Common but Innovative: Learning Features From Apps in Different Domains

Guohang Song,Liang Hu,Yuzhou Liu

doi:10.1109/access.2020.3029262

Guohang Song, Liang Hu + Show 1 more

Open Access

https://doi.org/10.1109/access.2020.3029262

Copy DOI

Abstract

With the popularity of Apps, the products in one domain become more and more similar to each other, and developers start to find the break from other domains. However, facing the large-scale data resource in App stores, it is difficult to identify the related domains, let alone gain useful features from the products in them. In this paper, we propose an approach to help developers learn information of features related to their App from the products in different domains. Firstly, we provide the method to extract features as well as their relationships from App descriptions to describe one domain. Then, the similar features shared by different domains are identified as the bridges for searching the potential information which may be re-used by the developers. Finally, we provide the framework of an interactive recommendation system to let developers gain and understand the information easily. To evaluate our approach, we conducted experiments based on the dataset on Google Play. The results show that the average precision of our approach for finding similar features between different domains can reach 82.38%, and the survey on developers indicate that the information recommended by our approach is useful for updating Apps and inspiring developers generate innovative ideas.

Highlights

With the development of smartphones, mobile applications (Apps) have become one of the most common software in our daily life
Since many methods used in the approach are validated in our previous work, such as the method for extracting features from App descriptions and the one for clustering features, we only need to discuss other steps and there are mainly three research questions here: RQ1: Whether our approach can find similar features shared by different domains?
By further analyzing the results, it can be seen that if the two domains can cover more extensive features, our method can find similar features between them more accurately, for example, the precision between domain ‘‘Social’’ and ‘‘Shopping’’ can reach highest value 96.67%; While if the two domains focus on more concrete functionalities separately, it becomes more difficult to find the features share by them, such as the precision between ‘‘Food’’ and ‘‘Video’’ is only 70%

Summary

INTRODUCTION

With the development of smartphones, mobile applications (Apps) have become one of the most common software in our daily life. The prosperity of App market brings huge economic benefit, and generates increasingly fierce competition [4]–[6] In such condition, the developers have to update their products continually to maintain existing users and attract new ones. We have proposed approaches to mine App features from the descriptions [15] and use them as the requirements for the App development [16], [17] in our previous work, but we still only recommend useful/popular features from the Apps in the same domain for updating developers’ product. The results indicate that our approach can recommend useful information from different domains for updating Apps and inspiring developer to generate new ideas.

RELATED WORK

ANALYSIS OF APP FEATURE IN ONE DOMAIN

IDENTIFY RECOMMENDABLE INFORMATION

FEATURE RECOMMENDATION

PRESENTING THE RECOMMENDATION INFORMATION

VIII. CONCLUSION AND FUTURE WORK