directions of this academic field are envisioned, with an advocation for clor synthesis of the related disciplines.
Keywords: Opinion Dynamics, Social Influence, Voter Model, Sznajd Model, Culture Dismination Model, Bounded Confidence Model
INTRODUCTION
In the last decades, there are growing interests on applying the models and tools that are developed in complex system studies to investigated various complex social phenomena, under the brands of, for example, “generative social science” (Epstein, 2007), “computational social science” (Bankes, et al, 2002), and “sociophysics” (Galam, 2008). One notable sub-field of this new rearch paradigm is so-called “opinion dynamics” (Hoylst, et al, 2001), which can roughly be defined as a rearch field in which mathematical-and-physical models and computational tools are utilized to explore the dynamical process of the diffusion and evolution of opinions in human population. In recent years, the studies on opinion dynamics have attracted wide attention, especially in the statistical physics community; and many interesting outcomes have been reported. Nevertheless, today this rearch field is still at its very infancy. As noted by (Castellano, et al, 2009), “the development of opinion dyna
mics so far has been uncoordinated and bad on individual attempts … without a general shared framework and often with no reference to real sociological studies”. In order to develop this rearch field, it is worthwhile to track its academic origins and development trails, and to envision its future trends.
With the prominence the related rearch endeavors, there have been some good academic reviews on opinion dynamics, such as (Castellano, et al, 2009), (Staufer, 2005), and (Lorenz,
2007), which are mostly bad on the viewpoint of statistical physics. Tho physics-centric efforts have demonstrated great significance as they may reveal that the societal process also obey the universal statistical laws that govern the motions of massive particles in nature; and it might become possible to bridge the disciplinary gap between natural and social sciences by adopting the well-established mathematical, physical and computational methods to study such societal process. Correspondingly, the review on opinion dynamics from the physics point of view is doubtlessly valuable. However, our contention is that tho physics-centered reviews may be insufficient for offering a comprehensive view on opinion dynamics, regarding the interdisciplinary nature of this rearch field. On one hand, the physical models on opinion dynamics, which are currently in the spotlight, are fundamentally bad upon earlier investigations in various social disciplines, especially
教跳舞in social psychology and sociology. Tho social investigations are critical to today’s physical modeling of opinion dynamics as the key problems and modeling assumptions of physical modeling are both originated from the social investigations. On the other hand, for the future development of opinion dynamics, it is derved to pursue deeper confluence between the “hard” disciplines (e.g. mathematics, physics and computer science) and the “soft” disciplines such as sociology, social psychology, communication studies, cultural anthropology, and politics. The insufficiency of such disciplinary confluence is the key drawback of the current physics-centered studies. This also requests a multidisciplinary view on this field. For the above reasons, we in this paper attempt to review this rearch field from a wider disciplinary scope, regarding opinion dynamics as confluence of multiple academic sources.
In tune with this aim, the remainder of this paper is organized as follows. In the next ction, we try to answer the question “where opinion dynamics comes from” by giving a multidisciplinary retrospect on the academic sources of opinion dynamics. Then, the key rearch endeavors on opinion dynamics are highlighted in the subquent ction and the state of the art is discusd in accordance with a modeling framework that is induced from the reviewed contributions. Some rearch directions are envisioned conquently and the paper ends up with a short concluding ction.
THE DISCIPLINARY ORIGINS OF OPINION DYNAMICS
The studies on opinion dynamics cover a wide range of social phenomena, e.g. emergence of fashions and fads, minority opinion survival and spreading, collective decision making, connsus building, emergence of political parties, diffusion of rumor, expansion of extremism, and propagation of cult. The phenomena are the persisting topics of interest in many areas of social sciences. In this ction, we try to take a brief retrospect to examine where opinion dynamics comes from.
肌肉紧张
Various studies in sociology and social psychology have great impact on the birth of opinion dynamics. One major theme of sociology is to explore the collective action of human population. Since the 19th century, the irrational and sometime maniac collective behaviors have been noticed and discusd, which were termed as the “madness of crowds” by Mackey (Mackey, 1841) and the “stupidity of crowds” by Le Bon (Le Bon, 1895). This obrvation was then followed by a ries of experimental studies in social psychology on “conformity”, including the “social norm” by Sherif (Sherif, 1936), the “group press” by Asch (Asch, 1951), and the “obedience” by Milgram (Milgram, 1974), and so on. The studies reveal that there is a common tendency that individuals’ attitudes and behaviors conform to the majority of the belonging group. In Economics, this conformity is often termed as “herd behavior”, as firstly
苗族习俗
introduced in Keynes’ (Keynes, 1936) “General Theory” and extensively investigated thereafter in the context of, e.g., corporate investment, financial markets and intra-firm decision-making (Scharfstein & Stein, 1990 and Banerjee, 1992). To explore the underlying mechanisms of the collective behaviors, various “threshold” or “critical mass” models have been developed in sociology and economics, e.g. (Granovettor, 1978), (Schelling, 1978), and (Brock & Durlauf2001). The models of collective behaviors have prominently influenced the growth of opinion dynamics studies since the formation of public opinions and social norms is often the premi of collective actions; and the studies of collective behaviors and of public opinions are actually inparable.
热点新闻Besides the aforementioned studies on psychological conformity and collective behaviors, the topic of opinions is more directly involved in a ries of sociological and social-psychological endeavors on the issues like group polarization (Moscovici & Zavalloni, 1969), minority influence (Moscovici, et al, 1969) and social influence models. The endeavors on group polarization and minority influence phenomenologically reveal the intrinsic complexity of opinion spreading in society; and the phenomena are also the key subjects of interest in today’s opinion dynamics rearches. The endeavors on social influence models, on the other hand, provide the early disciplinary treatments on the underlying mechanisms of opinion spreading. The sociological and socio-psychological end
eavors are the direct ancestors of today’s opinion dynamics rearches. One breakthrough work on social influence modeling is the two-step flow model of influence as formulated and elaborated by Lazarsfeld and his colleagues at Columbia University in the 1940-50s (Katz & Lazarsfeld, 1955). This model stress that the massive individuals, instead of being directly persuaded by the media, usually follow a smaller number of opinion leaders who bridge the media and the mass public. Another famous early work on social influence modeling is French’s (French, 1956) social power model. In this work, French propod a Markov chain model to describe the diffusion of social influence and the formation of public opinions in social networks, bad on a discussion and classification of “social power”. French’s social power model was further developed by French and Raven (French & Raven, 1959), who original rearch was concerned with situations in which a supervisor influences a worker. This social power model has become a good source for subquent social influence studies, which fall into five main areas, as summarized by Rashotte (Rashotte, 2007): (1) minority influence in group ttings, (2) rearch on persuasion, (3) dynamic social impact theory, (4) structural approach to social influence, and (5) social influence in expectation states theory.离职告别
Further progress were made on the social influence models in the 1980 and 1990s. In this respect,
Latané and his collaborators and followers made noticeable contributions. In 1981, Latané propod a “social impact theory”, which provides a microscopic analysis on how an individual is influenced by the surrounding social environments. For him, the impact of any information source on the individual relies on three factors: the number of others who make up that source, their immediacy, and their social strength (i.e. salience or power). Following this social impact theory, Latané and his collaborators ud cellular automata modeling and computer simulation to study the global effects of the individual attitude change (Nowak et al. 1990). This work was more systematically treated as the “dynamic social impact theory” by Latané (1996). In the dynamic social impact theory, social structure lf-organizes and evolves through the dynamic and iterative interpersonal influences. The overall process of opinion spreading begins from a random distribution of attitudes and beliefs. Within the social network, one is more likely to be influenced by someone nearby than by the ones far away; and localized cultures of beliefs may be a result of such influences. It can be argued that in the dynamic social impact theory a solid foundation has been established for the more
recent opinion dynamics rearches. During the same period, Friedkin and his colleagues took a structural approach to examine interpersonal influence within a larger social network of influence. Their endeavors constitute a “social influence network theory” (Friedkin 1998). This theory utilizes m
athematical models and quantification to measure the process of social influence. Within the social influence network, a “norm” opinion is formed as weighted average of the individuals’ private opinions on some issue; and in turn, the individuals modify their own opinions in respon to this norm by averaging their initial opinions and the social norm. The interpersonal influences are reprented as the weights that determine the formation of the norm and the change of the personal opinions. Friedkin’s work makes remarkable contribution to formalize social influence and opinion evolution in social networks; however, it is criticized by some authors, e.g. Heglmann & Krau (2002), due to the propod model is esntially a linear model which oversimplifies the real-world situations.
The prior introduction indicates that the social-influence studies carried out in sociology and social psychology have profound impact on the development of opinion dynamics. Various other disciplines, nevertheless, have also contributed to the rising of opinion dynamics, since this rearch field actually involves in a wide range of interests. One ries of work that is cloly related to opinion dynamics is the inquiries on connsus building in decision science (particularly in group decision-making studies). The studies on connsus building in the context of decision-making were initialized by De Groot (De Groot, 1974) and Lehrer (Lehrer, 1975) and a plenty of subquent inquiri
es in the 1970-80s (Chatterjee & Seneta, 1977, Wagner, 1978, Cohen & Newman, 1986). The focus of the efforts is to examine the process of connsus-building in a group and to develop methods to facilitate connsus-building. The modeling of non-connsus situations (i.e. disagreements) is less concerned.
In the contexts of economics and politics, one issue that critically influences the development opinion dynamics is the voting problem. Herbert Simon (Herbert Simon, 1954) propod an early model on this problem, which treated the “bandwagon and underdog effects” in voting. Simon’s model was later extended by Kuran (Kuran, 1987). Another well-noted work on the voting problem was originated in Clifford and Sudbury (Clifford & Sudbury 1973), which started the streamline of rearch on the “voter model” (Holley & Liggett, 1975, Dornic, et al, 2001, Lambiotte, et al, 2009). The voter model studies the evolution of personal position for a particular issue (e.g. the voting choice between two candidates) within a human collection. One individual’s position may adapt to tho of the neighbors’. Thus the distribution of the positions of the overall population evolves over time through the local interpersonal influences. In the last decades such voter model has been extensively investigated and interesting outcomes have been obtained. This streamline of rearch contributes noticeably on opinion dynamics (Ben-Naim, 2005). Besides the voter model, the culture
dismination model that was originally propod by Axelrod (Axelrod, 1997) is also cloly related to opinion dynamics. In his work, the term “culture” was ud to refer to a broad range of things that are subject to social influence, such as beliefs, attitudes, language, art, technical standards, and social norms. Axelrod’s model can be regarded as an opinion-dynamical model, in which opinions contain multiple dimensions. Axelrod’s work has been followed by many rearchers, who either gave further analysis on the cultural dynamics model, or suggested model extensions (Castellano, et al, 2000, Klemm, et al, 2005, Centora, et al, 2007). The previously-discusd models are all concerned about the aggregate dynamics of attitudes, voter choices, and cultures upon the interpersonal social influences. Opinion dynamics is growing from tho endeavors as a rearch field that covers the common features behind the
phenomena that are conventionally studies in multiple social-science disciplines. In this academic confluence, the u of mathematical, physical and computational tools plays a vital role. The term “opinion dynamics” is coined in a mathematical way since the spreading and evolution of opinions are mathematically modeled and studied as a dynamic process of a differential and/or difference system. To this end, the mathematical studies on the spreading of infectious dias, most famously the studies on the SIS and SIR models (Hethcote, 2000), may give implications to the studi
es on opinion dynamics. As noted, the key social force that enables the diffusion of opinions in a collection is all sorts of “social influence” or “social contagion” (Dodds & Watts, 2004), which can to some extent be analogous to the “infection” of dias; thus, the mathematical models of the spreading of infectious dias can be borrowed and revid to study the diffusion of opinions through contagions, e.g., cc. (Moreno, et al, 2005, Grönlund & Holme, 2004). The social process have also come into many physicists’ sight since the 1970s. In the physics community, the dynamics of group opinions have persistently been investigated by using statistical-mechanics models, e.g. the probability model of group polarization by Weidlich (Weidlich, 1971) and the Ising model for attitude change and connsus formation propod by Galam & Moscovici (Galam & Moscovici, 1991). More recently, with the blooming of complex adaptive system rearches, agent-bad computational modeling has become a new weapon for taming the social process of opinion diffusion and evolution (Stauffer, 2005). Correspondingly, there have been increasing “opinion dynamics” studies reported since the mid 1990s.
From the prior description, we can e that the studies on opinion dynamics were originated in various disciplines in social sciences, especially in sociology and social psychology. This subject of inquiry has recently attracted broader attentions with the participation of mathematicians, physicists
and complex-system rearchers. Up to now, as a key sub-field of complex social system studies, opinion dynamics has been extensively investigated. Various models have been established and the underlying mechanisms for the evolution of opinions have been investigated in accordance with the established models. Subquently we try to discuss the recent developments in more detail.
THE STATE OF THE ART ON OPINION DYNAMICS
As stated, opinion dynamics is growing from very diver rearch efforts in multiple disciplines; and tremendous endeavors have contributed to the development of this rearch field. However, the related endeavors are largely uncoordinated and prently it may be difficult to construct an integral and coherent framework for cover the important aspects of all the related endeavors. Therefore, we take an inductive method to discuss the key contributions on opinion dynamics. The endeavors that constitute the main body of opinion dynamics can roughly be divided into four major streamlines, i.e., the voter model, the Snzajd model, the culture dismination model, and the bounded confidence model. Thus we outline the key work on opinion dynamics in terms of the four ries of models. The models are then compared under a general modeling structure which is shared by the ries of models and we accordingly summarize the state of the art in this rearch field.
VOTER MODEL
冬瓜虾仁
As a prominent part of opinion dynamics, the studies on the voter model are abundant since Clifford and Sudbury’s (Clifford & Sudbury, 1973) and Holley & Liggett’s (Holley & Liggett, 1975) classical work. The voter model concerns the social dynamics of the public choices on a particular issue such as the voting on two candidates. Conquently, in the simplest form,