How to Personalize Conversational Coaches for Stress (PDF)

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/328677797 How to Personalize Conversational Coaches for Stress Management? Conference Paper · October 2018 DOI: 10.1145/3267305.3267698 CITATION 1 READS 238 5 authors, including: Some of the authors of this publication are also working on these related projects: Computers i nthe Human Interaction Loop (CHIL) project View project ANR VICTEAMS View project Jean-Claude Martin Computer Sciences Laboratory for Mechanics and Engineering Sciences 248 PUBLICATIONS 2,652 CITATIONS SEE PROFILE Sophie Rosset Computer Sciences Laboratory for Mechanics and Engineering Sciences 167 PUBLICATIONS 1,282 CITATIONS SEE PROFILE All content following this page was uploaded by Jean-Claude Martin on 05 November 2018. The user has requested enhancement of the downloaded file.  How to Personalize Conversational Coaches for Stress Management? Abstract Hundreds of well-being apps aim to manage stress. Despite some successes in developing personalized regimes for stress coaching, current apps are still far from offering a compelling user experience. We discuss the requirements and technical challenges underlying the design of a virtual coach, including the critical need to support both personalization and conversation. Author Keywords Stress; coach; personalization; conversational agent. ACM Classification Keywords H.5.m. Information interfaces and presentation. Introduction Coping with stress is critical to well-being. Chronic stress is the psychological and physiological response to emotional pressure following a prolonged period in which an individual perceives he or she has little or no control [1]. Chronic stress increases mortality, morbidity, immuno-suppression, anxiety, depression, and work absenteeism. Hundreds of well-being apps now aim to manage stress, e.g. by encouraging physical activity. We also see the emergence of novel sensing methods to automatically detect stress at work [2, 3]. However the theories underlying these apps are usually implicit [4], and there are challenges in making © 2018 Association for Computing Machinery. ACM acknowledges that this contribution was authored or co-authored by an employee, contractor or affiliate of a national government. As such, the Government retains a nonexclusive, royalty-free right to publish or reproduce this article, or to allow others to do so, for Government purposes only. UbiComp/ISWC'18 Adjunct, October 8–12, 2018, Singapore, Singapore © 2018 Association for Computing Machinery. ACM ISBN 978-1-4503-5966-5/18/10…$15.00 https://doi.org/10.1145/3267305.3267698. Illustrative conversation Coach: Hello. Your profile suggests that you know how to find solutions to the problems you encounter at work. You manage your time well, and you plan your deadlines very well. On the other hand, it appears that you find it difficult to express your emotions. What do you think? Do you agree with that assessment? User: Well, it’s true that I often find it easier to focus on work and that I get a lot of satisfaction out of being well- organized. It’s also true that I tend not to share my feelings. And actually sometimes I am not even sure that I really know how I feel. But I do sometimes feel isolated and I don’t have anyone to talk to about things when something upsets me. C: Do you want me to suggest exercises that could help you make progress with dealing with your emotions? Jean-Claude MARTIN LIMSI-CNRS, Univ. Paris Sud Orsay, 91400, France [email protected] Christine LESCANFF CIAMS, Univ. Paris Sud Orsay, 91400, France [email protected] Sophie ROSSET LIMSI-CNRS, Univ. Paris Sud Orsay, 91400, France [email protected] Marilyn WALKER University of California, Santa Cruz [email protected] Steve WHITTAKER University of California, Santa Cruz [email protected] 718  these apps compelling, as evidenced by lack of long term usage [5]. Despite some success in developing personalized adaptive stress coaching [6], current apps don’t yet offer compelling long-term user experiences. One solution is virtual coaches that provide individualized motivations to increase engagement and compliance. Such coaches involve participants in active problem solving about their condition, offer bespoke recommendations about remedial strategies, and actively monitor and encourage users as strategies are deployed. We motivate our theoretical approach, discussing requirements and technical challenges in designing such a coach, including the need to support both personalisation and conversation [7]. Psychological Theories Unlike most behavior change apps, our approach is theory-led, as we derive system designs directly from behavioral research and theory summarized here. Stress is a known risk factor for maladjustment and psychopathology [8]. The type of stress experienced depends on how people react to it [8], indicating the importance of coping. Stress is a transactional subjective phenomenon depending on perceiver interpretation [1]. Two core concepts are how a person evaluates the potential threat (primary appraisal) and then what one can do (secondary appraisals). The Transtheoretical Model [15] conceptualizes intentional behavior change as progress through stages, which can also inform tailoring of motivational messages [16]. Personality differences also explain stress vulnerability [11]. Individuals scoring high on trait neuroticism [12] are susceptible to stress [13]. Regulatory Focus Theory (RFT) [14] also proposes two motivational systems (regulatory foci): achievement (promotion) or preservation (prevention). Promotion-focus people aim to promote ideal selves (what they want to be) whereas revention-focus people give greater weight to “ought” selves (what they think they have to be) when enacting behaviors. For a given individual, one of the two systems is dominant, and in our work [27], this focus provides the basis for personalised motivational messages. Affective computing and conversations General frameworks for designing virtual eHealth coaches for behavior change and for well-being address autonomy, competence and relatedness [18]. Stress has also received attention from the Affective Computing research community. Experiments focus on acute stress detection [19], and systems to manage chronic stress [6,20,21,22] including the timing of interventional messaging. New sensors and eCoaching can help people cope better with stress at work [3], as can tailored motivational text messages based on behavior change theory [16]. Other work designs conversational agents offering social support about general healthy behaviors [24, 23, 25]. However current work provides limited personalised conversation and support for stress management [7]. Below, we outline theory-led requirements for a stress coach. Intervention outline Management involves three main steps: 1) initial profile collection (eg. personality), 2) initial assessment (suggestion and negotiation about initial strategies for coping), and 3) regular interactions: goal achievement and reconsideration of relevant strategies. Inspired by stress theories [1, 8], we designed a survey [26] that profiles people’s overall levels of stress and their stress U: Sure, I’d be willing to try something if you think it would help. I haven’t done anything like this for a long time. C: We can always make progress in all areas of life and that’s what is interesting, to keep developing throughout life. To familiarize yourself with your emotions, you can begin to turn your attention to yourself for a few minutes a day, with a meditation or relaxation exercise. I will give you an instruction sheet with some very simple exercises and then you can pick one of them and try it. U: When would be the best time to do it? C: It's up to you to decide what is the best time for you, perhaps in the morning before going to work, or during your lunch break, or during another break in the day. You could also do it in the evening, or when you are commuting. 719  coping strategies. The survey identifies: 1) how well a person can cope with a given problem, 2) how the person copes with stress through emotion regulation, and 3) how this person is able to access social support for coping. This questionnaire was filled in by 500 PhD students and 588 employees. It displays good reliability and face validity. Developing this personal profile allows us to design interactive dialogues which recommend tailored strategies to address experienced stresses. An example of dialog provided by a psychologist who is an expert in stress management is displayed in the sidebar (Figure 1). It illustrates initial strategies that patients could bring into play to manage their stress. Actual user conversations would of course be considerably messier. Challenges and potential solutions for personalised coaching conversations Effective persuasive technologies must provide tailored user interactions [27]. Such personalization should be achieved at multiple levels using individualised dialogue strategies focused on how a specific user should cope with stress, based on their user profile. Personalised conversations should be designed to negotiate user-appropriate strategies for stress management. E.g. a coach might say “I suggest that you attend a Yoga session”. Users might answer “No, I don’t think that Yoga works for me. I would rather cycle outdoors”. There is an important requirement that the system be able to understand the user’s responses, e.g. understanding rejection and that the user is suggesting a different strategy. There are also opportunities to explore mixed-initiative and politeness strategies with recommendations [28] Online stress forums might be used to develop models of how people express stress and how others respond to requests (Figure 2). Such data might improve the natural language understanding component of the coach. The motivational messages displayed during the conversations (e.g. “Nothing is impossible for a willing heart: You will succeed tomorrow”) also need to be tailored. It might be more efficient to select motivational messages that fit a user’s personality traits and chronic regulatory focus [27]. Other potential issues include the level of detail of proposed strategies. “I suggest that you do some physical exercise, this will help you get more relaxed” might not be specific enough - so a coach might distinguish between types of physical exercises: cardio, weights and stretching. The system also needs knowledge about the user’s preferences and capabilities regarding remedial activities. Possible solutions to gathering preferences include survey-based profiling, crowdsourcing preferred activities [2], or automatic tracking of activities to establish frequency and competence [30]. Other required conversational skills include assessing achievements and status (e.g. “It looks like you did no running today, why?”) leading to possible adjustments of goals. This might involve renegotiating goals that preserve self-efficacy [6]. In sum, we motivate the requirements for a stress coach, describing the challenges to supporting this critical aspect of well-being. References 1. R. S. Lazarus. 1966. Psychological stress and the coping process. McGrawHill Series Psychology. 2. V. Hollis et al. 2017. What Does All This Data Mean for my Future Mood? Hum.-Comput. Interact. 32, 5-6 (2017), 208-267. U: How long should I do it for? C: You can start with 1 or 2 minutes, then gradually increase to 5 or 10 minutes. Or if you can, you can try to do it twice a day for three minutes each time. The more you practice, the easier it will be. U: All right, I'll start tomorrow. I’ll see how it goes. C: Great. Get a notebook to track your progress. Write down what exercise you’ve decided on, how much time you spend on it, and what time of day you did it. We can look at it the next time and see how you are doing and how to make further progress. Figure 1: Example of dialogue strategies based on user profile from the survey. 720  3. S.J. Koldijk. 2016. Context-Aware Support for Stress Self-Management: SIKS Dissertation Series; No. 2016-43. RU Radboud Universiteit. 4. E. B. Hekler et al. 2013. Mind the theoretical gap. CHI '13. NY, 3307-3316. 5. R. Waller, S. Gilbody. 2009. Barriers to the uptake of computerized cognitive behavioural therapy: Psychol Med. 39(5):705-12. 6. A. Konrad et al. 2015. Finding the Adaptive Sweet Spot: CHI '15. NY, USA, 3829-3838. 7. T. Bickmore, et al 2010. Maintaining Engagement in Long-Term Interventions with Relational Agents. Applied artificial intelligence : AAI 24.6: 648–666. 8. M. Wang, K. J. Saudino. 2011. Emotion Regulation and Stress. J. Adult Development, 18(2), 95–103. 9. A. P. Allen et al.2014. Biological and psychological markers of stress in humans. Neuro. and Biobehavioral Reviews, 38(10), 94–124. 10. R. R. McCrae and P. T. Costa. 1987. Validation of the five-factor model of personality across instruments and observers. J. of Personality and Social Psychology, 52(1), 81–90. 11. T. R. Schneider. 2004. The role of neuroticism on psychological and physiological stress responses. Exp. Social Psy., 40(6), 795–804. 12. E. T. Higgins. 1997. Beyond pleasure and pain. American psychologist, 52(12), 1280. 13. J. O. Prochaska and C.C. DiClemente. 1983. Stages and processes of self-change of smoking. J Consult Clin Psychol 51(3):390. 14. R.A.J. de Vries, et al. 2017. A word of advice: how to tailor motivational text messages based on behavior change theory to personality and gender. Pers Ubiquit Comput (2017) 21: 675. 15. D. Peters et al. 2018. Designing for Motivation, Engagement and Wellbeing in Digital Experience. Front Psychol. 9:797. 16. J. Costa et al.. 2016. EmotionCheck: UbiComp '16. ACM, NY, USA, 758-769. 17. G. Mark, et al. 2014. Stress and multitasking in everyday college life. CHI '14. NY, USA, 41-50. 18. T. Umematsu et al. 2018. Improving Stress Forecasting using LSTM Neural Networks. EMBC’18, Honolulu, Hawaii. 19. A. Sano et al. 2017. Designing Opportune Stress Intervention Delivery Timing using Multi-modal Data. ACII’17, San Antonio, Texas, October 2017. 20. D. Schulman et al. 2011. An Intelligent Conversational Agent for Promoting Long-Term Health Behavior Change using Motivational Interviewing. AAAI 2011. 21. D. Schulman and T. Bickmore. 2009. Persuading users through counseling dialogue with a conversational agent. in Proceedings of 4th conference on persuasive technology, 22. A.S. Miner et al. 2016. Smartphone-based conversational agents and responses to questions about mental health, interpersonal violence, and physical. JAMA Intern Med. 2016 May 1;176(5):619-25. 23. B.B. Chang unpublished. PhD Thesis. Evaluation du stress au travail et méthodes de prévention. Université Paris Sud. Dir. Christine Le Scanff. 24. J.-C. Martin, C. Clavel. 2018. Tailored, Multimodal and Opportune Interactions on a Wearable Sport Coach: The WE-nner Framework. Global Thoughts, Local Designs. T. Clemmensen et al. (Eds.): INTERACT 2017, LNCS 10774, Springer, 24-32. 25. M. A Walker, J. E Cahn, S. J Whittaker. 1997. Improvising linguistic style: Social and affective bases for agent personality. AAMAS’97. 26. M. Rabbi, et al. MyBehavior: UbiComp '15. NY, 707-718. Q: Redditors with anxiety, what are simple yet effective ways that you deal with it? A1: Definitely exercise. If I get super anxious about something, I start to get very jittery and I feel like I just have way too much energy. Then when I go for a moderately strenuous bike ride, my mind feels so much calmer and I can think more clearly. A2: Being outside and exercise is incredibly underrated for helping all kinds of mental ailments. A3:. Even a short walk helps a lot if there's not enough time for exercise. A4: And cardio, not weights. Weights are fine for strength and tone, but nothing soothes an anxious mind like vigorous cardio, whether it be running, biking, ellipticals, or whatever. An added bonus is the ability, if you're on a stationary machine to just check out for 20-30 minutes. Figure 2: Examples of advice and social support on Reddit Forum. 721 View publication stats View publication stats