Bridging Semantic Gap Between App Names: Collective Matrix Factorization for Similar Mobile App Recommendation

With the increase of mobile apps, i.e. applications, it is more and more difficult for users to discover their desired apps. Similar app recommendation, which plays a critical role in the app discovering process, is of our main concern in this paper. Intuitively, name is an important feature to distinguish apps. So app names are often used to learn the app similarity. However, existing studies do not perform well because names are usually very short. In this paper, we explore the phenomenon of the ill performance, and dive into the underlying reason, which motivates us to leverage additional corpus to bridge the gap between similar words. Specifically, we learn app representation from names and other related corpus, and formalize it as a collective matrix factorization problem. Moreover, we propose to utilize alternating direction method of multipliers to solve this collective matrix factorization problem. Experimental results on real-world data sets indicate that our proposed approach outperforms state-of-the-art methods on similar app recommendation.