API Relationship Analysis Based Android Application Classification And Description

With the growing number and variety of Android applications,an application market with well-categorized and descriptive information management will help users search and recommend applications.At present,the classification of applications and the provision of descriptive information are mainly completed by manual operations.In the existing automated classification studies,the classification accuracy of the application is low due to improper feature selection or insufficient analysis of existing information.And in terms of descriptive information,there is still a lack of related automated research work.Incorrect classification or inaccurate descriptive information will cause users to download irrelevant applications,which is not conducive to maintaining market stability.Based on these situations,this paper proposes and implements an Android application classification and description technology based on API analysis.By using static analysis,the internal feature information of the application is extracted,and combined with the deep learning model,the application function classification technology and the application function descriptive word generation technology are implemented to provide a reference automated management method for the application market.The specific research contents are as follows:(1)An Android application classification method based on API relationships analysis is proposed.This method uses static analysis technology to generate the API call graph of the application program,so as to extract the API relationship information between functions and convert it into a novel matrix structure according to the designed algorithm.After that,two feature selection methods are applied to enhance the use of effective information in the features,transforming the matrix into a vector.Finally,application samples labeled by feature vectors and classes are used to train a convolutional neural network to build a classifier model.The experiment proves that the proposed method achieves an accuracy of 88.9%,which is better than the reference algorithms.(2)This paper proposes a method for generating function descriptive words for Android applications based on API combination analysis.This method first designs a word descriptive ability calculation method to select the words that can widely describe the functions of various applications,and then uses the set of these words as labels.A static analysis method was used to extract the API combination relationship in the application program,and a multi-label classification model was established in conjunction with the convolutional neural network algorithm to predict words capable of expressing multiple functions of the application program.In the experiment of generating 3 words,the precision is 80.8%,which has certain reference value...