Scholarlink

Nowadays, smartphones, tablets and car user interfaces are full of various applications that enable us to perform a wide variety of tasks. To increase comfort but also safety in some specific contexts such as driving, it is relevant to determine at any instant which is the next application that the user will use to facilitate the access to the predicted application. But if a user started very recently to use the device, we have very little data about him or her. Therefore, the Machine Learning (ML) model used to suggest the next application could quickly overfit and its suggestions would be inaccurate. To mitigate this problem, we propose an approach based on Data Augmentation (DA) using a variant of Generative Adversarial Networks (GAN) called DoppelGANger. Our ML model used to predict the next application is a Deep Neural Network (DNN) based on Long-Short term memory (LSTM) units. By adding the synthetic data generated by the GAN to the original training data, we obtain a mean improvement of 4.7% for the Top-3 Precision and 14.1% for the Top-3 Recall.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Next Application Prediction Enhanced for New Users by Using Data Augmentation with Generative Adversarial Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Antoine Foricher,
  • Andreas Heimrath,
  • Oliver Meyer,
  • Nicole Beringer

摘要

Nowadays, smartphones, tablets and car user interfaces are full of various applications that enable us to perform a wide variety of tasks. To increase comfort but also safety in some specific contexts such as driving, it is relevant to determine at any instant which is the next application that the user will use to facilitate the access to the predicted application. But if a user started very recently to use the device, we have very little data about him or her. Therefore, the Machine Learning (ML) model used to suggest the next application could quickly overfit and its suggestions would be inaccurate. To mitigate this problem, we propose an approach based on Data Augmentation (DA) using a variant of Generative Adversarial Networks (GAN) called DoppelGANger. Our ML model used to predict the next application is a Deep Neural Network (DNN) based on Long-Short term memory (LSTM) units. By adding the synthetic data generated by the GAN to the original training data, we obtain a mean improvement of 4.7% for the Top-3 Precision and 14.1% for the Top-3 Recall.