Behavior Recommender Systems


Sedentary lifestyles influence the onset of many serious health problems. Healthy behavior change is an arduous task, but behavior recommenders can help their users achieve it. This novel type of recommender system can suggest its users to pair up with an exercise partner, or propose a day-by-day plan to increase their physical activeness. We quest for understanding the factors that influence well-beings of people, and develop a behavior recommender system that exploits these factors to provide suggestions for healthier lifestyles. Toward this end, we analyse various aspects of activities of daily living, such as physical exercises, work, sleep, social interactions, and daily routines. We utilize state-of-the-art timeseries analysis methods to explore trends, repetitive patterns, regularities and correlations from data such as bluetooth proximity, cell-tower locations, calorie expenditure, and other various information collected from pervasive, wearable sensors. We also show that traditional factor analysis methods can also be applied to sensor data.

Recent results show that activity regularity, social interaction regularity and certain daily routines have strong influences on life satisfaction and well-being. Furthermore, we have successfully shown that people can be grouped into meaningful categories in terms of their daily activity routines. This opens up further possibilities, such as tailoring specialized recommendations to improve the level of activeness for each common daily routine category.


Activities of daily living, activity recognition, data mining, timeseries analysis, statistical analysis, behavior recommenders




Onur Yürüten and Pearl Pu




  • Onur Yürüten and Pearl Pu. A framework of behavior recommender systems. Journal submission in review, 2017
  • Onur Yürüten and Pearl Pu. Factoring the Habits: Comparing Methods for Discovering Behavior Patterns from Large Scale Activity Datasets. International Conference on Big Data Analytics, Data Mining and Computational Intelligence (BIGDACI). Part of Multi Conference on Computer Science and Information Systems (MCCSIS) (forthcoming) 2016
  • Onur Yürüten, Jiyong Zhang, and Pearl Pu. Decomposing Activities of Daily Living to Discover Routine Clusters. In the 28th AAAI Conference on Artificial Intelligence (AAAI-14), Quebec City, Canada, July 27-31, 2014
  • Onur Yürüten, Jiyong Zhang, and Pearl Pu. Predictors of Life Satisfaction Based on Daily Activities from Mobile Sensor Data. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’14), Toronto, Canada, April 26 – May 1, 2014.

Available Resources

Recommender System framework – source distribution

We provide to the research community the source code of the developed framework for behavior recommender. It is shared under the General Purpose License (GPL 3.0) and accessible online at We ask researchers who would use this code to cite the papers from above.

Activity and Snacking Datasets for Activity Analysis and Recommendations

We also make available for research purposes four datasets curated or obtained using the behavior recommender system framework and designed to discover distinct routines of activities of daily living. We call these datasets as HT-48, HT-83, YQZ, and SNACK. These datasets were curated by conducting longitudinal user studies. The construction process involved longitudinal observations of physical activities (HT-48, HT-83, YQZ) and snacking behavior (SNACK). During the longitudinal user study, the participants were subject to various interventions: becoming partners with another person (HT-48, HT-83), joining to an exercise group (YQZ) or receiving motivational messages (SNACK).

These datasets are distributed under the following terms of use:

  • The datasets can be used for research purposes only.
  • Do not redistribute the datasets further. Instead, please refer interested parties to Onur Yürüten ( or to the web page of the behavior recommender systems project:
  • Please kindly cite the papers above in your publications if you use any of these lexicons in your research.

To request these datasets, please send an e-mail to pearl.pu [at] and onuryuruten [at]

Initial activity datasets

We used two activity datasets for initial discovery of common behavior patterns: HT-48 and YQZ. You can find more information on these datasets and their terms of use on this page