User Modeling for Pervasive Alcohol Intervention Systems Ugan Yasavur Reza Amini Christine Lisetti School of Computing and School of Computing and School of Computing and Information Sciences Information Sciences Information Sciences Florida International University Florida International University Florida International University Miami, FL Miami, FL Miami, FL uyasa001@fiu.edu ramin001@fiu.edu lisetti@cs.fiu.edu ABSTRACT Explicit modeling is generally used in the initial user pro- In this paper, we have proposed a user model for com- file creation stage and does not require continuous updates. puter based drinking behavior change intervention and rec- Implicit modeling facilitates the maintenance of context- ommender systems. We discuss specific requirements of related variables in order to increase the context-awareness user modeling in health promotion and specifically alco- (e.g. users’ physical and social environments) of the system. hol interventions. We believe that making behavior change After initial creation of a user profile, context-related and systems available pervasively may lead to better and sus- affective features need to be kept up-to-date and other pro- tainable results. Therefore, our proposed user model takes file features must be updated less frequently. advantage of the target-behavior related features such as We focus on one target behavior, namely alcohol consump- contextual features (e.g., social interactions, location, and tion related behavior change. Therefore, our proposed user time). The proposed user model uses well-validated ques- model targets lifestyle change systems which aim to promote tionnaires to capture target-behavior specific aspects. We decreasing or stopping alcohol consumption. also introduced approaches for enhancing users’ experience In the following sections, first we study the state of the in the model creation stage by using Embodied Conversa- art in user modeling in life style change recommender sys- tional Agents(ECAs) and users’ affective states. tems and behavior change intervention systems in Section 2. Then, we extend the explicit(target-behavior specific) and implicit(target-behavior related) features to build and Keywords maintain a user model in Section 3. User modeling, tailoring, alcohol intervention, behavior change, lifestyle change recommender systems (LSCRS). 2. RELATED RESEARCH 1. INTRODUCTION Personalization and tailoring are used in variety of differ- ent domains including e-commerce[24], social networks [35], The positive effect of tailoring and personalization on lifestyle entertainment [18] [7] and health [26] [27]. Whereas collab- change systems is evidenced by several studies [20] [33] [34]. orative and content-based recommender systems provide a For effective tailoring in lifestyle change systems, compre- good level of personalization in e-commerce, social networks hensive user characteristics and personal profile/model re- and entertainment domains, the behavior change domain re- lated to the target behavior need to be acquired and main- quires a different approach. The demographic information, tained. user interests, goals, background information and individual Explicit and implicit modeling is needed in healthy be- traits are the most commonly used user profile features in havior promotion systems. In addition, the user model for recommender systems. While these features are still useful health behavior change systems must be specialized accord- in health behavior change systems, different target behav- ing to a target behavior (e.g excessive drinking, lack of ex- iors requires different modeling features (e.g. consequences ercise, obesity). Explicit ways to create a user model or of drinking and dependence on alcohol for drinking behavior user profile may include conducting assessments with the use change; and family history and Body Mass Index (BMI) for of validated questionnaires, psychometric instruments and obesity). screening instruments. Implicit ways to build user-profile In addition to personal information, it is useful to ben- may include tracking motivation, stage of change, affective efit from research on context-aware systems [2]. By the features, spatio-temporal events and some data interpreta- increase in usage of smart mobile phones and mobile so- tion and mining. cial network applications, it has recently become possible to track context-related information about users. The most widely used features in context-aware systems are location [7] and time [18] [7]. It is also useful for health promotion recommender systems to use findings of context-aware sys- Paper presented at the Workshop on Recommendation Technologies for tems which focus on inferring users’ states and activities Lifestyle Change 2012, in conjunction with the 6th ACM conference on including social interactions [36] [32]. From continuously Recommender Systems. Copyright c 2012 for the individual papers by the posted data on social networks, it is possible to detect social papers’ authors. This volume is published and copyrighted by its editors. interaction [6]. Lifestyle@RecSys’12, September 13, 2012, Dublin, Ireland. Recently, there has been an increasing interest in user modeling based on affective features [4]. The user’s affective 3.1 Target-Behavior Specific Features states can be an indicator for the relevance of the recom- Our target behavior in this paper is alcohol drinking. So, mended item to the user’s interest. in this section we focus on the assessment instruments which In behavior change systems, personalization according to can capture specifically the user’s alcohol consumption be- affective state plays a particularly important role because havior features. The assessments used in this paper are delivering appropriate messages according to current emo- standardized assessment measures proved to be effective in tions of the user can increase the effectiveness of health pro- alcohol consumption behavior change [40]. motion interventions [25]. In the health intervention systems which use Embodied 3.1.1 Consequences of Drinking Conversational Agents(ECAs)[13] as a user interface, addi- “Drinking Consequences” feature set assesses the negative tional personalization can increase the efficacy of the inter- consequences of the user’s drinking. Drinker’s Inventory of vention system. Several studies show that concordance of Consequences (DrInC) [28] is a reliable, valid, clinically use- patient and physician increases patient satisfaction [15] [23]. ful, and self-administered instrument to assess the negative Also, related research on race concordance of the virtual consequences of drinking. DrInC includes a set of ques- character and the user implies that racial adaption of ECA tions in five different areas: physical, inter-personal, intra- and user has positive impact on user’s satisfaction [25]. personal, impulse control, and social responsibility. In the context of the computer-based alcohol interven- The user answers each question in a 4-point Likert scale. tions, although there exists some effort in web-based alcohol Then, by adding up the responses in each area, we calculate interventions for personalization and tailoring, they mainly his/her score in that area. These scores show the severity of focus on personalization of feedback for conducted assess- an individual’s problems. ments [9] [27], [19]. The recommender system can use these scores in order to While all mentioned interventions provide personalized prepare the best personalized feedbacks and recommenda- feedback, few of them [27], [19] provide feedback based on tions based on the consequences that alcohol has had on the theoretical constructs (e.g., Transtheoretical Model of Be- user’s life. According to the [28] this feature set should be havior Change). Drinker’s Check Up (DCU) [19] provides updated on weeks 1, 8, 16, 26, 52, and 68 of intervention. personalized feedback based on available normative data Intra-Personal: This feature is assessed using 8 ques- and uses elements of behavior change models. Responsi- tions which reflect the subjective perceptions of the user ble Drinking Program[27] makes further personalization by about her/his drinking. These questions query the user’s dynamically tailoring feedback across multiple interactions feeling experienced because of drinking (bad, unhappy, or of the client. Although the explicit information acquired guilty), personality change experiences (e.g. aggressive, de- from the users is only used for tailoring the feedback, these pressive), interference with personal growth, moral life, in- brief interventions provide good sources for target-behavior terests and activities, and interested lifestyle. specific user modeling. They do not focus on user model- Inter-Personal: The focus of this feature is to find out ing and personalization in the course of long term behavior the impact of drinking on the user’s relationships. So, we change period. query the user’s experiences of damage/loss of friendship/love, It has been concluded by several extensive surveys on al- impairment of parenting and causing harm to the family, cohol interventions [8] [43] that computer based interven- concern about drinking from family or friends, damage to tions have positive effect on reducing or stopping drinking. reputation, and embarrassing actions while drinking. The To maintain motivation and make the behavior change sus- assessment of this feature is performed using 10 questions. tainable, we can use behavior change support systems in the Social Responsibility: We use this feature to describe form of social networks, mobile applications, lifestyle change the role-fulfillment of the user from the other people’s point recommender systems, and motivational systems. of view. We use 7 questions to query the user’s work/school In the next section we discuss our proposed comprehen- problems (missing days, poor quality, fired or suspended), sive user model which can be used as a reference for alcohol financial problems, and failings to meet expectations. intervention systems and behavior change support systems. Physical: This feature is assessed using 8 questions that reflect the negative physical states resulting from user’s drink- 3. THE PROPOSED USER MODEL ing. These questions query the user’s hangovers, sleeping Our proposed user model is shown in Figure 1. The model problems, sickness, harm to health, appearance, eating habits, is updated after each assessment and after perception of new sexuality, and injury while drinking. affective and contextual features of the user. Assessments Impulse Control: This feature includes 12 questions provide information about different aspects of the client’s about other unhealthy lifestyles exacerbated by drinking (e.g., drinking. We use some well-validated [19] assessment in- smoking, drugs, and overeating), risk taking and impulsive struments to gain understanding of the user’s drinking psy- actions of the user, troubles with law, and damages to people chometric aspects. In addition to assessment results, it is and property. beneficial to monitor the user’s affective states via a camera to be able to adapt the recommendations and messages with 3.1.2 Motivation to Change the user’s affective states. To assess the stage of user’s readiness and motivation to The proposed user model is composed of features grouped change, we use an instrument called SOCRATES [31]. This under two categories, target-behavior specific features (ex- instrument involves 19 questions categorized in three do- plicit features) and target-behavior related features (implicit mains: ambivalence, recognition, and taking steps. Ques- features). In the following sections, we explain the impor- tions are answered in a 5-point Likert scale. A behavior tance of each feature and the aspects of the problematic change recommender system can use these scores to capture drinking behavior that each feature captures. the readiness of the user to change before providing recom- Figure 1: User Model mendations to change the user’s behavior change. use, additional life problems, motivation for treatment, and Recognition: The recognition score shows the degree of history of problem development. Information derived from the user’s awareness about his/her drinking problems, and this feature set can be used in selecting the treatment ap- the degree of his/her desire to change. Therefore, higher proaches for user [29] in the behavior change recommender degrees of this feature show more desire and motivation to systems. According to the BDP manual [30], the non-static change from the user. features of this group should be updated every three months. Ambivalence: Ambivalence score shows the degree of Age of Onset Problems: This feature involves the uncertainty of the user about whether s/he drinks too much, user’s age in which s/he first took a drink, the age in which is in control, is hurting others, or is alcoholic. A high am- s/he first became drunk, and the age in which drinking bivalence score shows openness of the user to change. A low started affecting his/her life. This feature is static and does ambivalence score has two possible reasons: (1) user knows not need updates later. that his drinking is causing problems (high Recognition); or Family History: This feature includes the alcohol prob- (2) user knows that s/he does not have drinking problems lem history of the person’s family. User can place his/her (low Recognition). family drinking in different categories of abstainer, light Therefore, we can use this feature to decide whether the drinker, moderate drinker, heavy drinker, problem drinker, user is open to reflections and recommendations or is not or alcoholic. If the user’s family does not have any drink- ready yet. ing history, it means that his/her drinking patterns were ac- Taking Steps: This feature shows the degree of the user’s quired, not inherited. To assess genetic risk factors, the alco- successful experience in changing drinking behavior. So, hol problems of his/her other biological relatives are queried high “Taking Steps” score can be interpreted as (1) need too. help to persist on the change behavior, and (2) need help Drug Use: Since using other drugs can increase the risk to prevent backsliding to the previous drinking behaviors. of alcohol problems, the type and frequency of the possible On the other hand, low scores in this feature show no recent used drugs in the last 3 months is queried. behavior changes in user. AUDIT Score: Alcohol Use Disorders Identification Test (AUDIT) [5] is a 10-item questionnaire that we use to iden- 3.1.3 Dependence to Alcohol tify people whose alcohol consumption has become hazardous We assess the user’s degree of dependence to the alcohol or harmful to their health. The amount and frequency of using a self-administered 20-item questionnaire called Sever- drinking, alcohol dependence, and problems caused by alco- ity of Alcohol Dependence Questionnaire (SADQ-C) [41]. hol are queried using this instrument. Questions are scored This feature can be used to predict the likelihood of achiev- using a 5-point Likert scale. The total score is the summa- ing control-drinking goals, and likelihood of withdrawal. tion of all the answers. Table 1 shows the way AUDIT scores Questions are answered in a 4-point Likert scale, so the are interpreted. range of the score will be from 0 to 60. Scores higher than 30 The cut-off numbers may be different based on average for males and 25 for females show severe alcohol dependence body weight, gender, race, and cultural standards. and probable need of medical intervention. Scores in 16-30 range show moderate dependence. Otherwise, the user has 3.1.5 Frequency of Drinking mild physical dependency. This category of features describes the user’s drinking pat- terns and amount of alcohol consumption. So, the alcohol 3.1.4 Risk Factors behavior change recommender systems can use them as indi- We use the Brief Drinker Profile (BDP) [30] to assess some cators of the user’s drinking pattern and provide more per- information about the family drinking history, other drug sonalized recommendations for the user. We can build rapport with the user by calling the user Table 1: AUDIT score interpretation. with his/her name during the intervention and personalize AUDIT Score Interpretation his/her experience. score < 4 No drinking problems For the systems that use ECAs as the interface, they can 4 ≤ score ≤ 8 Harmful for ages under 18 and females adapt the ECA’s race and gender to the user’s. Research score > 8 Alcohol dependence shows that patient-physician race concordance can lead to 8 < score ≤ 15 Should be advised to reduce drinking better health outcomes [15] and that people respond to the 16 ≤ score ≤ 19 Should be suggested counseling ethnicity of ECAs in the same ways of that of humans. score ≥ 20 Should be warranted further diagnose 3.2.2 Affective Features The problem drinkers, who experience intense feeling of Drinking Pattern: A drinker may have one of the two depression, discontent and indifference to the world around drinking patterns: steady or periodic. A drinker with steady them, report that they drink to relax or reduce anxiety drinking pattern drinks at least once a week and about the symptoms [39]. Another research found that emotions and same amount every week. A drinker with periodic drinking affective states of a person, depending on personality types, pattern drinks less often than once a week and is abstinent predict motives for problem drinking [16]. Therefore emo- between drinking episodes. tions and affective states of a problem drinker is crucial for Drinks in Last 4 Weeks: This feature includes the num- the user model. They can help to fine-tune appropriateness ber of standard drinks that a user had per week in the last of recommendations and interventions and improve context four weeks. A standard drink is a 12 oz beer (5% alcohol), a awareness. 5 oz wine (12.5% alcohol), or a 1.5 oz liquor (40% alcohol). The emotions and affective states can be also used to im- Relative Drinking: This feature shows the user’s sta- prove user’s experience in the systems which use ECAs as tistical standing relative to the other U.S. people with the the user interface. The user’s experience may affect implic- same gender. itly the amount and accuracy of the disclosed information. Peak BAC: Blood Alcohol Concentration (BAC) is the Building a close relationship with the user facilitates his/her amount of alcohol contained in a person’s blood and is mea- behavior change and affects the accuracy of the information sured as weight per unit of volume. Widmark’s [44] basic disclosed [38], [42]. formula for calculating BAC is as follows: While the instruments demonstrated can be used as self- 5.14 administered via form-based interface, the suggested style %BAC = (A × × r) − 0.015 × H (1) W to administer them is to be delivered via a face-to-face in- Where, “A” is the total number of liquid ounces of alco- terview [28]. The face to face interviews can be conducted hol that the person has drunk since the commencement of by ECAs [25] which can build a close relationship with the drinking. It is calculated by multiplying the number of liq- user and have positive effects on the interview process. uid ounces of drink by its percentage of alcohol. “W” is the Monitoring the facial expressions and mood helps to de- person’s weight in pounds. “r” is the alcohol distribution termine the user’s emotions and affective states. In the next ratio which is 0.73 for men and 0.66 or women. “H” is the section, we described each of these non-verbal signals in number of hours between commencement of drinking and more details. the time of BAC calculation. Facial Expressions: According to [3], the facial expres- sions are the most important modalities in human behav- 3.2 Target-Behavior Related Features ioral judgment. Thus, including facial expressions in human These features are not specific to the target behavior but affect analysis can increase the accuracy [12] of the analysis. they are implicitly related with the target behavior. For Using facial expressions, the behavior change recommender example, demographic information of the user have signif- system can recognize the effect of the recommended mes- icant role in personalizing the recommendations and using sage/feedback on the user, and his/her affective state. the normative data to interpret the target-behavior specific The user’s emotional facial expressions can be recognized features. As a concrete example, the normative data used for through a camera using a real-time facial expression recog- rating the dependence to alcohol and consequences of drink- nition system and categorized into the universal emotion ing depend on the user’s gender, race, and age. In addition categories [17]: happy, sad, angry, surprised, and neutral. to the demographic information, we studied affective and Mood: Mood is the user’s background state of well-being contextual features which provide important target-behavior which is often modeled on a bipolar scale of positive-negative related information. valence. Mood changes much slower than emotion and lasts longer time (e.g, minutes to days). Therefore, unlike fa- 3.2.1 Demographic Features cial expressions that are updated in real-time, mood can be Demographic features can be used to improve interpreta- updated less frequently (e.g., every 5 minutes) in the user tion of the other feature scores and to improve interaction model. with the user. Studies [28], [22] show that people of dif- To capture the user’s mood, we suggest to get the average ferent genders, ages, and ethnicities experience different of the user’s categorized emotional facial expressions in a types of negative consequences after drinking. For example, time window and to classify the user’s emotions to positive women have more sleeping problems after drinking while and negative emotions. men have more sexual and money problems after drinking. Therefore, taking the demographic data into account in 3.2.3 Contextual Features the user model enables recommending more accurate feed- The advancement of the technology on mobile devices, back and exercises to the user. increasing usage of mobile applications, and location-based social networking systems such as Facebook Location1 and implicit indicators of topical relevance. In Proceedings FourSquare2 introduced new possibilities in development of of the 17th ACM international conference on the context-aware systems. Other than location and time Multimedia, MM ’09, pages 461–470, New York, NY, information, social networking and micro-blogging services USA, 2009. ACM. (Twitter3 ) also offer possibilities to track mood [11], social [5] T. F. Babor, J. C. Higgins-Biddle, J. B. Saunders, and interactions, relationships, and social ties of the user. M. G. Monteiro. AUDIT: The Alcohol Use Disorders Recently, increased popularity of the music-based social Identification Test. Guidelines for use in primary networks4 and their tight integration to the general purpose health care. World Health Organization, Department social networks introduced new possibilities to improve con- of Mental Health and Substance Dependence, 2 text awareness of the systems. Research [14] shows that edition, 2001. listening some music genres is positively associated with al- [6] N. Banerjee, S. Agarwal, V. Bahl, R. Chandra, cohol use. It is also possible to identify personal song lists A. Wolman, and M. D. Corner. Virtual Compass: which lead to alcohol use by tracking multiple context re- relative positioning to sense mobile social interactions. lated parameters. For example variation of mood depending In Proc. Eighth International Conference on Pervasive on the listened songs and music genres might give important Computing, number LNCS 6030, pages 1–21, Helsinki, insight about the factors which prepare appropriate psycho- Finland, May 2010. logical conditions for alcohol use. [7] V. Bellotti, B. Price, P. Rasmussen, M. Roberts, D. J. The location, time of the day, social interactions and mood Schiano, A. Walendowski, B. Begole, E. H. Chi, tracking [11] can help to understand specific conditions which N. Ducheneaut, J. Fang, E. Isaacs, T. King, M. W. result in alcohol use such as physical environment, psycho- Newman, and K. Partridge. Activity-based logical conditions, and social conditions. serendipitous recommendations with the Magitti Several studies show the relationship between reasons and mobile leisure guide. Proceeding of the twenty-sixth motivations for drinking [1], [22], [21]. Their results imply annual CHI conference on Human factors in that contextual awareness will have positive effect on inter- computing systems - CHI ’08, page 1157, 2008. vention and support systems. [8] B. M. Bewick, K. Trusler, M. Barkham, A. J. Hill, These results implies that personalization and tailoring, J. Cahill, and B. Mulhern. The effectiveness of based on the contextual factors, are crucial for the alcohol web-based interventions designed to decrease alcohol intervention and behavior support recommender systems. consumption–a systematic review. Preventive Thus, in our proposed user model, we propose to use avail- medicine, 47(1):17–26, July 2008. able information from social networking services and mobile [9] B. M. Bewick, K. Trusler, B. Mulhern, M. Barkham, applications to monitor drinking related contextual features. and A. J. Hill. The feasibility and effectiveness of a web-based personalised feedback and social norms 4. CONCLUSION alcohol intervention in UK university students: a In this paper we proposed a user model for alcohol re- randomised control trial. Addictive behaviors, lated lifestyle change recommender systems. We proposed 33(9):1192–8, Sept. 2008. target-behavior specific features and target-behavior related [10] T. Bickmore. Establishing and maintaining long-term features for the user model. We identified the importance human-computer relationships. ACM Transactions on of each feature group for the alcohol related intervention Computer-Human, 12(2):293–327, 2005. and recommender systems. We proposed a user model com- [11] J. Bollen, H. Mao, and X. Zeng. Twitter mood posing of eight different groups of features, consequences of predicts the stock market. Journal of Computational drinking, motivation to change, dependence to alcohol, risk Science, 2(1):1–8, Mar. 2011. factors, frequency of drinking, demographic features, affec- [12] G. Caridakis, L. Malatesta, L. Kessous, N. Amir, tive features, and contextual features. A. Raouzaiou, and K. Karpouzis. Modeling naturalistic affective states via facial and vocal 5. REFERENCES expressions recognition. In Proceedings of the 8th international conference on Multimodal interfaces, [1] A. Abbey, M. J. Smith, and R. O. Scott. The pages 146–154. ACM, 2006. relationship between reasons for drinking alcohol and [13] J. Cassell and E. J. Sullivan, S. Prevost, & E. alcohol consumption: An interactional approach. Churchill. Embodied Conversational Agents. Social Addictive Behaviors, 18(6):659 – 670, 1993. Psychology, 40(1):26–36, 2000. [2] G. Adomavicius and A. Tuzhilin. Recommender [14] M.-J. Chen, B. a. Miller, J. W. Grube, and E. D. Systems Handbook. Springer US, Boston, MA, 2011. Waiters. Music, substance use, and aggression. [3] N. Ambady and R. Rosenthal. Thin slices of Journal of studies on alcohol, 67(3):373–81, May 2006. expressive behavior as predictors of interpersonal [15] L. a. Cooper, D. L. Roter, R. L. Johnson, D. E. Ford, consequences: A meta-analysis. Psychological Bulletin, D. M. Steinwachs, and N. R. Powe. Patient-centered 111(2):256–274, 1992. communication, ratings of care, and concordance of [4] I. Arapakis, I. Konstas, and J. M. Jose. Using facial patient and physician race. Annals of internal expressions and peripheral physiological signals as medicine, 139(11):907–15, Dec. 2003. 1 http://www.facebook.com/about/location [16] M. L. Cooper, V. B. Agocha, and M. S. Sheldon. A 2 https://foursquare.com motivational perspective on risky behaviors: the role 3 http://www.twitter.com of personality and affect regulatory processes. Journal 4 http://www.spotify.com of personality, 68(6):1059–88, Dec. 2000. [17] P. Ekman, W. V. Friesen, and P. Ellsworth. What (SOCRATES). Psychology of Addictive Behaviors, emotion categories or dimensions can observers judge 10(2):81–89, 1996. from facial behavior? In P. Ekman, editor, Emotion in [32] E. Miluzzo, N. D. Lane, K. Fodor, R. Peterson, H. Lu, the human face, pages 39–55. Cambridge University M. Musolesi, S. B. Eisenman, X. Zheng, and A. T. Press, New York, 1982. Campbell. Sensing meets mobile social networks: the [18] D. Gavalas and M. Kenteris. A web-based pervasive design, implementation and evaluation of the cenceme recommendation system for mobile tourist guides. application. In Proceedings of the 6th ACM conference Personal and Ubiquitous Computing, 15(7):759–770, on Embedded network sensor systems, SenSys ’08, May 2011. pages 337–350, New York, NY, USA, 2008. ACM. [19] R. K. Hester, D. D. Squires, and H. D. Delaney. The [33] S. Noar, C. Benac, and M. Harris. Does tailoring Drinker’s Check-up: 12-month outcomes of a matter? Meta-analytic review of tailored print health controlled clinical trial of a stand-alone software behavior change interventions. Psychological bulletin, program for problem drinkers. Journal of substance 133(4):673, 2007. abuse treatment, 28(2):159–69, Mar. 2005. [34] S. M. Noar, N. Grant Harrington, S. K. Van Stee, and [20] P. Krebs, J. O. Prochaska, and J. S. Rossi. A R. Shemanski Aldrich. Tailored Health meta-analysis of computer-tailored interventions for Communication to Change Lifestyle Behaviors. health behavior change. Preventive medicine, American Journal of Lifestyle Medicine, 5(2):112–122, 51(3-4):214–21, 2010. Nov. 2010. [21] E. Kuntsche, R. Knibbe, G. Gmel, and R. Engels. [35] D. Quercia and J. Ellis. Using mobile phones to Who drinks and why? A review of socio-demographic, nurture social networks. IEEE Pervasive Computing, personality, and contextual issues behind the drinking 09:12–20, 2010. motives in young people. Addictive behaviors, [36] K. K. Rachuri, M. Musolesi, C. Mascolo, P. J. 31(10):1844–57, Oct. 2006. Rentfrow, C. Longworth, and A. Aucinas. [22] J. W. LaBrie, J. F. Hummer, and E. R. Pedersen. Emotionsense: a mobile phones based adaptive Reasons for drinking in the college student context: platform for experimental social psychology research. the differential role and risk of the social motivator. In Proceedings of the 12th ACM international Journal of studies on alcohol and drugs, 68(3):393–8, conference on Ubiquitous computing, Ubicomp ’10, May 2007. pages 281–290, New York, NY, USA, 2010. ACM. [23] T. a. Laveist and A. Nuru-Jeter. Is doctor-patient race [37] C. R. Rogers. The necessary and sufficient conditions concordance associated with greater satisfaction with of therapeutic personality change. Journal of care? Journal of health and social behavior, Consulting Psychology, 21(2):95–103, 1957. 43(3):296–306, Sept. 2002. [38] B. G. Silverman, J. Holmes, S. Kimmel, C. Branas, [24] G. Linden and B. Smith. Amazon. com D. Ivins, R. Weaver, and Y. Chen. Modeling emotion recommendations: Item-to-item collaborative filtering. and behavior in animated personas to facilitate human Internet Computing, IEEE, 7(February):76–80, 2003. behavior change: the case of the HEART-SENSE [25] C. Lisetti, U. Yasavur, C. D. Leon, and R. Amini. game. Health care management science, 4(3):213–28, Building an On-Demand Avatar-Based Health Sept. 2001. Intervention for Behavior Change. Proceedings of the [39] M. M. Spada and A. Wells. Metacognitions, emotion Twenty-Fifth International Florida Artificial and alcohol use. Clinical Psychology & Psychotherapy, Intelligence Research Society Conference, 2012. 12(2):150–155, Mar. 2005. [26] M. L. a. Lustria, J. Cortese, S. M. Noar, and R. L. [40] D. D. Squires and R. K. Hester. Using technical Glueckauf. Computer-tailored health interventions innovations in clinical practice: the Drinker’s delivered over the Web: review and analysis of key Check-Up software program. Journal of clinical components. Patient education and counseling, psychology, 60(2):159–69, Feb. 2004. 74(2):156–73, Feb. 2009. [41] T. Stockwell, D. Murphy, and R. Hodgson. The [27] L. Mauriello. An Internet-Based Computer-Tailored severity of alcohol dependence questionnaire: its use, Intervention to Promote Responsible Drinking: reliability and validity. British journal of addiction, Findings from a Pilot Test with Employed Adults. 78(2):145–155, 1983. Alcoholism Treatment, 30(1):1–15, 2011. [42] a. L. Suchman, K. Markakis, H. B. Beckman, and [28] W. Miller and J. Tonigan. The drinker inventory of R. Frankel. A model of empathic communication in consequences (DrInC). 1995. the medical interview. JAMA : the journal of the [29] W. R. Miller and R. K. Hester. Matching problem American Medical Association, 277(8):678–82, Feb. drinkers with optimal treatments. In W. R. Miller and 1997. N. Heather, editors, Treating addictive behaviors [43] A. White, D. Kavanagh, H. Stallman, B. Klein, Processes of change, pages 175–203. Plenum Press, F. Kay-Lambkin, J. Proudfoot, J. Drennan, 1986. J. Connor, A. Baker, E. Hines, and R. Young. Online [30] W. R. Miller and G. A. Marlatt. Brief Drinker Profile. alcohol interventions: A systematic review. J Med Psychological Assessment Resources, Odessa, FL, Internet Res, 12(5):e62, Dec 2010. 1984. [44] E. M. P. Widmark. Principles and Applications of [31] W. R. Miller and J. S. Tonigan. Assessing drinkers’ Medicolegal Alcohol Determination. Biomedical motivation for change: The Stages of Change Publications, Davis, California, 1981. Readiness and Treatment Eagerness Scale