Abstract: In the current streaming mobile application distribution system, the distribution of applications relies manually install/uninstall either by mobile user or administrator, which does not take mobile users’ context information into account. To address these problems, it proposes a context-aware streaming mobile application recommendation mechanism. It collects the user’s context information data and uses the Xgboost algorithm to identify user’s context. Then, it recommends applications to user according to the identified context information. At the same time it uses the user’s feedback information to improve the accuracy of user-specific application recommendations. The experimental results show that the accuracy and time overhead of Xgboost algorithm is better than traditional algorithms, is practical in streaming application distribution system.

Key words: streaming application, context-aware, Xgboost, machine learning, personalized recommendation

摘要： 在目前流式应用分发系统中，客户端的移动应用分发都是依靠系统后台管理员人工操作或者简单地依靠位置信息为用户分发应用，没有考虑到用户在不同的情境活动下对应用的需求差异问题。针对上述问题，提出一种基于用户情境感知的流式应用推荐机制。该机制通过采集流式应用场景下用户的情境信息数据，利用机器学习Xgboost算法识别用户情境活动，并根据识别的用户情境来为用户推荐应用。同时，利用用户的反馈信息进一步提高用户个性化应用推荐的准确度。实验结果表明，Xgboost算法在准确率和时间开销上性能优于传统算法，在流式应用分发系统中有很高的实际应用价值。

关键词: 流式应用, 情境感知, Xgboost, 机器学习, 个性化推荐