1. Information about the paper


Srinivasan, Vijay, Saeed Moghaddam, Abhishek Mukherji, Kiran K. Rachuri, Chenren Xu, and Emmanuel Munguia Tapia. “Mobileminer: Mining your frequent patterns on your phone.” In Proceedings of the 2014 ACM International Joint Conference on Pervasive and Ubiquitous Computing, pp. 389-400. ACM, 2014.


Smartphones can collect considerable context data about the user, ranging from apps used to places visited. Frequent user patterns discovered from longitudinal, multi-modal context data could help personalize and improve overall user experience. Our long term goal is to develop novel middleware and algorithms to efficiently mine user behavior patterns entirely on the phone by utilizing idle processor cycles. Mining patterns on the mobile device provides better privacy guarantees to users, and reduces dependency on cloud connectivity. As an important step in this direction, we develop a novel general-purpose service called MobileMiner that runs on the phone and discovers frequent co-occurrence patterns indicating which context events frequently occur together. Using longitudinal context data collected from 106 users over 1-3 months, we show that MobileMiner efficiently generates patterns using limited phone resources. Further, we find interesting behavior patterns for individual users and across users, ranging from calling patterns to place visitation patterns. Finally, we show how our co-occurrence patterns can be used by developers to improve the phone UI for launching apps or calling contacts.






2. My review of the paper


This paper tried to utilize co-occurrence mobile context and use the pattern to predict next user activity and further provide related suggestion. Examples of the data collected in this paper are: media play log, setting log, place log, call/sms log, app usage log, physical activity log. Then it will collect co-occurrence pattern, in other word: mobile context from these logs which are frequently happening at the same time. For example: {morning, at home, listen to music}. Then it will predict next mobile context even that is likely to happen. They develop a weighted mining algorithm called weighted mining of temporal patterns (WeMiT).

There are four main contributions from this paper:

  1. design and implementation of the MobileMiner pattern mining service.
  2. experimental validation of the feasibility of running the MobileMiner implementation on a phone using context logs that were collected from 106 users over 1-3 months.
  3. analyzing patterns from individual users to understand their utility in multiple use cases, and also analyzing which patterns occur frequently across all 106 users.
  4. implement two UI improvements using predictions from MobileMiner. It predicted the next outgoing call or app launch, and provide users with convenient predictive shortcut icons for the next contacts to be called or the next apps to be launched.



  1. It is a good idea to utilize idle time of the phone because it lower the burden on the user and the phone itself.
  2. Finally, I see a paper that considers privacy. This paper considers privacy by running all mining process on the phone. Therefore, there is no sensitive data usage sent to a server.
  3. It reduces the need for cloud connectivity because the mining and predicting pattern are done entirely on the phone.
  4. Their algorithm performed better than previously available Apriori algorithm. They concluded that WeMiT is 15 times faster than the Apriori algorithm. For a subset of users, Apriori requires a running time of 20 minutes on the phone, while WeMiT discovers the same patterns in 78.5 seconds.
  5. It is also a good idea to provide an API for developers to take advantage of this learning algorithm.


  1. The downfall of entirely running it on the phone is: it will be fully personalized. It will not be able to get more general prediction by comparing pattern from all users. Some pattern may be better personalized. However, previous paper in class showed that both general and personalized pattern can be more useful if combined. This paper itself concluded some general patters across subset of users. Such general context could be used to help users with lower number of context data. Then again, it will need cloud connectivity. And to preserve privacy, the data collected should be encrypted and anonymized.
  2. It also can not benefit from social context since it does not share context data between users. However, sharing context data should be done while preserving privacy. It should ask permission and the users should confirm each other thus agreeing to share context data between them.

 Discussions and Future Directions:

  1. I was asking why do they bother implementing their idea in Tizen platform which is not popular (yet). And at the near end of the paper, they want to port this application Android (which is more popular). Why don’t they implement it in Android from the beginning? However, the answer is clear, since the authors worked for Samsung Research.
  2. This paper answer all of our research questions: what are the usage data streams, how to retrieve data efficiently, what can we tell from the usage data, how can we improve/redesign mobile systems to better fit personal pattern. However, it focuses more on three questions other than how to retrieve data efficiently. This other question has been discussed more in previous class.
  3. Further research could be done to collect user responses to the suggested application and contact. Then use this information to better predict what is the next application to be used or the contact to be called.