Personalized User Modelling for Context-Aware Lifestyle Recommendations to Improve Sleep Vaibhav Pandey∗ Dhruv Deepak Upadhyay∗ vaibhap1@uci.edu ddupadhy@uci.edu University of California, Irvine University of California, Irvine Irvine, CA, USA Irvine, CA, USA Nitish Nag Ramesh Jain University of California, Irvine University of California, Irvine Irvine, CA, USA Irvine, CA, USA nagn@uci.edu jain@ics.uci.edu ABSTRACT Recommender Systems (HealthRecSys’20), Online, Worldwide, September 26, Sleep is a significant contributor to leading a healthy lifestyle. Each 2020. , 7 pages. day, most people go to sleep without any idea about how their night’s rest will be or how they can leverage their data to improve it. For an activity that humans spend near a third of their life do- ing, there is a surprising amount of mystery around it. Despite 1 INTRODUCTION current research, creating personalized sleep models in real-world Lifestyle factors have a high impact on our health outcomes. Our settings has been challenging. Existing literature provides several eating, sleeping, and movement patterns determine large parts of connections between daily activities and sleep quality. Unfortu- our short and long-term health [Hills et al. 2015; Nag 2020; Nag nately, these insights do not generalize well in many individuals. et al. 2018; Shochat 2012]. Keeping track of how we behave in For these reasons, it is essential to create a data-driven personalized different contextual situations, and the impact these behavioral sleep model. This research proposes a sleep model that captures patterns have on our health is difficult for medical professionals causal relationships between daily activities and sleep quality and and individuals. At the same time, with the increasing prevalence of presents the user with specific feedback recommendations to im- chronic diseases such as diabetes and hypertension, understanding prove sleep quality. Using N-of-1 experiments on longitudinal user the effect of lifestyle on different aspects of our health becomes a data and event mining, the model generates a probabilistic un- critical research challenge [Lee et al. 2017; Sarris et al. 2014]. The derstanding between lifestyle choices (exercise, eating, circadian rising popularity of wearable and IoT devices provide us with an rhythm, environmental selection) and their respective impact on opportunity to address this problem computationally. There is a sleep quality. Our experimental results identified and quantified re- considerable body of research dedicated to finding and logging life lationships while extracting confounding variables through a causal events from multimodal data streams [Gurrin et al. 2014; Oh and framework. We then utilize the generated model to provide lifestyle Jain 2017; Sellen and Whittaker 2010]. A multitude of consumer recommendations to optimize sleep outcomes in a context-aware devices such as smartwatches and smart home systems measure health recommendation system. aspects of our daily life as events and data streams and control our local environment [Casino et al. 2018]. Using the data streams and KEYWORDS events captured by these devices, we can find recurring behavioral sleep recommendation system, health recommendation system, patterns and associated health outcomes to create an explainable context aware, event mining, hypothesis verification, N-of-1 exper- rule-based model of the person [Pandey et al. 2018]. Explainabil- iments, n=1, causal inference, healthy lifestyle, personalized health, ity is a desired quality in health prediction and recommendation precision health. systems as it can verify the quality of predictions and builds user engagement and trust in the system. ACM Reference Format: We can use such a model to provide the right guidance at the right Vaibhav Pandey, Dhruv Deepak Upadhyay, Nitish Nag, and Ramesh Jain. time for health management. Health recommendation systems pro- 2020. Personalized User Modelling for Context-Aware Lifestyle Recommen- vide us a way to apply cybernetic principles to manage a person’s dations to Improve Sleep. In Proceedings of the 5th International Workshop on Health Recommender Systems co-located with 14th ACM Conference on health [Nag et al. 2017]. Using lifestyle interventions, we can build a navigation system that guides us through our day much in the ∗ Both authors contributed equally to this research. same way that modern navigation systems inform drivers about the most optimum path towards their destination [Nag and Jain 2019]. We need to design a context-aware personal recommenda- HealthRecSys ’20, September 26, 2020, Online, Worldwide tion system that changes the person’s context with every event that © 2020 Copyright for the individual papers remains with the authors. Use permitted happens during the day. The dynamic context allows us to provide under Creative Commons License Attribution 4.0 International (CC BY 4.0). This volume is published and copyrighted by its editors. an optimal recommendation at every point of the day, and calibrate the recommendations as different events occur. HealthRecSys ’20, September 26, 2020, Online, Worldwide Pandey and Upadhyay, Nag, Jain In this work, we present a sleep recommendation system that study, we used a combination of actigraphy and sound to record considers various lifestyle factors as contextual variables and possi- movements during sleep events. Several sleep applications use au- ble recommendations for optimizing a sleep parameter. We create dio from the phone mic to record and report sleep quality (e.g., a rule-based model to understand the effects of different lifestyle SleepCycle, Sleepscore). factors (such as exercise during the day, and mealtimes) on sleep Previous works have attempted to predict sleep quality using parameters. These rules are used in a recommendation system smartphone mic data in conjunction with machine learning [Min framework for providing the most effective interventions at any et al. 2014]. Other studies have attempted to use actigraphy graphs time throughout the day. These interventions could be lifestyle and utilize Deep Learning to predict sleep quality [Sathyanarayana recommendations presented to the user or a command to one of et al. 2016]. There are even studies that attempt to forgo the idea of the IoT or smart devices that control the user’s environment (e.g., tracking sleep and use factor graph models based on daily activity to HVAC systems, Light bulbs, Music or ambient sounds). predict sleep quality with 78% accuracy [Bai et al. 2012]. While these models are useful, they do not address the individual variability in 2 RELATED WORKS sleep and do not provide a way to incorporate different lifestyle Our work spans across context-aware and health recommendation aspects. systems, sleep specific monitoring and prediction, and causal event mining. We review the current approaches and limitations below. 2.3 Effect of Lifestyle Activities on Sleep Quality 2.1 Context Aware and Health Current literature has made many attempts to identify daily activ- Recommendation Systems ities that affect a person’s sleep quality. Studies have shown that The field of sleep and health recommendation systems is relatively sleep and exercise are related, and higher physical activity levels new. Studies have explored the pitfalls of using the conventional can lead to better sleep latency [Yang et al. 2012] [Kline 2014][Kel- recommendation systems for health and devised alternatives us- ley and Kelley 2017]. A systematic review has also shown that ing entity properties and relationships [López-Nores et al. 2012]. dietary patterns and the types of food eaten throughout the day Context-awareness is an essential quality for health recommenda- lead to better sleep quality and duration [St-Onge et al. 2016]. The tion systems [Schäfer et al. 2017]. Context-aware recommendation environmental factors (temperature and humidity) are also vital to systems (CARS) have been explored in different domains [Villegas our sleep duration and quality [Troynikov et al. 2018]. From these et al. 2018]. CARS have traditionally incorporated context infor- studies and many more, it is clear that choices made throughout mation in collaborative filtering models in one of three ways, 1) the day affect the quality of the next night’s sleep. Contextual Pre-filtering, 2) Contextual Post-filtering, and 3) Con- textual Modelling [Adomavicius and Tuzhilin 2015]. There can be different types of contextual information relevant to a recom- 3 CAUSAL RULE-BASED MODELLING: EVENT mendation system. These usually fall into one of the following MINING categories: temporal, location, individual (user characteristics), ac- Creating a model of the person’s behavior and health is central tivity (about the activity), and relational (when multiple entities to building personalized health recommendation systems. In this are involved)[Villegas et al. 2018]. We have adopted a contextual work, we present an approach to build a rule-based explainable modeling approach and incorporated the contextual information in model for predicting sleep outcomes in different contextual situ- the rule-based model itself. Multiple studies have explored person- ations. We apply event mining [Jalali 2016] principles to perform alized recommendation systems for different aspects of user-health, N-of-1 experiments on a user’s data [McDonald et al. 2017] that such as diet [Khan et al. 2019]. These utilize different learning tech- allows us to find causal relationships between different lifestyle niques to develop personalized models for individuals but usually events and biological outcomes. The process is described in figure lack explainability, which is an important characteristic of health 1. recommendation systems. Event mining allows us to discover and specify patterns between different events in a person’s life. We utilize these patterns to create 2.2 Sleep Monitoring and Prediction hypotheses that might describe a person’s behavior. A hypothesis Applications needs to specify the intervention event and the associated confound- Polysomnography is considered to be the gold standard for under- ing factors that affect the relationship between the intervention standing sleep quality. The test records various metrics such as and the outcome. The confounding factors are specified using the brain waves, oxygen levels in the blood, heart rate, breathing, and temporal delay operator, Δ[𝑡𝑏 , 𝑡𝑒 ], that relates the events that occur eye and leg movements [Kushida et al. 2005]. It requires a sleep within the specified time interval [𝑡𝑏 , 𝑡𝑒 ]. The confounding factors expert and multiple medical sensors. The study’s accuracy does are specified as a set of patterns 𝑃 between the lifestyle events and come at the cost of needing too many resources and equipment to the outcome. Thus, a hypothesis would be specified as 𝐸𝑖 − → 𝐸𝑜 , 𝑃 be performed every night reliably. Actigraphy is another popular where we want to measure the causal effect of intervention 𝐸𝑖 on technique used to measure sleep quality. It measures sleep quality the outcome event 𝐸𝑜 while controlling for the events specified by using a wearable device (e.g., a watch) by recording movement the set of patterns 𝑃. These patterns and hypotheses can be derived during a sleep event. Its simplicity comes at the cost of accuracy, as from existing knowledge and human intuition, allowing us also to it can only infer sleep quality via movement measurements. For our leverage the results of population studies performed in clinical and Personalized User Modelling for Context-Aware Lifestyle Recommendations to Improve Sleep HealthRecSys ’20, September 26, 2020, Online, Worldwide Figure 1: Rule based personal model for predicting sleep outcomes. We divide the occurrences of the outcome events into smaller subsets based on the values of co-occurring contextual factors. This minimizes the variance in the outcome due to the confounding variables within each subset. Subsets that exhibit significantly different distribution for different values of input events are converted to rules and added to the model. controlled settings. the changing user parameters and trained using the latest observed Combining the event mining operators with causal inference prin- data. We can easily update the rules by updating the outcome vari- ciples allows us to perform N-of-1 experiments on the user’s longi- able’s distribution whenever the rule matches the user’s current tudinal data. We utilize the candidate hypothesis specification to context. create different subsets of data based on the values of co-occurring confounding events. These subsets minimize the variance in the outcome due to confounding factors and mimic a version of the 4 MULTI-ITEM HEALTH do-operator [Pearl 2009]. We can use different statistical techniques, RECOMMENDATIONS such as linear regression or t-tests to find the effect of the interven- The rule-based health model allows us to find the health outcomes tion event on the outcome within each subset. This allows us to in different contextual situations. The user’s activities during the find the effect of the intervention on the outcome in an unbiased day (such as exercise, meals, work-related stress) and their local manner, and if we can capture all the confounding variables in the environmental parameters (such as temperature, humidity, and set of patterns, we would obtain the causal effect of the intervention ambient light and sounds) determine these contextual variables. on the outcome. The result of the test is stored as a conditional rule Thus, we can utilize this model in a recommendation system setting that uses confounding variables and the intervention event as the to determine the set of parameters (both user behavior and envi- predicate. The distribution of the outcome events in the subset is ronmental variables) for optimizing a health outcome (e.g., sleep used to make a prediction. quality). A set of these contextual rules can be used to predict health Every action taken by the user and every environmental exposure outcomes. We would need to find the most relevant rule by match- changes the user’s health state [Nag 2020], which changes the con- ing the user’s current context with the set of rules and utilize it text for future actions and recommendations. We need to retrieve to make the prediction. We describe it in more detail in the next the relevant events from the user’s events and data streams that section. A rule-based model, while lacking in complexity, offers the impact their health state [Pandey et al. 2020]. Different contextual advantage of explainability and online training. Every prediction parameters are defined as aggregations of these events. For exam- and recommendation generated from this model can be explained ple, Total Screen Time during the day is an important confounding using the associated contextual factors, thus eliminating recom- factor for understanding an individual’s sleeping habits. It can be mendations based on spurious relationships. This is an essential determined by aggregating the duration of all the screen activity characteristic of health models and recommendation systems. As events (such as working, watching TV, and social media activity) the user behavior changes over time, the model needs to adapt to during the day. These aggregations can be performed using events HealthRecSys ’20, September 26, 2020, Online, Worldwide Pandey and Upadhyay, Nag, Jain Figure 2: Live context calculation. The system updates user context every time they log an event. We retrieve all the sub-events and parameters relevant for context calculation (e.g., time of meal from dinner event). The retrieved contextual information is added to the existing context, and the updated context is used to generate a new set of recommendations. These recommen- dations are then sent either to the user or to a device controlling the user’s environmental factors. based triggers encoded as condition-action rules. As new events recommendations are recomputed anytime an event changes the appear in the person’s events log, the retrieved events can be ag- user’s context. This process is depicted in figure 2. Thus, at any gregated to change the user’s live context parameters. We can use point during the day, the recommendation system would provide the latest context values to provide a set of recommendations that a list of timestamped actions to be performed by different agents would optimize the user’s health outcomes. (the user, or an automated device) to optimize the sleep outcomes. We match the live contextual parameters for the person with the contextual parameters of the various rules present in the model. If the current context matches multiple rules, then we utilize the rule with the highest likelihood of the desired outcome. Once we have identified the rules that match the current context, we can 5 EXPERIMENTS AND RESULTS utilize the unmatched contextual parameters and the intervention We ran experiments to create a personal rule-based model for opti- event to find the set of parameters that can lead to the optimal mizing a person’s sleep quality metrics by providing lifestyle and outcome. We can either present the recommendation to the user (if local environmental recommendations. We utilized data collected the recommendation is an action to be taken by the user) or commu- by one individual for more than two years using readily available nicate with a smart device that controls the user’s environmental consumer applications and wearable and IoT devices. We performed context (e.g., smart home devices, HVAC systems, smart bulbs). two sets of experiments on the collected dataset to create and eval- The recommendation system produces a set of actions that would uate the model. The first set of experiments find the average causal maximize the likelihood of the optimal outcome; thus, the proposed effect of input variables on sleep quality metrics. We used Welch’s recommendation system is different from typical recommendation t-tests and a p-value of 0.05 to determine statistical significance. systems as the recommendation consists of multiple items. The second set of experiments tested the prediction accuracy for a Since any event during the day can change the user’s context, the static pre-trained model vs. an online training model. Personalized User Modelling for Context-Aware Lifestyle Recommendations to Improve Sleep HealthRecSys ’20, September 26, 2020, Online, Worldwide Table 1: Sleep Quality Measure and Event Thresholds Table 2: Lifestyle Factors and Event Thresholds Variable Classification Ranges Event Name Classifications Variables Event Name [0, 15] Good Ranges Sleep Latency (15, 30] Average [0] None (30, ∞) Poor Exercise Minutes (0, 50] Poor [0, 20] Good Per Day (50, 150] Average Awake Minutes (150, ∞) Good (20, ∞) Poor [0, 1] Good [0, 150] Poor Awakenings >5 mins Exercise Minutes (1, ∞) Poor (150, 300] Average Per Week [0.85, 1.00] Good (300, ∞) Good Sleep Efficiency [0] Missing [0, 0.85) Poor Interval Between (0, 180] Poor Eating and Sleeping (180, ∞) Good [0, 900] LT 15 Hours 5.1 Data Set Minutes Awake (900, 1020] Btwn 15-17 Hours Between Sleep Events The data set includes exercise and lifestyle parameters for a 31 year (1020, ∞) GT 17 Hours old male collected continuously over 2 years via the user’s Garmin [0, 60] Cold Fenix 5 smart watch, their smartphone, and an IoT sensor that Starting Temperature (60, 67] Comfortable collected local temperature and humidity values. Sleep Cycle was (67, ∞) Warm primarily used to keep track of sleep events. Apple Health Kit was [0, 30] Low used to help compile sleep quality measures recorded by the Garmin Starting Humidity (30, 50] Ideal smart watch, daily step counts, and daily floors climbed. The ac- (50, 100] High clerometer measures of the smartwatch and the audio recordings of Sleep Cycle were used to create sleep quality measures. Strava was used to keep track of exercise events. An image based food log the two distributions and use the combined distribution at the time recorded feeding times with phone camera metadata, and a Sensor- of contextual matching. Push IoT sensor was used to collect temperature and humidity The results of these experiments are in Figure 3. One interesting during sleep events. All of these data sources were then temporally observation is that an average temperature(60-67 𝐹 𝑜 ) seems to matched to record lifestyle events that took place throughout the improve every sleep quality measure except for sleep latency. This day accurately. is an important observation as it shows that not all quality measures We used the thresholds mentioned in Table 1 and Table 2 to convert are correlated with each other and that an improvement in one does the data streams to relevant events for the event mining analysis. not necessarily equate to an improvement in all other sleep quality We used nine lifestyle/environmental events: Previous Night’s Sleep measures. Another interesting insight is that exercise improves Quality Measures, Exercise Minutes in the Day, Interval Between sleep latency the most. On average, we can tell that exercising a lot Eating and Sleeping, Minutes Awake Between Sleep Events, Tem- will reduce sleep latency by 10.5 minutes with just a small workout perature, and Humidity when going to bed. The possible output will help reduce sleep latency by an average of 8 minutes. events are sleep quality measures (Table 1). We used 70% of the data to build the model, and 30% of the data to test the model. The 5.3 Context Matching and Sleep Predictions train-test split was created based on temporal order. We also want to demonstrate the contextual matching of rules and test the accuracy of the model’s predictions, as that will determine 5.2 Causal Rules and Effects from N-of-1 the efficacy of any recommendations we generate. We train a linear Experiments regression model corresponding to every rule in the model, and use We perform different N-of-1 tests on the user’s data to find the aver- the data subset that matches the rule to train the model. This model age effects of different lifestyle and environmental events on sleep is then used to predict sleep outcomes for situations matching with quality parameters while controlling for other lifestyle parameters the rule. as confounding factors. We treat one of the input event’s possible We used two training strategies for the prediction model; 1) Pre- values as the baseline and compare the distribution of the outcome trained static models, and 2) the warm start online training. We variable for other values of the event with the baseline distribution. expect that over time the user’s sleep behavior would change, and If changing the input event value causes a significant change in the thus an online learning strategy would eventually start outperform- outcome distribution while controlling for confounding variables, ing the pre-trained model. the rule is deemed significant. This gives us the causal effect of dif- The model’s input features are Exercise minutes during the day, ferent values of an input event on the observed outcome. We repeat Feeding Time, Time Awake, Humidity, and Temperature while this experiment while controlling for different sets of variables and going to bed. We match the user’s context with the context of the aggregate the causal effects to find the input event’s average causal rules, and the most significant rule that matches the context is used effect. If the difference is not found to be significant, then we merge to provide the recommendation. HealthRecSys ’20, September 26, 2020, Online, Worldwide Pandey and Upadhyay, Nag, Jain Figure 3: Average Effects that each input event has on the output event when compared to each input event’s base category. If a metric is 0 then no significant relations were found. quality. With enough data, this model can be potent. Coupled with the context-aware health recommendation system, it can give peo- ple control over their sleeping habits that have not been previously possible. The insights from the model are easily understandable and should promote user engagement as the recommendations are not coming from a black-box model but are simple relationships be- tween daily habits. The context-aware recommendation approach allows us to provide recommendations at different points during the day. Even if the user fails to follow any recommendations, we can provide them with a new set of recommendations and modify their local environment to best suit their sleep requirements. This helps us move from recommendation-based guidance to navigational guidance. Figure 4: Comparison of pre-trained model vs. online train- Although this framework incorporates many useful data sources ing. Online training allows the model to adapt to user’s and provides useful insights into users’ sleep behavior, there are changing sleep behavior resulting in lower error in predic- many ways to improve. Many other lifestyle factors affect our sleep tions. outcomes but are not included in our study, such as stress and nutritional intake. Our events based framework would allow us We create a set of contextual variables at the end of the day for to include these events and data streams with minimal additional each day in the dataset. These values are then used to find a match- effort. ing rule. If multiple matches are found, then we used the rule with We have proposed a recommendation system to optimize one health a higher statistical significance. The linear model associated with outcome. However, in a real-world application, the users may want the matched rule would then be used to predict the sleep outcome to optimize multiple outcomes simultaneously. This can be an excit- parameter. The key difference between the pre-trained and online ing opportunity for the recommendation systems research commu- models is that the online model would be updated continuously nity, and we hope to stimulate future work expanding to a larger using the data in the test set. This way, the online model has the user base and with different applications. opportunity to adapt to the user over time. The results of the model predictions are in 4. The results illustrate an improvement in the REFERENCES performance of the online model over the pre-trained model. Even- Gediminas Adomavicius and Alexander Tuzhilin. 2015. Context-Aware Recommender Systems. In Recommender Systems Handbook. Springer US, Boston, MA, 191–226. tually, we expect the online model would achieve a much smaller https://doi.org/10.1007/978-1-4899-7637-6{_}6 MSE as it adapts to the changing sleep behavior exhibited by the Yin Bai, Bin Xu, Yuanchao Ma, Guodong Sun, and Yu Zhao. 2012. Will You Have a user. Good Sleep Tonight? Sleep Quality Prediction with Mobile Phone. Technical Report. https://www.andrew.cmu.edu/user/yuzhao1/papers/2012BodyNets1.pdf Fran Casino, Constantinos Patsakis, Edgar Batista, Octavian Postolache, Antoni 6 DISCUSSION AND FUTURE WORK Martínez-Ballesté, and Agusti Solanas. 2018. Smart Healthcare in the IoT Era: A Context-Aware Recommendation Example. In 2018 International Symposium We have shown the need for and created a sleep model that uti- in Sensing and Instrumentation in IoT Era, ISSI 2018. Institute of Electrical and lizes event mining and causal inference principles to provide useful Electronics Engineers Inc. https://doi.org/10.1109/ISSI.2018.8538106 feedback about the relationships between lifestyle events and sleep Personalized User Modelling for Context-Aware Lifestyle Recommendations to Improve Sleep HealthRecSys ’20, September 26, 2020, Online, Worldwide Cathal Gurrin, Alan F. Smeaton, and Aiden R. Doherty. 2014. LifeLogging: Personal of the 2018 ACM Multimedia Conference. Association for Computing Machinery, Inc, big data. , 125 pages. https://doi.org/10.1561/1500000033 New York, New York, USA, 1993–2002. https://doi.org/10.1145/3240508.3241913 Andrew P. Hills, Steven J. Street, and Nuala M. Byrne. 2015. Physical Activity and Hyungik Oh and Ramesh Jain. 2017. From multimedia logs to personal chronicles. Health: "What is Old is New Again". In Advances in Food and Nutrition Research. In MM 2017 - Proceedings of the 2017 ACM Multimedia Conference. Association for Vol. 75. Academic Press Inc., 77–95. https://doi.org/10.1016/bs.afnr.2015.06.001 Computing Machinery, Inc, New York, New York, USA, 881–889. https://doi.org/ Laleh Jalali. 2016. Interactive Event-driven Knowledge Discovery from Data Streams. 10.1145/3123266.3123375 (2016). Vaibhav Pandey, Nitish Nag, and Ramesh Jain. 2018. Ubiquitous event mining to George A. Kelley and Kristi Sharpe Kelley. 2017. Exercise and sleep: a systematic enhance personal health. In UbiComp/ISWC 2018 - Adjunct Proceedings of the 2018 review of previous meta-analyses. Journal of Evidence-Based Medicine 10, 1 (2 2017), ACM International Joint Conference on Pervasive and Ubiquitous Computing and 26–36. https://doi.org/10.1111/jebm.12236 Proceedings of the 2018 ACM International Symposium on Wearable Computers. ACM Mansura A Khan, Ellen Rushe, Barry Smyth, and David Coyle. 2019. Per- Press, New York, New York, USA, 676–679. https://doi.org/10.1145/3267305.3267684 sonalized, Health-Aware Recipe Recommendation: An Ensemble Topic Model- Vaibhav Pandey, Nitish Nag, and Ramesh Jain. 2020. Continuous Health Interface ing Based Approach. Technical Report. https://github.com/MAK273/ Event Retrieval. In Proceedings of the 2020 International Conference on Multimedia SupportingFileForHealthRecsys2019 Retrieval. Association for Computing Machinery (ACM), New York, NY, USA, 486– Christopher E Kline. 2014. The bidirectional relationship between exercise and sleep: 494. https://doi.org/10.1145/3372278.3390705 Implications for exercise adherence and sleep improvement. American journal of Judea Pearl. 2009. Causal inference in statistics: An overview. Statistics Surveys 3, 0 lifestyle medicine 8, 6 (11 2014), 375–379. https://doi.org/10.1177/1559827614544437 (2009), 96–146. https://doi.org/10.1214/09-SS057 Clete A Kushida, Michael R Littner, Timothy Morgenthaler, Cathy A Alessi, ; Den- Jerome Sarris, Adrienne O’Neil, Carolyn E. Coulson, Isaac Schweitzer, and Michael Berk. nis Bailey, ; Jack Coleman, Leah Friedman, Max Hirshkowitz, Sheldon Kapen, 2014. Lifestyle medicine for depression. , 13 pages. https://doi.org/10.1186/1471- Milton Kramer, Teofilo Lee-Chiong, ; Daniel, L Loube, Judith Owens, ; Jeffrey, 244X-14-107 P Pancer, and ; Merrill Wise. 2005. Practice Parameters for the Indications for Aarti Sathyanarayana, Shafiq Joty, Luis Fernandez-Luque, Ferda Ofli, Jaideep Srivas- PSG-AASM Practice Parameters SLEEP Practice Parameters for the Indications for tava, Ahmed Elmagarmid, Teresa Arora, and Shahrad Taheri. 2016. Sleep Quality Polysomnography and Related Procedures: An Update for 2005. Technical Report 4. Prediction From Wearable Data Using Deep Learning. JMIR mHealth and uHealth https://academic.oup.com/sleep/article-abstract/28/4/499/2696969 4, 4 (11 2016), e125. https://doi.org/10.2196/mhealth.6562 Duck chul Lee, Angelique G. Brellenthin, Paul D. Thompson, Xuemei Sui, I. Min Lee, Hanna Schäfer, Santiago Hors-Fraile, Raghav Pavan Karumur, André Calero Valdez, and Carl J. Lavie. 2017. Running as a Key Lifestyle Medicine for Longevity. , Alan Said, Helma Torkamaan, Tom Ulmer, and Christoph Trattner. 2017. Towards 45–55 pages. https://doi.org/10.1016/j.pcad.2017.03.005 Health (Aware) Recommender Systems. In Proceedings of the 2017 International Martín López-Nores, Yolanda Blanco-Fernández, José J. Pazos-Arias, and Alberto Gil- Conference on Digital Health - DH ’17. ACM Press, New York, New York, USA, Solla. 2012. Property-based collaborative filtering for health-aware recommender 157–161. https://doi.org/10.1145/3079452.3079499 systems. Expert Systems with Applications 39, 8 (6 2012), 7451–7457. https://doi. Abigail J. Sellen and Steve Whittaker. 2010. Beyond total capture: A constructive org/10.1016/j.eswa.2012.01.112 critique of lifelogging. Commun. ACM 53, 5 (5 2010), 70–77. https://doi.org/10. Suzanne McDonald, Francis Quinn, Rute Vieira, Nicola O’Brien, Martin White, Derek W. 1145/1735223.1735243 Johnston, and Falko F. Sniehotta. 2017. The state of the art and future opportunities Tamar Shochat. 2012. Impact of lifestyle and technology developments on sleep. , for using longitudinal n-of-1 methods in health behaviour research: a systematic lit- 19–31 pages. https://doi.org/10.2147/NSS.S18891 erature overview. , 307–323 pages. https://doi.org/10.1080/17437199.2017.1316672 Marie-Pierre St-Onge, Anja Mikic, and Cara E Pietrolungo. 2016. Effects of Diet on Jun-Ki Min, Afsaneh Doryab, Jason Wiese, Shahriyar Amini, John Zimmerman, and Sleep Quality. Advances in Nutrition 7, 5 (9 2016), 938–949. https://doi.org/10.3945/ Jason I. Hong. 2014. Toss ’n’ turn. In Proceedings of the 32nd annual ACM conference an.116.012336 on Human factors in computing systems - CHI ’14. ACM Press, New York, New York, Olga Troynikov, Christopher G. Watson, and Nazia Nawaz. 2018. Sleep environments USA, 477–486. https://doi.org/10.1145/2556288.2557220 and sleep physiology: A review. , 192–203 pages. https://doi.org/10.1016/j.jtherbio. Nitish Nag. 2020. Health State Estimation. Ph.D. Dissertation. University of California, 2018.09.012 Irvine. http://arxiv.org/abs/2003.09312 Norha M. Villegas, Cristian Sánchez, Javier Díaz-Cely, and Gabriel Tamura. 2018. Char- Nitish Nag and Ramesh Jain. 2019. A Navigational Approach to Health: Actionable acterizing context-aware recommender systems: A systematic literature review. Guidance for Improved Quality of Life. Computer 52, 4 (4 2019), 12–20. https: Knowledge-Based Systems 140 (1 2018), 173–200. https://doi.org/10.1016/J.KNOSYS. //doi.org/10.1109/MC.2018.2883280 2017.11.003 Nitish Nag, Vaibhav Pandey, Hyungik Oh, and Ramesh Jain. 2017. Cybernetic Health. Pei Yu Yang, Ka Hou Ho, Hsi Chung Chen, and Meng Yueh Chien. 2012. Exercise (5 2017). http://arxiv.org/abs/1705.08514 training improves sleep quality in middle-aged and older adults with sleep problems: Nitish Nag, Vaibhav Pandey, Preston J. Putzel, Hari Bhimaraju, Srikanth Krishnan, and A systematic review. Journal of Physiotherapy 58, 3 (9 2012), 157–163. https: Ramesh Jain. 2018. Cross-Modal Health State Estimation. In MM 2018 - Proceedings //doi.org/10.1016/S1836-9553(12)70106-6