Semantic Learning Approaches Based On Axiomatic Fuzzy Set Theory

With the development of artificial intelligence,many industries encounter new opportunities and challenges.Artificial intelligence models are used to assist or replace human decisionmaking,but in some cases,companies need to understand the reason why such decisions are made,which makes explainable artificial intelligence increasingly attractive.Axiomatic fuzzy set(AFS)theory,as a tool to extract semantic information effectively from data,provides a new framework for explainable artificial intelligence.In this thesis,supervised learning,semisupervised learning,and unbalanced data analysis are studied on the basis of the AFS theory,and the research mainly includes the following topics:(1)A novel interpretable classifier is designed based on the AFS theory and the concept of entropy,which is capable of achieving sound classification performance and interpretability.Besides,the interpretable classifier provides a new framework of classifier design that can adapt more human-oriented recognition mechanisms.The existing AFS-based classifiers are weak in obtaining the optimal semantic description.To address this drawback,a new measure based on the AFS theory,named semantic entropy extended in Shannon's entropy,is developed to evaluate the discriminatory capabilities of semantic descriptions for each category.Moreover,an evaluation index is used to prune descriptions to deliver a promising performance.Compared to the previous AFS-based classifiers,the proposed approach offers a semantic entropy to measure the information derived from semantic descriptions of data rather than filter category descriptions by thresholds,so that the optimal semantic description of each class can be obtained.For the purpose of illustrating the effectiveness of the classifier,several data sets are utilized to facilitate a comparative analysis of the proposed approach and other state-of-the-art classifiers.The experimental studies demonstrate that the proposed approach can achieve the semantic descriptions of each class and the performance of the proposed approach is comparable with the performance of other approaches.(2)A semantic semi-supervised learning(Semantic SSL)approach is proposed based on the AFS theory,which is targeted at unifying two machine learning paradigms in a mutually beneficial way,where classical support vector machine(SVM)learns to reveal primitive logic facts from data,while the AFS theory is utilized to exploit semantic knowledge and correct the wrongly perceived facts for improving the machine learning model.It is known that disagreement-based semi-supervised learning can be viewed as an excellent schema so that a cotraining approach with SVM and the AFS theory can be utilized to improve the resulting learning performance.Compared with other semi-supervised approaches,the proposed approach can build a structure to reflect data distributed information with unlabeled data and labeled data,so that the hidden information embedded in both labeled and unlabeled data can be sufficiently utilized and can potentially be applied to achieve good descriptions of each category.Experimental results demonstrate that this approach can offer a concise,comprehensible,and precise semi-supervised learning frame,which strikes a balance between interpretability and accuracy.(3)In applications,the choice of wart treatment is studied based on the AFS theory.Wart disease is a kind of skin illness that is caused by Human Papillomavirus(HPV).Many medical studies are being carried out with the aid of machine learning and data mining techniques to find the most appropriate and effective treatment for a specific wart patient.However,the imbalanced distribution of medical data may lead to misclassification in this field.In this thesis,a Synthetic Minority Over-sampling(SMOTE)method is adopted to deal with the unbalanced data and combined with the AFS theory to predict whether patients can respond to treatment or not.Compared with other existing approaches,the proposed approach can provide descriptive information of the patients which can help to predict the response towards the treatment.Furthermore,the proposed approach can assist doctors in treatment,save medical resources,and improve the quality of treatment.