Comparison of Text-Based and Feature-Based Semantic Similarity Between Android Apps

Textual description of mobile apps comprised of apps' core functionality and describes the key features of any given app. Finding semantically similar apps help app users searching and looking for alternatives in the app stores as well as help the developers learn from their similar apps. A number of works have been done on apps' text-based semantic similarity. However, no attempts have been taken yet to find similar apps based on semantic similarity of their features. Moreover, varying length of textual descriptions affect semantic approach significantly. In this paper, we employ two different semantic similarity techniques to both the text-based and feature-based app descriptions. Together we investigate four different techniques and compare the similarities generated between the given android apps. We compare our results with a truth-set built on 50 similar and 50 non-similar apps in the same domain and across the other domains respectively. Experimental results demonstrate that feature-based approach generate better results in determining app similarity in the same domain and across the other domains irrespective of the length of apps' textual descriptions.