Order this book
CRESS, University of Surrey
A wide range of topics and methods is covered throughout the 53 papers collected in this volume. In the introduction, the editors grouped the authors' interests into seven categories: "social networks", "modelling", "machine learning and data mining", "social behaviours", "public health", "cultural aspects", and finally "effects and search". This is an unworkable taxonomy - given the juxtapositions we are faced with here. But the main problem is that the volume itself is not structured along these classifications - an unfortunate slip given the length of the book. More important is the manifest involvement of social scientists in social computing, and the productive collaboration between them and natural scientists and engineers. Several papers are either written or co-authored by psychologists, sociologists and economists. This is a rising trend that consolidates the importance of social computing and mechanism-based explanations of social phenomena across disciplines.
To provide a thorough discussion of the volume content is an impossible task. Most of the papers I highlight deal with agent-based simulations and social network analysis, particular interests of mine. In order to structure this review, I distinguish among complex thought experiments, behavioural models oriented towards policymaking, and empirically calibrated models.
Complex thought experiments are those computational models in which researchers test whether some proposed mechanisms bring about the macrostructures of interest. Although informative, these models do not incorporate data to falsify them. What they provide is a computational demonstration that a given mechanism is in fact sufficient to generate the observed pattern. In this volume, Lingzhi and colleagues' agent-based model is an example. They study the effect of leaders in ethno-religious conflicts. In their model, agents are embedded in a social network and play the iterated prisoner's dilemma. Their results indicate that even when a high fraction of the population is willing to cooperate with other ethnic groups, if the leaders were not willing to cooperate, there will be a high potential of intergroup conflict.
Behavioural models oriented towards policymaking are also included in this volume. Navarro-Barrientos and colleagues' system dynamic model shed light on the necessary conditions in which an intervention to reduce obesity might succeed. Hu and Puddy's artificial society simulates the dynamics of child maltreatment and child maltreatment prevention, using a social network approach. Chen and colleagues investigate the coevolution of individual behaviour, social network structures and viral epidemics. Their model suggests two patterns. Firstly, a limited supply of antiviral drugs can be distributed optimally between the hospitals and the market. Thus, when 40% of the stockpile is assigned to hospitals, the number of infected agents reaches its minimum. After that, there is no decrease from a larger hospital allocation. Secondly, the model allows the authors to isolate the effects of changes in individual behaviour caused by the fear of getting infected, and the effects of changes in the social contact network, on the spread of the disease. The results show that dependence of demand on epidemic prevalence postpones the epidemic peak and that changes in social networks caused by isolation measures reduce the peak of the epidemic curve.
Contrary to complex thought experiments, such as those reviewed above, empirically calibrated models take real-world data into account. Modellers use statistics either to gauge the generative sufficiency of the proposed mechanisms or to adjust the simulation parameters. The volume contains several papers in this vein. However, most of them apply pattern recognition methods, with machine learning algorithms, Bayesian inferences and neural networks as the preferred techniques. Just a few of them suggest mechanism-based explanations.
Liberman and Alt advocate for empirically calibrated artificial societies by using survey data, such as the World Values Survey. They model a social network in which the strength of the connections is given by the degree of similarity among the agents. Similarity is established by sociodemographic and cultural variables in that database. Consequently, there is a one-to-one ratio of agents to original survey respondents. The problem with this exercise lies in the scientific validity of the enterprise: it is problematic to extract conclusions from a network built upon a database that does not include social network information. Lussier and colleagues study the relationship between social networking and user-generated content. More than six million friend relationships and half a million users in the social news website "digg.com" are analysed. Two findings are particularly worth mentioning. First, individuals' consumption habits influence their friend networks, but as well there are effects on the extended network (i.e., friends of friends). Second, the level of reciprocity is highly imbalanced, meaning that person A "diggs" person B's stories far more frequently than B "diggs" A's stories. The authors hypothesise that this pattern, uncommon in offline networks, is critical to the intended function of the site, as non-reciprocity helps generate the "buzz" associated with popular articles.
In synthesis, this volume offers such a wide range of methods and applications, that anyone involved in social computing will find many of these papers stimulating. Given the quality of the contributing authors, this volume stands as a strong statement about the current place of social computing.
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2011