Abstract

Fuzzy cognitive map (FCM) is an important approach for modeling the behavior of dynamic systems. FCM’s ability to represent casual relationships between the concepts (factors, attributes, etc.) has attracted the interest of researchers from different disciplines. The construction process of FCMs is mostly initialized with expert knowledge because FCMs can conveniently incorporate available information and expertise in the determination of vital parameters and relations of the system. However, their higher dependence on expert knowledge may significantly influence the reliability of the model due to the increase in subjectivity. In order to avoid weaknesses depending on expert knowledge, learning algorithms that search for the appropriate relationships between the concepts have been used with FCM studies. In this paper, a FCM analysis was performed for tobacco use to understand the cause–effect relationships between demographic characteristics of people (such as gender, age range, and residence type) and likelihood to tobacco use. In order to reduce the impact of external interventions (from experts), a multi-stage learning procedure was applied by integrating two different learning algorithms (nonlinear Hebbian learning algorithm and extended Great Deluge algorithm). The results showed that the multi-stage learning procedure increased the accuracy of the model and provided more reliable maps for the studied system. They also proved that the multi-stage learning procedures can help to reduce the dependency to expert knowledge and improve the robustness of the study.

中文翻译：

---

基于多阶段学习的烟草使用模糊认知图

摘要

模糊认知图（FCM）是对动态系统行为进行建模的重要方法。FCM表示概念（因素，属性等）之间的随意关系的能力吸引了来自不同学科的研究人员的兴趣。FCM的构建过程主要是通过专家知识来初始化的，因为FCM可以方便地将可用信息和专业知识整合到确定重要参数和系统关系中。但是，由于主观性的提高，他们对专家知识的较高依赖可能会极大地影响模型的可靠性。为了避免依赖专家知识的弱点，FCM研究中使用了搜索概念之间适当关系的学习算法。在本文中，对烟草使用进行了FCM分析，以了解人们的人口特征（例如性别，年龄范围和居住类型）与吸烟可能性之间的因果关系。为了减少外部干预（来自专家）的影响，通过集成两种不同的学习算法（非线性Hebbian学习算法和扩展的Great Deluge算法）来应用多阶段学习过程。结果表明，多阶段学习过程提高了模型的准确性，并为所研究的系统提供了更可靠的映射。他们还证明了多阶段学习程序可以帮助减少对专家知识的依赖并提高研究的稳定性。