Abstract:

With the rapid development of mobile devices, and the number of mobile applications has grown very rapidly, how to recommend appropriate mobile applications which users are really interested in becomes more and more important. It is found that user's age is one of the factors that can affect the choice of mobile Apps of users, after analyzing some questionnaires and the mobile industry reports released by Airei. Based on this result, this paper proposes to integrate user's age information into traditional recommendation algorithms for personalized mobile application recommendation. The prediction of user's age is regarded as a classification problem. It uses multi-view learning approach to expand training dataset based on questionnaire and user behavior information. Then user age prediction model can be built through classification model based on user's historical behavior information. This paper proposes two recommendation models: AgeBPR and AgeSocialMF, to incorporate age information to traditional recommendation models. Experiments conducted on a large number of user's real behavior data show that the proposed models AgeBPR and AgeSocialMF gain a higher recommendation accuracy compared with models without using age information, suggesting that the proposed strategy, integrating user's age information into traditional recommendation models, is effective to improve the recommendation accuracy.

Key words: age prediction, implicit feedback, recommendation system, matrix factorization

摘要：

随着移动端设备的快速发展，手机应用呈爆炸式增长，如何在包含百万App的手机应用商城中将用户喜爱的App准确推荐给用户显得尤为重要。通过对艾瑞发布的多项移动行业报告以及对用户的调查分析，发现用户的年龄信息是影响用户选择App的因素之一。针对此现象，提出采用用户年龄信息与传统推荐算法相结合的方式来给用户推荐App。将预测用户年龄看成分类问题。利用用户在应用商城中的多项行为特征构建用户年龄预测模型。提出了两个基于用户年龄的推荐模型AgeBPR模型和AgeSocialMF模型。通过大量用户的真实数据集上的实验结果表明，提出的两个模型的推荐准确度较基准模型均有一定幅度的提升，说明提出的将用户年龄信息考虑到推荐模型中的有效性。

关键词: 年龄预测, 隐式反馈, 推荐系统, 矩阵分解