Name
Title | ||
|---|---|---|
Explainable Visualization For Interactive Exploration Of Cnn On Wikipedia Vandal Detection |
Abstract | ||
|---|---|---|
Machine learning (ML) algorithms have greatly improved the performances of many computationally expensive operations, including detection of frauds and attacks for security applications. However, these algorithms are often hard for users to interpret or modify due to their hidden processes. This paper presents an explainable visualization approach to provide a set of developer tools that enable users to apply ML algorithms efficiently without requiring users to understand the algorithms completely. We use an algorithm of convolutional neural network (CNN) for Wikipedia vandal detection as an example. Our highly coordinated visualizations serve as a visual analytics interface for developers to analyze the behaviors of algorithms and large-scale data. It enables users to study hidden relationships among the spaces of data, algorithm parameters, and results that are essential to understand the CNN detection mechanism. We provide several case studies to demonstrate how our approach can be used to explore the clusters of parameter space, identify outlier of parameters or data, study the parameter sensitivity ranges, and select suitable ranges of parameter for combined criteria (e.g., covering selected important cases and maintaining relative high success ratio) that are hard to describe with single objective functions. Due to the complexity of the problem, we also discuss the limitations of our approach and future work. |
Year | DOI | Venue |
|---|---|---|
2019 | 10.1109/BigData47090.2019.9006128 | 2019 IEEE INTERNATIONAL CONFERENCE ON BIG DATA (BIG DATA) |
Keywords | Field | DocType |
Explainable visualization, CNN, vandal detection, Wikipedia user behavior | Data mining,Computer science,Visualization,Convolutional neural network,Visual analytics,Outlier,Artificial intelligence,Parameter space,Single objective,Machine learning | Conference |
ISSN | Citations | PageRank |
2639-1589 | 0 | 0.34 |
References | Authors | |
0 | 2 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Zerong Liu | 1 | 0 | 0.34 |
| Aidong Lu | 2 | 353 | 30.18 |