AC-Net: Assessing the Consistency of Description and Permission in Android Apps

Abstract

With Android applications (apps) becoming increasingly popular, there exist huge risks lurking in the app marketplaces as most malicious software attempt to collect users' private information without their awareness. Although these apps request users' authorization for permissions, the users can still face privacy leakage issues due to their limited knowledge in distinguishing permissions. Thus, accurate and automatic permission checking is necessary and important for users' privacy protection. According to previous studies, analyzing app descriptions is a helpful way to examine whether some permissions are required for apps. Different from those studies, we consider app permissions from a more fine-grained perspective and aim at predicting the multiple correspondent permissions to one sentence of app description. In this paper, we propose an end-to-end framework for assessing the consistency between descriptions and permissions, named Assessing Consistency based on neural Network (AC-Net). For evaluation, a new dataset involving the description-to-permission correspondences of 1415 popular Android apps was built. The experiments demonstrate that AC-Net significantly outperforms the state-of-the-art method by over 24.5% in accurately predicting permissions from descriptions. INDEX TERMS Android security, app descriptions, app permissions, consistency assessment, text classification, deep learning.

FCDP: Fidelity Calculation for Description-to-Permissions in Android Apps

“…They did not quantify the fidelity of description-to-permissions, which cause that the stakeholders are difficult to utilize the results of those works for further decision. Among the aforementioned methods, AC-Net [36] is closest to our present study. However, unlike AC-Net, our proposed method captures the semantics of app descriptions using the attention mechanism in deep learning.…”

“…For instance, the word "contact" can be a verb, as in the sentence "contact with us by email," whereas it is a noun in the sentence "add friends from your contacts," indicating the usage of the permission READ_CONTACTS. Feng et al proposed AC-Net [36], which uses the GRU model to predict permissions from app descriptions. It covers 16 dangerous permissions, which is currently the maximum number of labeled permissions.…”

“…Android Package Kit (APK) files are decompiled using apktool 3 to obtain the AndroidManifest.xml file, in which the requested permissions are declared. In this study, we predict 11 permission groups (involving 16 sensitive permissions as in [36]) from app descriptions, as shown in Table 2. The end-users concern these 16 permissions which allow the app to access sensitive data and operations [38].…”

“…It will be fed into deep learning model for word embedding. As manually labeling description-topermissions is labor-intensive, we selected the labeled dataset of AC-Net provided by Feng et al [36] to train our model. It contains 1,415 unique apps with 24,726 sentences.…”

“…To evaluate the effectiveness of FCDP in predicting permissions from app descriptions, we compared it with the following deep learning models: AC-Net [36], Bi-LSTM, Convolutional Neural Networks for text classification (textCNN) [48], Hierarchical Attention Networks (HAN) [24] and Recurrent Convolutional Neural Networks (RCNN) [49]. To perform a valid comparison, we set identical hyper-parameters for each model, including corpus for training word vectors, the dimensionality of word embeddings, and different random seeds for shuffling and splitting datasets.…”