Human Behavior Understanding for Inducing Behavioral Change: Social and Theoretical Aspects Bruno Lepri1,2 , Albert Ali Salah3 , Fabio Pianesi1 , and Alex Sandy Pentland2 1 FBK, via Sommarive 18, Povo, Trento, Italy {lepri,pianesi}@fbk.eu 2 MIT Media Lab, 20 Ames Street, 02-139 Cambridge, MA, USA pentland@mit.edu 3 Bo˘gazi¸ci University, Department of Computer Engineering, Istanbul, Turkey salah@boun.edu.tr Abstract. The 2nd International Workshop on Human Behavior Un- derstanding (HBU’11) focuses on inducing behavioral change via com- puter systems that can analyse human behavior and communicate per- suasive messages accordingly. While analysis techniques that involve pat- tern recognition, signal processing and machine learning are very rele- vant to this aim, the underlying psychological and sociological aspects of inducing behavioral change cannot be neglected. This paper provides a framework for assessing the impact of social factors for these applications, and discusses the role of social mediation of behaviors and attitudes. 1 Introduction People routinely engage in relationships whereby they influence and are influ- enced by other humans but they are just starting confronting with machines that have this capability, due to the recent attempts at building persuasive systems. Most of the research on persuasive technologies is comprised under the umbrella of the term ‘captology’ [20], which refers to the study of machines designed to influence people’s attitudes and behaviors. A notable difference between persua- sion in human-human interaction as opposed to human-machine interaction, is the limited (if any) resort of machines to real-time understanding of people’s individual traits, activities and social dynamics. As a consequence, most of the current persuasive systems lack flexibility and cannot personalize and adapt their message to the broader context the target person(s) is (are) in.4 4 R. Wichert, K. van Laerhoven, J. Gelissen (eds.) AmI 2011 Workshops, CCIS 277, pp.252-263, 2012. Copyright by Springer-Verlag Berlin Heidelberg. This is the unedited author’s proof, please do not distribute.  The automatic analysis of human behavior, in turn, has been progressing in the last few years, thanks to the shared awareness that computer systems can provide better and more appropriate services to people only if they can understand much more about their attitudes, preferences, personality, social re- lationships etc., as well as about what people are doing, the activities they have been engaged in the past, their routines and lifestyles. This paper proposes a set of social and theoretical perspectives and consider- ations to develop systems and applications that rely on human behavior analysis for inducing behavioral change. These systems have the potential to re-define the relationship between the computer and the human, moving the computer from a passive observer role to a socially active one and enabling it to drive inter- actions that influence the attitudes and behaviors of people in their everyday environments [51]. In a quite general sense, the goal of this paper is to contribute to further advancing ubiquitous information societies where computers and humans are part of one and the same ecosystem. One crucial property of entities living in the same ecosystem is that they mutually influence and affect each other’s behavior, as well as internal states (e.g., attitudes) in a host of different ways and through varied means, including implicit and indirect ones whose mechanisms are not necessarily fully grasped by the target. It seems to us that there is still a good deal of reluctance in facing the puzzling possibility that, in order to approach autonomous human-like behavior, machines must, among other things, also be able to use different means to affect human behaviors and attitudes. Yet, this possibility is intrinsic to the overall vision of machines as autonomous agents in constantly changing environments, which lies at the heart of research areas such as artificial intelligence, cognitive systems, embedded systems, ambient intelligence, pervasive and ubiquitous computing, and so on. Such capabilities are also an essential ingredient of applications that aim to turn those technological and scientific advances into valuable services for the users. This paper is structured as follows: Section 2 proposes a possible approach to social influence and computer-induced behavioral change. Section 3 deals with the possibility of endowing machines with the skills needed to social percep- tion, and underlines the importance of personality assessment. Section 4 shortly discusses the sensing modalities for social behavior, and provides pointers for fur- ther reading. Section 5 discusses the technological, scientific and societal impact of the discussed framework. Finally, the last section draws our conclusions. 2 Bringing about change: perspectives on social influence In an attempt to lay the foundations of a theory for computer-induced human change, we will operate under several assumptions that incidentally set this work apart from most current efforts in the same direction. First of all, our focus is on the modifications of the social structure and dynamics of small and large groups (friends, colleagues, families, students, and so on) and on changes in individuals (behaviors and attitudes) that occur because of their membership in social en- tities. Social settings are, in most respects, more challenging than those based on a single individual because of the dynamic and bidirectional individual-group relationships: in many respects, understanding a group’s characteristics implies understanding the characteristics of its members, and social change implies in- dividual change. In social psychology, researchers study the psychological processes involved in persuasion, conformity, and other forms of social influence, but they have seldom modeled the ways influencing unfolds when multiple sources and multi- ple targets interact over time [43]. On the other hand, researchers in sociology, economics, network science and physics have developed models of influence flow in populations and groups without relying on any detailed understanding of the participating individuals. For example, the social diffusion phenomenon, in which a behavior spreads over a social network, is explained by a mechanism of behav- ioral cascading whereby the probability for a group member to adopt a behavior is affected by the adoption behavior of the other group members. In many as- pects, this approach is similar to a popular model of spreading epidemics: subject X adopts the same behavior as the other group members if his/her exposure to it exceeds a given threshold [10, 13, 24, 57]. Obviously, at this level of modeling, details of individuals are neglected. Recently, some proposals have been advanced to incorporate a detailed micro- level understanding of influence processes derived from social psychology within the broader picture of multidirectional, dynamic influences typical of social net- work studies [43, 21]. For example, Friedkin proposed to merge social-psychological approaches to the attitude-behavior link with the behavioral cascades diffusion models [21]. The attitude-behavior link, in turn, is accounted for by means of the Theory of Planned Behavior (TPB) [2]. In a simplified form, TPB main- tains that actual behavior is explained by behavioral intentions that, in turn, are influenced by (i) the specific attitudes toward that behavior (e.g., attitude towards smoking); (ii) subjective norms (beliefs concerning how the people one cares about view the behavior in question); and (iii) perceived behavioral control (e.g., whether people think it will be easy for them to stop smoking). Once the relationships from attitudes to behavior is accounted for, the re- verse link, from behavior back to attitudes, can be modeled using, e.g., the Self Perception Theory (SPT) [8]. According to SPT, the behavior-attitude link is activated in situations where people do not already have clear ideas about their own attitudes, so that they rely on external observations to infer about then. For instance, STP’s view of the behavior-attitude link fits well situations in which individuals’ attitudes are not yet well formed. Another possible shift in the conception of computers as actors of social change is the idea that they behave as a sort of peripheral device, exploiting different kinds of minimalist strategies to bring about change through the small- est amount of human-computer interaction. Minimalist strategies are motivated by the desire that even in the presence of a change inducing system, the main activity of people (their ‘primary task’) remain that of interacting with other people (e.g. friends, colleagues, relatives, and so on). Minimalist strategies for change could exploit the accumulating evidence about behaviors occurring au- tomatically and without conscious effort [5, 11]. For instance, people who were unconsciously exposed (primed) to the stereotype of an elderly person walked slower than a control group [6] and participants listening to male-typical words while driving drove faster than participants listening to neutral words [54]. Im- portantly, direct investigations about the participants’ perception of their own behavior revealed that they were not aware of these induced changes. Because of its pervasiveness, the so-called direct perception-behavior link [11] can be exploited to bring about desired behavioral changes by means of prim- ing procedures (based on gender-related, cultural, age-related, etc., stereotypes) and/or by facilitating mimicry, the tendency to mimic other individuals’ behav- iors without awareness or intent. One might also venture that at least some of the phenomena usually addressed under the rubric of behavioral cascades and social contagion (i.e. the propagation of behaviors and attitudes in a socially mediated manner) can be profitably addressed by means of the concept of mimicry and of the perception-behavior link. The idea of agents that pursue change through indirect and minimalist strate- gies is also related to the peripheral route to persuasion of the Elaboration Like- lihood Model (ELM) [46]. This often-discussed socio-psychological model of be- havioral/attitudinal change posits two ways in which attitudinal and behavioral change can take place: a central route and a peripheral route, respectively. The central route requires that the influencee attentively attends to an argumenta- tive communication and thinks about the arguments presented. The efficacy of this communication depends on the coherence, logic, and depth of the arguments presented, as well as the knowledge and reasoning abilities of the receiver. The peripheral route, in turn, involves strategies that influence people by means of peripheral cues, e.g., the status of the communication source, its attractiveness and credibility, etc. Some recent works have attempted at exploring these ideas by devising sys- tems that present members of a working group with information about their own social behavior, with the goal of changing it and making it more conducive to better group dynamics [16, 33, 47, 56]. Moreover, an interesting and recent work [1] tested an experimental intervention for inducing changes in fitness- related, physical activities and habits. The intervention was based on a novel social mechanism in which subjects were rewarded based on their peers’ perfor- mance rather than their own. The results suggested that (i) social factors have an effect on physical activity behavior over time, (ii) social incentives strengthen social influence among subjects, and (iii) a phenomenon similar to contagion emerges as it relates to the pre-existence of social ties among the subjects. This work is a good example of the way two of the general concepts discussed so far can work to drive behavioral chenges: (i) the importance of social factors in inducing individual changes and (ii) the effectiveness of strategies based not on argumentation and logical reasoning but on mimicking and learning through imitation of the others’ behavior. 3 Understanding social behavior From the discussions of the previous section, two major dimensions emerge that characterize change in our framework: (i) social vs. individual change and (ii) attitudinal vs. behavioral change. Regarding the first dimension, we pursue both social and individual change, with the latter addressed through the mediation of the group. It therefore pops up with all its importance the goal of endowing behavior-inducing systems with the ability to understand social behaviors and social relations. It is impossible to mention here the many cognitive and social psychology theories that have been formulated to account for human social behavior. It is more important to point out that some of these theories are already providing the backbone to computational models and are going to be used in change-inducing systems. A telling example in this respect are dominance and other dimensions related to social verticality [22]. Dominant behavior, in fact, is a key determi- nant of a group’s social structure and dynamics [7]. Quite straightforwardly, the recognition of dominant people could be useful to decide about the right sub- jects to address in order to maximize the chance of success of group persuasion attempts. Following the lead of much work in social psychology [25, 44], computer sci- entists focused their attention on studying dominance and role-based status cues [27, 28, 31, 49, 50] and functional social roles [59, ?] in small-group, task- oriented meetings, using audio-visual nonverbal Another important piece of knowledge that can be used to to build effective change-inducing systems is personality: people, in fact, react differently to per- suasive stimuli according to their personality. For instance, studies about the relationship between the trait of self-esteem and susceptibility to social influ- ence have reported that people with low self-esteem are most easily influenced and those with high self-esteem are much less so [29, 30]; other works found that people with medium self-esteem values are most open to influence [26, 14]. Locus of Control, too, has been found to empirically relate to social influence susceptibility. Internals, who believe they control their behavioral outcome dur- ing their lifetime, seem to be more resistant to social influence than externals, who attribute their behavioral outcome to external factors such as fate, luck, or powerful others [26, 14]. Moreover, the relative efficacy of one or the other persuasive route in the ELM model might also depend on the personality of the target subject: for instance, people who are deemed as “in need for cognition”, a trait indicating that people enjoy thinking about complex problems [9], are more amenable to persuasion through the central route than people who score low in such a trait [9, 46]. Conversely, people with low need for cognition are more susceptible to be persuaded using unconscious stimuli (e.g. mimicry and priming strategies). On the computational side, several works have started exploring the auto- matic recognition of personality, mostly targeting the Big Five model [41, 42, 37, 36, 47, 12] and/or traits such as the Locus of Control [36, 47], using differ- ent sources of behavioral data, e.g. visual and acoustic cues and mobile phone data. All these approaches to the automatic recognition of personality more or less implicitly share the so-called ‘person perspective’ on personality [19]: for a given behavioral sample, they attempt to classify whether that sample belongs to an extrovert or introvert (or equivalently, to a neurotic or an emotionally stable person, and so on). The problem with this approach is that it assumes a direct and stable relationship between, e.g., being extravert and acting extravert- edly. Extraverts, however, can sometimes act introvertedly, while introverts can at times exhibit extraverted behaviors. Similarly, people prone to neuroticism need not always exhibit anxious behavior, while agreeable people can sometimes be aggressive. While the person perspective has often dismissed these fluctua- tions of actual behavior as statistical noise, it has been recently suggested by Fleeson [19] that they can be meaningful. The social psychology literature has coined the term personality states to refer to concrete behaviors (including ways of acting, feeling and thinking) that can be described as having a similar con- tent to the corresponding personality traits. In other words, a personality state describes a specific behavioral episode wherein a person behaves more or less introvertedly/extravertedly, more or less neurotically, etc. Personality can then be reconstructed through density distributions over personality states, with pa- rameters such as means, standard deviations, etc., summarizing what is specific to the given individual. Recently, this paradigm shift has started being consid- ered by computational approaches, as by Staiano et al. who investigated the automatic recognition of personality states in small group meetings [55]. Regarding the second dimension, a change inducing system could aim to modify individual minds by changing their attitudes towards the relevant issues (or person, group of people, etc.). It might also try to change people’s behavior directly. We note that the distinction between mental and behavioral modifica- tions is not unlike the differences in effect durations; mental modifications last longer than those that affect only behavior. Recently, some works started to model behavioral and attitudinal changes in individuals [38, 40, 39]. In [40], the authors described the use of mobile phones to model and understand the link between exposure to peers and weight gain among a group of undergraduate students, where a positive correlation was es- tablished between the change in an individual’s Body Max Index (BMI) with face-to-face exposure to social contacts who themselves gained weight. Along the same direction, Madan et al. [39] measured the spread of political opinions (republicans vs. democrats) during the 2008 US presidential election campaign to model specific behaviors and changes in political opinions. 4 Sensing social signals The previous section showed that we can endow machines with the skills to understand and predict human social behaviors, individual characteristics and attitudes. In this section, we reviews ways we can provide them with the ability of perceiving and sensing the social signals upon which such an understanding can be built. Research on first impression formation has demonstrated that even when observing a person in social interactions for only a short amount of time and even without knowing him/her, we are capable to accurately assess aspects of his/her personality [3, 23], emotions, motives and intentions, cognition, future behavior and the types of social relationship he/she is used to get involved in [53, 25]. This quite precise “zero-acquaintance” appreciation (or rapid cognition) of the internal properties of another person is based on short sequences of expressive behavior called “thin slices of behavior” [4], is almost automatic and largely exploits nonverbal behavior [3, 32], such as posture, position, facial expressions, location, prosodic features, and so on. There is encouraging evidence that computers can be made capable of ex- ploiting thin slices of behavior to detect individual traits such as personality [35, 47] and dominance [27, 28, 31], group properties like social roles [17], interactions’ outcomes [15, 36], etc. Many of these works have employed so-called ‘honest sig- nals’ - social signals that, being too difficult for humans to control, can provide a reliable source of information about socially relevant aspects [58, 45]. In [51], the authors have discussed the different modalities used to sense sig- nals pertaining to individual behavior or social behavior. Traditional sensors like cameras and microphones provide valuable information, yet sensing social data almost invariably is more challenging than sensing the behavior of a single indi- vidual, and temporal dynamics is found to be of the utmost importance for social signals [52]. For instance turn-taking behavior, interruptions, relative sound en- ergy, usage of gaze and attention, are all found to be relevant in transmitting status and dominance [22]. Recently, the rich sensor sets installed in smartphones have been harnessed to transform the smartphone into veritable data mines for assessing social inter- action information. Not just the usage patterns of these phones (as in the reality mining study of Eagle and Pentland [18]), but also proximity and communication data have been found to be useful [38, 40, 39]. Other types of sensors have been packaged into wearable units, the so-called ‘sociometric badges’, to specifically target small-group interactions [34]. These developments allow for a cost-effective, real-time, extensive, and (mostly) unobtrusive monitoring of human social signals, and open up the way to the im- plementation of new socially sensing machines. 5 Technological, scientific and societal impact The framework proposed here contributes to transform science, technology and societies in several different ways, and it is difficult to underestimate the societal impact of computer systems that induce behavioral change. Such systems can contribute to enforce individual and social values by helping individuals reach target behaviors through interaction and immersion. Apart from technological challenges, these systems require careful preservation of privacy and the main- tenance of appropriate levels of control by the users. The enormous potential impact is partly due to the ubiquity and informality of such systems; they can be everywhere and can function through subtle and inconspicuous influences. It is also due to the richness of the application domain, as these technologies stand to transform vital sectors such as healthcare, organizational and individ- ual psychology, management, education, entertainment, commercial advertising, politics, and many more. The societal impact will not be limited to the offer of new and revolutionary technologies, as the relationship between systems and users will change in fundamentally different ways. Once the ethical and legisla- tive framework is set in place, researchers from different fields will have access to first hand, empirical data on how computer-induced change works in realistic contexts, as well as models and paradigms to initiate new projects of behavioral change. The marriage between social sciences that systematically chart out principles of human behavior and computer-related disciplines (such as multimodal analy- sis, signal processing, pattern recognition, machine learning) that aim to provide computer systems with the ability of automatically analyzing behavior, can pro- duce long-lasting effects on both sides. Social psychology and social sciences have worked out much of the theoretical foundations for the issues discussed in this work. However, present approaches like TPB, thin slices in rapid social cognition, the perception-behavior link, peer pressure, social contagion, etc., are mostly descriptive frameworks that lack formal and computational modeling. The framework proposed in this paper is mostly geared towards laying the foun- dations of computational modeling of psycho-social and sociological concepts and theories with the goal of actually incorporating them in working real-world systems. The combination of psychology and technology at this level is a major de- parture from the tradition and practice of captology. Although the Computer as a Social Actor (CASA) framework [48] has convincingly argued that we tend to attribute computers characteristics that are typically human, it has done so in quite restricted scenarios, mostly limited to single computer-single human set- tings. The tradition of captology has taken the media equation and CASA as enabling factors for computer-induced change. By exploiting concepts such as the peripheral route to attitude, the automaticity of behavior, the peer pressure of cascade models, we depart from this tradition, requiring forms of sociality in the human-computer relationship, and subsequently filling a gap in our under- standing of the social/non-social continuum in human-machine interaction. 6 Concluding remarks Throughout the paper, we have been stressing the importance of social and personality factors for building systems that interact with individuals in ways that would result in behavioral change. When we design ‘persuasive’ systems, the dialectic nature of human-computer interaction cannot be ignored, and we converge, in many respects, onto models of human-human interaction, i.e. the twin domains of psychology and sociology. The influencing of a single person can happen via different routes. It is pos- sible to directly influence a certain behavior, or to cause changes in an attitude that will in time result in the modification of a certain behavior, or start a be- havioral cascade, where the behavioral change in the individuals social circle (for instance through the dominant people in the group) will eventually cause the individual to adapt to its environment by accepting the same behavior change. The persuasive message can require conscious processing and elaboration, or it may be subliminal and peripheral. We have cited many empirical studies, which establish that there is no general methodology that can be adopted for persua- sion, but each individual will, depending on his or her personality, his or her context and other such factors, be influenced by different strategies to differing extents. A point of consideration is that even very short expressions of behavior can be extremely useful in selecting the appropriate strategies. These cues can give us the key to detecting dominant people in groups or personality traits of interacting individuals, thereby laying the foundation for inducing behavioral change. Old and new sensor technologies are combined in innovative ways to capture these social signals. It is our firm belief that the resulting systems will have a significant impact on the confluence of humans and computational systems. 7 Acknowledgments Bruno Lepri’s research is funded by PERSI project inside the Marie Curie CO- FUND – 7th Framework. References 1. Aharony, N., Pan, W., Ip, C., Khayal, I., Pentland, A.: Social fMRI: Investigating and shaping social mechanisms in the real world. Pervasive and Mobile Computing (2011) 2. Ajzen, I.: From intentions to actions: A theory of planned behavior. In: Kuhl, J., Beckmann, J. (eds.) Action Control, from Cognition to Behavior. pp. 11–39. Springer Verlag (1985) 3. Ambady, N., Hallahan, M., Rosenthal, R.: On judging and being judged accu- rately in zero-acquaintance situations. Journal of Personality and Social Psychol- ogy 69(3), 518–529 (1995) 4. Ambady, N., Rosenthal, R.: Thin slices of expressive behavior as predictors of interpersonal consequences: A meta-analysis. Psychological Bulletin 111(2), 256– 274 (1992) 5. Bargh, J.: The automaticity of everyday life. Lawrence Erlbaum Associates Pub- lishers (1997) 6. Bargh, J., Chen, M., Burrows, L.: Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of Personality and Social Psychology 71(2), 230 (1996) 7. Barsade, S.: The ripple effect: Emotional contagion and its influence on group behavior. Administrative Science Quarterly pp. 644–675 (2002) 8. Bem, D.: Self-perception theory. Advances in Experimental Social Psychology 6, 1–62 (1972)  9. Cacioppo, J., Petty, R.: The need for cognition. Journal of Personality and Social Psychology 42(1), 116–131 (1982) 10. Centola, D., Macy, M.: Complex contagions and the weakness of long ties. American Journal of Sociology 113(3), 702 (2007) 11. Chartrand, T., Bargh, J.: The chameleon effect: The perception–behavior link and social interaction. Journal of Personality and Social Psychology 76(6), 893 (1999) 12. Chittaranjan, G., Blom, J., Gatica-Perez, D.: Who’s who with big-five: Analyzing and classifying personality traits with smartphones. In: Proc. 15th Annual Int. Symp. on Wearable Computers (ISWC) (2011) 13. Christakis, N., Fowler, J.: The collective dynamics of smoking in a large social network. New England Journal of Medicine 358(21), 2249–2258 (2008) 14. Cox, D., Bauer, R.: Self-confidence and persuasibility in women. Public Opinion Quarterly pp. 453–466 (1964) 15. Curhan, J., Pentland, A.: Thin slices of negotiation: Predicting outcomes from conversational dynamics within the first 5 minutes. Journal of Applied Psychology 92(3), 802–811 (2007) 16. DiMicco, J., Bender, W.: Group reactions to visual feedback tools. Persuasive Technology pp. 132–143 (2007) 17. Dong, W., Lepri, B., Cappelletti, A., Pentland, A., Pianesi, F., Zancanaro, M.: Using the influence model to recognize functional roles in meetings. In: Proc. 9th Int. Conf. on Multimodal interfaces. pp. 271–278. ACM (2007) 18. Eagle, N., Pentland, A.S., Lazer, D.: Inferring friendship network structure by using mobile phone data. Proc. National Academy of Sciences 106(36), 15274– 15278 (2009) 19. Fleeson, W.: Toward a structure-and process-integrated view of personality: Traits as density distributions of states. Journal of Personality and Social Psychology 80(6), 1011–1027 (2001) 20. Fogg, B.: Persuasive technologies. Communications of the ACM 42(5), 27–29 (1999) 21. Friedkin, N.: The attitude-behavior linkage in behavioral cascades. Social Psychol- ogy Quarterly 73(2), 196–213 (2010) 22. Gatica-Perez, D.: Automatic nonverbal analysis of social interaction in small groups: A review. Image and Vision Computing 27(12), 1775–1787 (2009) 23. Gifford, R.: A lens-mapping framework for understanding the encoding and decod- ing of interpersonal dispositions in nonverbal behavior. Journal of Personality and Social Psychology 66(2), 398–412 (1994) 24. Granovetter, M.: Threshold models of collective behavior. American Journal of Sociology pp. 1420–1443 (1978) 25. Hall, J., Coats, E., LeBeau, L.: Nonverbal behavior and the vertical dimension of social relations: A meta-analysis. Psychological bulletin 131(6), 898 (2005) 26. Harvey, O., Consalvi, C.: Status and conformity to pressures in informal groups. The Journal of Abnormal and Social Psychology 60(2), 182–187 (1960) 27. Hung, H., Jayagopi, D., Yeo, C., Friedland, G., Ba, S., Odobez, J., Ramchandran, K., Mirghafori, N., Gatica-Perez, D.: Using audio and video features to classify the most dominant person in a group meeting. In: ACM Multimedia. vol. 3. Citeseer (2007) 28. Hung, H., Jayagopi, D., Ba, S., Odobez, J., Gatica-Perez, D.: Investigating auto- matic dominance estimation in groups from visual attention and speaking activity. In: Proc. 10th Int. Conf. on Multimodal interfaces. pp. 233–236. ACM (2008) 29. Janis, I.: Personality correlates of susceptibility to persuasion. Journal of Person- ality 22(4), 504–518 (1954) 30. Janis, I.: Anxiety indices related to susceptibility to persuasion. The Journal of Abnormal and Social Psychology 51(3), 663 (1955) 31. Jayagopi, D., Hung, H., Yeo, C., Gatica-Perez, D.: Modeling dominance in group conversations using nonverbal activity cues. Audio, Speech, and Language Process- ing, IEEE Transactions on 17(3), 501–513 (2009) 32. Kenny, D., Horner, C., Kashy, D., Chu, L.: Consensus at zero acquaintance: Repli- cation, behavioral cues, and stability. Journal of Personality and Social Psychology 62(1), 88–97 (1992) 33. Kim, T., Chang, A., Holland, L., Pentland, A.: Meeting mediator: enhancing group collaborationusing sociometric feedback. In: Proc. 2008 ACM Conf. on Computer supported cooperative work. pp. 457–466. ACM (2008) 34. Kim, T., Olgu´ın, D., Waber, B., Pentland, A.: Sensor-based feedback systems in organizational computing. In: Proc. Int. Conf. on Computational Science and En- gineering. vol. 4, pp. 966–969. IEEE (2009) 35. Lepri, B., Mana, N., Cappelletti, A., Pianesi, F.: Automatic prediction of individual performance from thin slices of social behavior. In: Proc. 17th ACM Int. Conf. on Multimedia. pp. 733–736. ACM (2009) 36. Lepri, B., Mana, N., Cappelletti, A., Pianesi, F., Zancanaro, M.: Modeling the personality of participants during group interactions. User Modeling, Adaptation, and Personalization pp. 114–125 (2009) 37. Lepri, B., Subramanian, R., Kalimeri, K., Staiano, J., Pianesi, F., Sebe, N.: Em- ploying social gaze and speaking activity for automatic determination of the ex- traversion trait. In: Int. Conf. on Multimodal Interfaces and the Workshop on Machine Learning for Multimodal Interaction. pp. 7:1–7:8. ICMI-MLMI, ACM, New York, NY, USA (2010) 38. Madan, A., Cebrian, M., Lazer, D., Pentland, A.: Social sensing for epidemiological behavior change. In: Proc. 12th ACM Int. Conf. on Ubiquitous Computing. pp. 291–300. ACM, New York, NY, USA (2010) 39. Madan, A., Farrahi, K., Gatica-Perez, D., Pentland, A.: Pervasive sensing to model political opinions in face-to-face networks. In: Proc. 9th Int. Conf. on Pervasive Computing. pp. 214–231. Pervasive’11, Springer-Verlag, Berlin, Heidelberg (2011) 40. Madan, A., Moturu, S.T., Lazer, D., Pentland, A.S.: Social sensing: obesity, un- healthy eating and exercise in face-to-face networks. In: Wireless Health 2010. pp. 104–110. WH ’10, ACM, New York, NY, USA (2010) 41. Mairesse, F., Walker, M.: Automatic recognition of personality in conversation. In: Proceedings of the Human Language Technology Conference of the NAACL. pp. 85–88. Association for Computational Linguistics (2006) 42. Mairesse, F., Walker, M.: A personality-based framework for utterance generation in dialogue applications. In: Proc. of the AAAI Spring Symposium on Emotion, Personality and Behavior (2008) 43. Mason, W., Conrey, F., Smith, E.: Situating social influence processes: Dynamic, multidirectional flows of influence within social networks. Personality and Social Psychology Review 11(3), 279 (2007) 44. Mast, M.: Dominance as expressed and inferred through speaking time. Human Communication Research 28(3), 420–450 (2002) 45. Pentland, A.: Honest signals: how they shape our world. The MIT Press (2008) 46. Petty, R., Cacioppo, J.: The elaboration likelihood model of persuasion. Advances in Experimental Social Psychology 19(1), 123–205 (1986) 47. Pianesi, F., Zancanaro, M., Not, E., Leonardi, C., Falcon, V., Lepri, B.: Multimodal support to group dynamics. Personal and Ubiquitous Computing 12(3), 181–195 (2008) 48. Reeves, B., Nass, C.: The Media Equation: How people treat computers, television, and new media like real people and places. CSLI Publications and Cambridge University Press (1996) 49. Rienks, R., Heylen, D., van der Weijden, E.: Argument diagramming of meeting conversations. In: Multimodal Multiparty Meeting Processing, Workshop at the 7th Int. Conf. on Multimodal Interfaces. pp. 85–92 (2005) 50. Rienks, R., Zhang, D., Gatica-Perez, D., Post, W.: Detection and application of influence rankings in small group meetings. In: Proc. 8th Int. Conf. on Multimodal Interfaces. pp. 257–264. ACM (2006) 51. Salah, A., Lepri, B., Pianesi, F., Pentland, A.: Human behavior understanding for inducing behavioral change: Application perspectives. In: Salah, A.A., Lepri, B. (eds.) HBU 2011. LNCS, vol. 7065. pp. 1–15. Springer, Heidelberg (2011) 52. Salah, A., Pantic, M., Vinciarelli, A.: Recent developments in social signal process- ing. In: Proc. IEEE Int. Conf. on Systems, Man, and Cybernetics (2011) 53. Schmid Mast, M., Murphy, N., Hall, J.: A brief review of interpersonal sensitivity: Measuring accuracy in perceiving others. pp. 163–185. Trinidad: University of the West Indies Press (2006) 54. Schmid Mast, M., Sieverding, M., Esslen, M., Graber, K., J¨ ancke, L.: Masculinity causes speeding in young men. Accident Analysis & Prevention 40(2), 840–842 (2008) 55. Staiano, J., Lepri, B., Kalimeri, K., Sebe, N., Pianesi, F.: Contextual modeling of personality states’ dynamics in face-to-face interactions. In: Proc. IEEE Social Computing (2011) 56. Sturm, J., Herwijnen, O., Eyck, A., Terken, J.: Influencing social dynamics in meetings through a peripheral display. In: Proc. 9th Int. Conf. on Multimodal Interfaces. pp. 263–270. ACM (2007) 57. Watts, D., Dodds, P.: Influentials, networks, and public opinion formation. Journal of Consumer Research 34(4), 441–458 (2007) 58. Zahavi, A., Zahavi, A., Balaban, A., Ely, M.: The handicap principle: a missing piece of Darwin’s puzzle. Oxford University Press, USA (1999) 59. Zancanaro, M., Lepri, B., Pianesi, F.: Automatic detection of group functional roles in face to face interactions. In: Proc. 8th Int. Conf. on Multimodal Interfaces. pp. 28–34. ACM (2006)