Abstract:The rapid growth of mobile apps has significantly promoted the use of third-party libraries in mobile app development. However, mobile app developers are now facing the challenge of finding useful third-party libraries for improving their apps, e.g., to enhance user interfaces, to add social features, etc. An effective approach is to leverage collaborative filtering (CF) to predict useful third-party libraries for developers. We employed Matrix Factorization (MF) approaches - the classic CF-based prediction approaches - to make the predictions based on a total of 31,432 Android apps from Google Play. However, our investigation shows that there is a significant lack of diversity in the prediction results - a small fraction of popular third-party libraries dominate the prediction results while most other libraries are ill-served. The low diversity in the prediction results limits the usefulness of the prediction because it lacks novelty and serendipity which are much appreciated by mobile app developers. In order to increase the diversity in the prediction results, we designed an innovative MF-based approach, namely LibSeek, specifically for predicting useful third-party libraries for mobile apps. It employs an adaptive weighting mechanism to neutralize the bias caused by the popularity of third-party libraries. In addition, it introduces neighborhood information, i.e., information about similar apps and similar third-party libraries, to personalize the predictions for individual apps. The experimental results show that LibSeek can significantly diversify the prediction results, and in the meantime, increase the prediction accuracy.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to address the issue of low-power data transmission in wireless neural recording. Specifically, the paper proposes a deep learning framework based on Quantized Compressed Sensing (QCS), called BW-NQ-DNN, for efficient compression and recovery of neural signals. The main objectives are as follows: 1. **Sparse Representation Problem**: - Neural spikes are not sparse in commonly used dictionaries (such as Discrete Cosine Transform basis and Discrete Wavelet Transform basis), leading to significant performance degradation when using these dictionaries for recovery. - The paper proposes to directly learn a binary measurement matrix through deep learning to improve the effectiveness of sparse representation. 2. **Measurement Matrix Design Problem**: - Traditional random measurement matrices (such as Gaussian matrices or Bernoulli matrices) are effective but not optimal. - The paper trains a neural network to directly learn a binary measurement matrix, which maintains hardware implementation efficiency while improving performance. 3. **Quantization Problem**: - Uniform quantization is simple but performs poorly at low bit rates. - The paper introduces a non-uniform quantization method and optimizes the quantizer to adapt to the characteristics of neural signals, especially performing better in low bit rate scenarios. 4. **Efficient Recovery Algorithm Problem**: - Traditional algorithms (such as convex optimization, greedy algorithms, etc.) have high computational complexity, making real-time processing difficult. - The paper proposes using deep neural networks as recovery solvers, significantly reducing computational complexity and achieving efficient real-time recovery. ### Summary The main contribution of the paper is the proposal of a new deep learning framework, BW-NQ-DNN, which effectively addresses the problem of signal compression and recovery in wireless neural recording, especially suitable for low-power devices. By jointly optimizing the binary measurement matrix, non-uniform quantizer, and non-iterative recovery solver, this framework not only significantly reduces the number of transmission bits but also performs excellently in high compression ratio tasks, demonstrating high practical value.

Diversified Third-Party Library Prediction for Mobile App Development

Accurate Library Recommendation Using Combining Collaborative Filtering and Topic Model for Mobile Development.

Diversified Mobile App Recommendation Combining Topic Model and Collaborative Filtering

Large-Scale Third-Party Library Detection in Android Markets

iLibScope: Reliable Third-Party Library Detection for iOS Mobile Apps

A Novel Social Recommendation Method Fusing User’s Social Status and Homophily Based on Matrix Factorization Techniques

Automated Detection and Classification of Third-Party Libraries in Large Scale Android Apps

LibSeeker:Detecting Android Third-party Libraries Using Parameter Auto-tuning

JSLibD: Reliable and Heuristic Detection of Third-party Libraries in Miniapps

Version-sensitive Mobile App Recommendation

Understanding Third-Party Libraries in Mobile App Analysis

Personalized Mobile App Recommendation by Learning User's Interest from Social Media

User-Specific Rating Prediction For Mobile Applications Via Weight-Based Matrix Factorization

Weighted meta-graph based mobile application recommendation through matrix factorisation and neural networks

Mobile Application Rating Prediction Via Feature-Oriented Matrix Factorization

A Hybrid Approach Based on Collaborative Filtering to Recommending Mobile Apps

Edge-cloud-enabled Matrix Factorization for Diversified APIs Recommendation in Mashup Creation

Detecting third-party libraries in Android applications with high precision and recall

APPCOMMUNE: Automated Third-Party Libraries De-duplicating and Updating for Android Apps

DeepApp

Federated Privacy-preserving Collaborative Filtering for On-Device Next App Prediction