A Comprehensive Taxonomy for Explainable Artificial Intelligence: A Systematic Survey of Surveys on Methods and Concepts

Gesina Schwalbe,Bettina Finzel

DOI: https://doi.org/10.1007/s10618-022-00867-8

2023-01-09

Abstract:In the meantime, a wide variety of terminologies, motivations, approaches, and evaluation criteria have been developed within the research field of explainable artificial intelligence (XAI). With the amount of XAI methods vastly growing, a taxonomy of methods is needed by researchers as well as practitioners: To grasp the breadth of the topic, compare methods, and to select the right XAI method based on traits required by a specific use-case context. Many taxonomies for XAI methods of varying level of detail and depth can be found in the literature. While they often have a different focus, they also exhibit many points of overlap. This paper unifies these efforts and provides a complete taxonomy of XAI methods with respect to notions present in the current state of research. In a structured literature analysis and meta-study, we identified and reviewed more than 50 of the most cited and current surveys on XAI methods, metrics, and method traits. After summarizing them in a survey of surveys, we merge terminologies and concepts of the articles into a unified structured taxonomy. Single concepts therein are illustrated by more than 50 diverse selected example methods in total, which we categorize accordingly. The taxonomy may serve both beginners, researchers, and practitioners as a reference and wide-ranging overview of XAI method traits and aspects. Hence, it provides foundations for targeted, use-case-oriented, and context-sensitive future research.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The problems that this paper attempts to solve are as follows: With the rapid growth of Explainable Artificial Intelligence (XAI) methods, researchers and practitioners need a comprehensive classification system to understand, compare, and select XAI methods suitable for specific application scenarios. Specifically, this paper aims to:

1. **Provide a comprehensive XAI method classification system**: Help beginners obtain a good initial overview and provide a starting point for more in - depth research.
2. **Support practitioners in selecting appropriate XAI methods**: By providing a classification scheme, enable practitioners to select appropriate XAI methods according to specific use cases.
3. **Assist researchers in identifying new research directions**: Help researchers discover the combined characteristics of XAI methods that have not been or rarely been considered.

To achieve these goals, the authors conducted extensive literature analysis and meta - research, synthesizing more than 50 of the most frequently cited and the latest review articles on XAI methods, metrics, and method characteristics. Eventually, they constructed a structured and detailed XAI method classification system and illustrated it with more than 50 different example methods.

### Main contributions of the paper

- **Complete XAI method classification system**: Provided a structured, detailed, and in - depth classification system of XAI method characteristics (see Figure 7). In particular, this classification system covers all application - related XAI methods and evaluation characteristics mentioned in the literature so far.

- **Extensive XAI method meta - research**: Reviewed more than 50 articles on XAI - related topics, which can be used as a starting point for in - depth understanding of XAI and its sub - fields. To the best of the authors' knowledge, this is the most extensive and detailed meta - research specifically for XAI and XAI methods currently.

- **Extensive XAI method review**: Collected, reviewed, and classified more than 50 carefully selected and diverse XAI methods (see Table 8). This proves the practical applicability of their classification system structure and the identified method characteristics.

### Specific methods for solving problems

1. **Systematic literature analysis**: Conduct keyword searches through databases such as Google Scholar and screen out literature related to the XAI classification system.
2. **Meta - research**: Conduct a detailed review of the screened - out literature, summarize and combine terms and concepts to form a unified classification system.
3. **Classification system construction**: Construct a structured classification system according to the method characteristics in the literature and illustrate it with specific XAI method examples.

### Conclusion

This paper promotes the further development of the XAI field by providing a comprehensive XAI method classification system, which helps researchers, practitioners, and beginners better understand and select XAI methods.