Background: We study mechanisms underlying the collective emotional behavior of Bloggers by using the agent-based modeling and the parameters inferred from the related empirical data. Methodology/Principal Findings: A bipartite network of emotional agents and posts evolves through the addition of agents and their actions on posts. The emotion state of an agent,quantified by the arousal and the valence, fluctuates in time due to events on the connected posts, and in the moments of agent's action it is transferred to a selected post. We claim that the indirect communication of the emotion in the model rules, combined with the action-delay time and the circadian rhythm extracted from the empirical data, can explain the genesis of emotional bursts by users on popular Blogs and similar Web portals. The model also identifies the parameters and how they influence the course of the dynamics. Conclusions: The collective behavior is here recognized by the emergence of communities on the network and the fractal time-series of their emotional comments, powered by the negative emotion (critique). The evolving agents communities leave characteristic patterns of the activity in the phase space of the arousal--valence variables, where each segment represents a common emotion described in psychology.
The Internet experience in recent years has revolutionized the mechanisms that an individual can exploit to participate in global social dynamics. Consequently, new techno-social phenomena emerge on the Web [1,2,3,4], boosting an intensive multidisciplinary research. In technology research, for example, new generation of services are developing in the direction to integrate human capabilities in a service-oriented manner [5]. Behavior of the users in the virtual world has impact on real-life events, which becomes a concern of both social sciences and every day's practice. On the other hand, the data collected from massive use of the Web provide the basis to study human behavior "experimentally" at unprecedented scale. For instance, from the high-resolution data stored at various Web portals (social networks, Blogs, forums, chat-rooms, computer games, etc) information related to user preferences, patterns of behavior, attitudes, and emotions can be inferred for each individual user and user communities gathered around certain popular subjects [6,7,8,9,10,11,12,13,14,15,16].
Physics of complex systems and, in particular, the statistical physics of social dynamics, are focused on the dynamical processes in which human collective behaviors emerge from large number of individual actions [4,6,7,9]. Combining the concepts of statistical physics with the machine-learning methods for the emotion detection in texts of messages [13,17], we have recently performed analysis of large datasets from bbcblog.com and digg.com and determined quantitative measures of the collective behaviors in which the emotions are involved [14,15]. Complementary to our work in Refs. [14,15] where the empirical data are analyzed to extract various complexsystems properties, the present work is a theoretical study of the processes, underlying the emergence of the collective emotional behavior of Blog users, within the framework of agent-based modeling.
The quantitative analysis of users collective behavior in the empirical data from diggs.com and bbcblog.com in Refs. [14,15] has been enabled by mapping the high-resolution data onto bipartite networks of users and posts, as two natural partitions. The idea of bipartite networks makes the “firm ground” also in the present theoretical model, where the agents interact indirectly over the posts. We also make use of several other features, observed in various empirical data, that are relevant for designing the dynamic rules of the theoretical model:
• Universality of user’s behavior related with the action-delay and the circadian cycles [9,6,10,7];
• User communities occurring in the cyberspace are reminiscent to the ones in real life, however, different time scales and grouping mechanisms might be involved [10,14,18,15];
• Quantitative measures of emotions have been introduced in psychology research [19]. In particular, based on Russell’s multidimensional model of affect [20], each known emotion can be represented by a set of numerical values in the corresponding multidimensional space. Two fundamental components of emotion, to which we refer in this work, are the arousal, related to reactivity to a stimulation, and the valence, measuring intrinsic attractiveness or aversiveness to a stimulation. These components of emotion can be measured in laboratory based on the related psychophysiological and neurological activity [21,22]. Moreover, a systematic association has been recognized [23] between individual emotional characteristics and word use. The arousal and valence components of an emotion can be retrieved from a written text by suitable machine-learning methods, which are being developed for a specific type of data [22,24,25].
Systematic analysis of the patterns of user behaviors and the emotion contents in the texts of comments in the empirical data from popular Blogs [14], discussion-driven Diggs [15], and Forums [26], suggests that negative emotions (critique) drive the activity on these Web portals. However, the mechanisms working behind this global picture have not been well understood. In order to elucidate the role of emotions in the blogging interactions, and to point out potential parameters and levels where the process can be controlled, we devise an agent-based model. The agents are spreading their emotions in a bipartite network environment. The agent’s properties, the rules and the parameters of the model are closely related with the empirical data from Blogs and Diggs.
Agent-based modeling [27,28,4], where different properties of agents influence their actions, provides suitable theoretical framework for numerical simulations of social phenomena. Recently a model for product-review with the emotional agents in a mean-field environment has been introduced [29], with the agents emotional states described by two state variables (a i , v i ). These variables correspond to the psychological values of the arousal and the valence, respectively, in view of the Russell’s two dimens
This content is AI-processed based on open access ArXiv data.
