Learning Mobile App Embeddings Using Multi-task Neural Network

Bajaj, Ahsaas; Krishna, Shubham; Tiwari, Hemant; Vala, Vanraj

doi:10.1007/978-3-030-23281-8_3

Cited by 1 publication

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They proposed a probabilistic model, named AppLDA, to generate app representations while excluding noise in reviews. Inspired by their work, Bajaj et al [5] jointly modeled app descriptions and user reviews to evaluate their use in predicting other indicators like app categories and ratings. Then they proposed a multitask neural architecture to learn and analyzed the influence of apps' textual data to predict other categorical parameters.…”

Section: A Methods Of Using App's Metadatamentioning

confidence: 99%

“…Existing research proposes the use of automated categorization techniques to categorize apps. Considering that apps usually use text information to guide users to use their functions, a large part of researches [5]- [11] use natural language processing to categorize apps by mining app description information, user comments, and so on. However, many popular apps do not have adequate or accurate text descriptions while the newly released app does not have sufficient comments.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

DroidARA: Android Application Automatic Categorization Based on API Relationship Analysis

et al. 2019

View full text Add to dashboard Cite

An application (app) market with well-managed categorization will help users with app search and recommendation. Current categorization methods in app markets mainly rely on manual operation. Existing approaches for automatic Android app categorization suffer from low efficiency and low accuracy due to insufficient analysis of features, or inappropriate choice of features. This will mislead users to download unrelated apps and is not conducive to market stability maintenance. In this paper, we propose DroidARA, an efficient automatic categorization method for Android apps based on API relationships analysis. Considering that the app category can be characterized by API relationships which represent the combinations and links among APIs, we design a complete system to generate API call graphs, extract the API relationships information, transform them into feature vectors and train the classifier. Firstly, the API calls are obtained through static analysis to generate API call graphs that contain the relationships among APIs. A novel matrix structure as well as related algorithm are designed to extract the API relationships information from API call graphs. After that, the matrix is transformed into vector according to two feature selection methods we designed for strengthening the use of effective information in the API relationships. A convolutional neural network (CNN) model is then trained with labeled samples of such feature vectors. To validate the feasibility of DroidARA, we conduct several categorization experiments on 19949 real apps of Google Play. The results demonstrate that DroidARA can achieve an average 88.9% accuracy in categorizing the apps into 24 categories, which outperforms existing methods by 18.5%.INDEX TERMS Android, categorization, API relationship, static analysis.

show abstract

Section: A Methods Of Using App's Metadatamentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%