基于元路径嵌入的移动应用需求偏好分析方法

随着互联网和移动应用平台的快速发展,围绕移动应用所产生的海量用户数据已经成为精确分析用户需求偏好的重要数据源.尽管已有不少学者从这些数据中分析和挖掘用户需求,但现有的方法通常只研究了数据的少数维度的特征,未能有效地挖掘多维移动应用信息以及他们之间的关联.提出一种基于元路径嵌入的移动应用需求偏好分析方法,能够为用户进行个性化移动应用推荐.具体地,首先分析移动应用的文本信息中的语义主题,挖掘用户需求偏好的分析维度.其次,将移动应用信息的语义特征构建了一个融合移动应用多维信息的概念模型,涵盖了能够表征用户需求偏好的多维度数据.基于概念模型的语义,设计了一组有意义的元路径集合,以精确地捕捉用户需求偏好的语义.最后,通过使用元路径嵌入技术进行用户行为画像,进而实现个性化的移动应用推荐.使用苹果应用商店包括1507个移动应用和153501条用户评论的真实数据集进行实验评估.实验结果表明所提的方法在各指标上均优于现有模型,其中平均F1值提升0.02,平均归一化折损累计增益(normalized discounted cumulative gain,NDCG)提升0.1.

A User Requirements Preference Analysis Method of Mobile Applications Based on Meta-Path Embedding

With the rapid development of the Internet and mobile application platforms,massive user data has been generated by mobile applications.Such data has become an important data source for accurately analyzing user requirements preference.Many researchers have analyzed and mined user requirements preference from user data.However,the existing studies do not link the multi-dimensional information of mobile applications,and only explore the characteristics of a few dimensions of the data.In this paper,we propose a method to analyze user requirements preferences based on meta-path embedding,which can personally recommend mobile applications for users.Specifically,we first analyze the semantic topics in the text information of mobile applications,which enriches the analysis dimension of user requirements preferences.Second,we construct a conceptual model that integrates multi-dimensional information for mobile applications,including multi-dimensional data that affects user choices.Based on the conceptual model,we design a series of meaningful meta-paths to accurately capture the semantics of user requirements preferences.Finally,we analyze user preferences based on the meta-path embedding technique to recommend personalized mobile applications for users.In this paper,we use the real data set obtained from the Apple App Store to evaluate our model,which contains 1507 mobile applications and 153501 user reviews.The experimental results show that our method outperforms the existing models in all metrics,in which the average F1-measure increases by 0.02,and the average NDCG increases by 0.1.