Nihao: A Predictive Smartphone Application Launcher

Chunhui Zhang, Xiang Ding, Guanling Chen, Ke Huang, Xiaoxiao Ma, Bo YanMobiCASE 2012 · DOI

TL;DR

This paper introduces Nihao, a predictive smartphone application launcher that uses a Bayesian Network to anticipate user app needs based on context like time, location, and previous app usage. It demonstrates that Nihao can significantly reduce app lookup time and improve user experience by intelligently ranking app icons, suggesting prediction accuracy is more crucial than UI dynamism.

What Problem Does This Paper Address?

As smartphones become ubiquitous and packed with an ever-growing number of applications, users face a significant challenge in efficiently finding and launching the apps they need. Traditional launchers often rely on alphabetical order or installation time, forcing users to scroll through lengthy lists or multiple home screens. This manual app discovery process consumes valuable time and effort. The paper addresses this by proposing an intelligent launcher that predicts user intent and proactively surfaces relevant applications, aiming to streamline the app lookup and management experience.

Methods

The core of Nihao is a Bayesian Network model that predicts the likelihood of a user opening a specific app. This model leverages several contextual factors:

  • Time of Day: Categorized into broad periods (morning, noon, afternoon, dinner, evening).
  • Day of Week: Simplified to a binary variable (weekday/weekend).
  • User's Location: Identified using a significant place recognition algorithm that combines cellular network information and WiFi access points.
  • Last Used App: Captures immediate app usage correlation.

Nihao calculates a "score" for each app based on the conditional probability derived from these contexts. The apps are then ranked and displayed in a grid view, with the most likely apps appearing first. The system employs a client-server architecture, with the client handling UI and local data logging, and the server managing historical data and potentially more complex model learning.

Datasets

The paper describes a field study conducted over six weeks with seven users. This study involved collecting app usage data and related contextual information. The data was used for both training the prediction models and evaluating the system's performance. No specific publicly available datasets are mentioned as being used for training or evaluation; the data was collected directly from the study participants.

Key Insights and Ideas

  • Context is King for App Usage: The study strongly validates the hypothesis that app usage patterns are highly dependent on context (time, location, previous app). This is a fundamental insight for building personalized mobile experiences.
  • Prediction Accuracy Trumps UI Flashiness: Surprisingly, the research found that a more dynamic or frequently changing UI (higher "UI difference") did not necessarily lead to longer app lookup times. Instead, the accuracy of the prediction itself was the most critical factor in reducing lookup time.
  • User Behavior Varies Greatly: The study highlighted significant diversity in how users interact with their phones. Some users heavily rely on a few core apps, making frequency-based prediction effective for them, while others have broader usage patterns that benefit more from context-aware prediction like Nihao.
  • Search is Underutilized: The study observed that users preferred browsing for apps over searching, suggesting that a well-organized and predictive launcher is more aligned with typical user behavior.

Key Results

  • Reduced App Lookup Time: Nihao demonstrated an average app lookup time of 4.94s for downloaded apps, compared to 5.51s for the default launcher. While seemingly small, this is considered a significant achievement, especially given the default UI's user familiarity.
  • Improved App Position Accuracy: Nihao successfully placed 76% of app launches within the top four positions (first row) of the grid view, significantly outperforming a frequency-based predictor (58%).
  • User Satisfaction: A user survey revealed that five out of seven users were "quite satisfied" with Nihao's predictions and expressed a desire to continue using it as their personal app management assistant.
  • Contextual Prediction Superiority: Nihao's context-aware prediction model outperformed simpler models (frequency-only, time-only, location-only, app correlation-only) in terms of placing frequently used apps in more accessible positions.

Limitations and Future Work

  • Fixed Model for All Users: Currently, Nihao uses a single model for all users. The authors acknowledge that more personalized models could exist for different users.
  • Limited Contextual Factors: While Nihao uses several contexts, the paper suggests exploring richer contextual information for even better prediction.
  • UI Predictability vs. Dynamism: The paper notes that while Nihao's UI is more dynamic, this doesn't negatively impact lookup time if predictions are accurate. However, understanding the optimal balance is an ongoing area.

Future work includes:

  • Developing customized prediction models that dynamically switch between different predictors based on individual user behavior.
  • Experimenting with different models and richer contexts to further improve prediction accuracy.
  • Testing the system with a larger and more diverse user base.
  • Exploring the potential of Nihao as a widget for even quicker access.

Summary updated

BibTeX
@inproceedings{zhang2012nihao,
  author = {Chunhui Zhang and Xiang Ding and Guanling Chen and Ke Huang and Xiaoxiao Ma and Bo Yan},
  title = {{Nihao: A Predictive Smartphone Application Launcher}},
  booktitle = {MobiCASE},
  year = {2012},
  doi = {10.1007/978-3-642-36632-1\_17},
}
Figure