1. Information about the paper


LiKamWa, Robert, Yunxin Liu, Nicholas D. Lane, and Lin Zhong. “Moodscope: building a mood sensor from smartphone usage patterns.” In Proceeding of the 11th annual international conference on Mobile systems, applications, and services, pp. 389-402. ACM, 2013.


We report a first-of-its-kind smartphone software system, MoodScope, which infers the mood of its user based on how the smartphone is used. Compared to smartphone sensors that measure acceleration, light, and other physical properties, MoodScope is a “sensor” that measures the mental state of the user and provides mood as an important input to context-aware computing. We run a formative statistical mood study with smartphonelogged data collected from 32 participants over two months. Through the study, we find that by analyzing communication history and application usage patterns, we can statistically infer a user’s daily mood average with an initial accuracy of 66%, which gradually improves to an accuracy of 93% after a two-month personalized training period. Motivated by these results, we build a service, MoodScope, which analyzes usage history to act as a sensor of the user’s mood. We provide a MoodScope API for developers to use our system to create mood-enabled applications. We further create and deploy a mood-sharing social application.


2. My review of the paper


This paper tried to infer user’s mood using several context data. The context data collected in this paper are: Application usage, phone calls, email messages, SMSes, web browsing histories, and location changes. The authors then developed nice simple app to collect mood data to train the learning algorithm. They collected the data from 32 participants. Then they tried several model and found that the best approach is personalized model (93% accuracy). However, to perform well, it needs long time to train the classifier. Therefore, they apply general model at the early stage of training. They also found that phone call and categorized application usage are the features that contribute most for the mood detection.

There are five main contributions from this paper:

  1. Demonstrated the feasibility of inferring mood from smartphone usage, paving the way for energy-efficient, privacy-preserving systems that automatically infer user mood.
  2. Designed a user-friendly mood journaling tool to effectively collect user mood data.
  3. Showed that how mood affects smartphone usage is personal; a general model performs significantly worse than personalized models. Moreover, categorized application usage and phone calls are strong indicators of mood for our participants.
  4. Described a lightweight, power-efficient, and easy-to deploy realization of MoodScope, consuming only 3.5 milli-Watt-hours over a day.
  5. Developed an API for developers to interface with mood, which was used to develop and deploy a sample mood-sharing application.



  1. It is a good idea to provide an API for developers, thus enabling many recommendation services according to user’s mood.
  2. Unlike previous work, it performs at good accuracy without the need of additional hardware and low energy consumption.
  3. It has been cited by 23 following works and was featured in several technology news site.
  4. It considers privacy seriously, although it has only done hashing private data which it admits is not secure enough for public deployment.
  5. It has also considered offloading training computation to the cloud which reduce the cost on the phone. Then the phone just apply the model to the user’s personal data.


  1. Some figures and tables are not located at the same page where they are discussed. A little effort to put figures and tables at the same page will increase readability.

Discussions and Future Directions:

  1. It is interesting to see Microsoft researchers used iPhone and Android platform instead of Windows Phone.
  2. It is interesting to see that the author is aware of publicity. In his webpage, roblkw.com, he listed the paper accompanied with slides, video of the conference presentation, and links to news coverage.
  3. Future research might be done to include more context data and better privacy preserving features.
  4. 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 fir personal pattern.
  5. The paper did collect number of words in text messages and emails which suggests that the contents are being scanned. This could be a privacy issue. However, we can not argue with the fact that the participant got paid $75 and a chance to get an iPad 2. In addition, the author promise to adhere privacy very seriously.
  6. Another privacy issue is that sharing mood should be under permission of the user, not automatically. The paper has already done good consideration by letting the user to choose how frequent they want to share their mood.