Table of Content

  

20 May 2024
Volume 59 Issue 5

Multi-label learning based on granular neural networks

CHEN Yumin, ZHENG Guangyu, JIAO Na

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  1-11.  doi:10.6040/j.issn.1671-9352.7.2023.239

Abstract ( 307 )   HTML ( 4 )   PDF (2876KB) ( 403 )   Save

References | Related Articles | Metrics

This paper introduces the theory of granular computing and proposes a multi-label learning method based on granular neural networks. This method utilizes similarity granulation to capture the structural correlations in the data. Samples are granulated into granules on individual features, and granules across multiple features form granule vectors. Operations on granules and granule vectors are defined. On this basis, a granular loss function is introduced and a granular neural network is constructed for multi-label learning. Experiments are conducted on multiple Mulan multi-label datasets and compared with existing multi-label classification algorithms across various evaluation metrics. The results demonstrate the effectiveness and feasibility of the granular neural network multi-label learning algorithm.

Neighborhood recommendation algorithm based on three-way causality force

FAN Min, QIN Qin, LI Jinhai

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  12-22.  doi:10.6040/j.issn.1671-9352.7.2023.0001

Abstract ( 174 )   HTML ( 1 )   PDF (2425KB) ( 138 )   Save

References | Related Articles | Metrics

A neighborhood recommendation algorithm under three-way causality force by combining the three-way decision ideas, causal force theory, and formal concept analysis are proposed. Considering the influence of extreme user rating on recommendation accuracy, we classify users by defining the degree of leniency and severity, and correct extreme user rating. Based on the modified score matrix, the three cosine similarity and the similarity structure importance of nodes are calculated to find the expert nodes. Under the objective function and constraint conditions that the weak concept of the object needs to meet, the cluster is carried out to obtain the neighborhood, and the key conditional attributes and decision attributes are identified according to the attribute density in the neighbourhood, and the confidence between them is calculated. The three-way causality force extraction recommendation rules are combined to carry out neighborhood recommendation for community members. The experimental results show that the proposed algorithm is significantly better than other traditional recommendation algorithms in terms of accuracy, recall, and F1.

Global and local relationships based on multi-label classification algorithm with label-specific features

ZHANG Shandan, WENG Wei, XIE Xiaozhu, WEI Bowen, WANG Jinbo, WEN Juan

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  23-34.  doi:10.6040/j.issn.1671-9352.7.2023.082

Abstract ( 205 )   PDF (2299KB) ( 192 )   Save

References | Related Articles | Metrics

To address the problem of neglecting the second-order label relation in the local label correlation, we propose a new algorithm called global and local relationships based on multi-label classification algorithm with label-specific features(LFGML). Specifically, the label-specific features are firstly obtained through the perspective of global relations, then the neighbourhood information of each instance is calculated using the weighted average method. The second-order label relationship in the local relationship are extracted using Jaccard similarity. The LFGML algorithm is tested on ten multi-label datasets: Genbase, Medical, Arts, Health, Flags, Cal500, Yeast, Image, Education and Emotions. The results demonstrate that our proposed algorithm outperforms other comparison algorithms in multi-label classification.

A neighbourhood granular fuzzy C-means clustering algorithm

ZHENG Chenying, CHEN Yingyue, HOU Xianyu, JIANG Lianji, LIAO Liang

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  35-44.  doi:10.6040/j.issn.1671-9352.7.2023.343

Abstract ( 186 )   HTML ( 0 )   PDF (5985KB) ( 288 )   Save

References | Related Articles | Metrics

Aiming at the problem that the sensitivity of initial value and noise lead to the decline of fuzzy C-means clustering, fuzzy C-means clustering method of neighborhood granule is proposed by introducing the theory of granular computation and using the neighborhood granulation technique. In the sample, the neighborhood granule is constructed by using the neighborhood granulation technique on single feature, and the neighborhood granular vector is formed by using granulation on multi-features.A variety of granule distance formulas are defined to measure the distance between granules. According to the granule distance measurement, a granular fuzzy C-means clustering method is proposed, and a granular fuzzy C-means clustering algorithm is designed. Multiple data sets are used to perform experiments, and the fuzzy C-means clustering algorithm is compared with the classical clustering algorithm. The results verify the feasibility and effectiveness of the proposed neighborhood granular fuzzy C-means clustering method.

Perturbation three-way clustering based on natural nearest neighbors

ZHU Jin, FU Yu, GUAN Wenrui, WANG Pingxin

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  45-51.  doi:10.6040/j.issn.1671-9352.4.2023.137

Abstract ( 140 )   HTML ( 1 )   PDF (3877KB) ( 109 )   Save

References | Related Articles | Metrics

By using samples natural nearest neighbors, a three-way clustering algorithm is proposed based on samples perturbation theory. The proposed algorithm combines natural nearest neighbor information with samples perturbation to generate two datasets. By randomly selecting parts of the samples feature, different clustering results are obtained through K-means clustering algorithms. The stability of each sample is calculated based on the defined frequencies. The universe is divided into stable set and unstable set based on the samples stability. Then, we use different strategies to obtain the core region and fringe region of each cluster. The testing results on five open datasets verify the effectiveness of the proposed algorithm through comparative tests with two traditional clustering methods.

Multi-granularity rough set attribute reduction algorithm based on optimized discernibility matrix

SONG Suyang, YE Jun, ZENG Guangcai, SUN Qing

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  52-62.  doi:10.6040/j.issn.1671-9352.0.2023.398

Abstract ( 171 )   HTML ( 1 )   PDF (5146KB) ( 239 )   Save

References | Related Articles | Metrics

An improved multi-granularity attribute reduction algorithm based on optimized discernible matrix is proposed to solve the problem of excessive computation for constructing discernible matrix in multi-granularity rough sets. Attribute importance is used as similarity to construct different particle size spaces, and kernel attributes in the optimized discernibility matrix of each particle size space are output to solve the final reduction, and reverse redundancy detection is performed on the reduced set to avoid redundant attributes. The results show that this algorithm can effectively reduce the time complexity and improve the reduction efficiency. Examples and experimental results of several UCI data sets demonstrate the effectiveness of the proposed algorithm.

Rule extraction based on linguistic concept lattice with fuzzy object

WU Jiang, LIU Deshan,YU Yingying, PANG Kuo, LI Xiaofeng

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  63-69.  doi:10.6040/j.issn.1671-9352.7.2023.188

Abstract ( 159 )   HTML ( 0 )   PDF (4048KB) ( 86 )   Save

References | Related Articles | Metrics

Aiming at the problem of linguistic decision-making information processing under uncertain environment, a method of rule extraction with confidence in inconsistent linguistic formal decision context with fuzzy object is proposed. The inconsistent linguistic formal decision context with fuzzy object is established to express the linguistic evaluation information of experts hesitation between adjacent linguistic values. Based on the finer relation between linguistic concept lattices with fuzzy object, the consistency of linguistic formal decision context with fuzzy object is discussed. The definition of the confidence in rules is given. A method of rule extraction in inconsistent linguistic formal decision context with fuzzy object is proposed based on the confidence of linguistic rule.The effectiveness and practicability of the proposed method are illustrated by an example of predicting the overall performance of school students and an experiment.

Online multi-label feature selection based on sub-correlation features and neighborhood mutual information

CHENG Yuxuan, MAO Yu, ZHANG Xiaoqing, ZENG Yixiang, LIN Yaojin

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  70-81.  doi:10.6040/j.issn.1671-9352.7.2023.4523

Abstract ( 161 )   HTML ( 0 )   PDF (7674KB) ( 149 )   Save

References | Related Articles | Metrics

To fully mine the features neglected by the single metric algorithm but beneficial to the classifier, this paper proposes an online multi-label feature selection algorithm based on sub-correlation features and neighborhood mutual information. By calculating the importance and correlation of newly arrived features, the difference between the significance of new features is analyzed, and the features are divided into salient features and sub-correlation features. Redundancy analysis is performed on newly arrived features and selected feature sets using neighborhood interaction information, and features with low dependencies are eliminated, to gradually improve the quality of feature subsets. This paper also constructs a measurement index based on the global linear and nonlinear relationship and uses it to calculate the local correlation of features, effectively mining the sub-correlation features. Strip the sub-correlation features from the feature set and save them separately, so that they will not be eliminated from the feature set during the redundancy analysis stage due to the high sensitivity of the salient features to the measurement index. Using established feature selection indicators and iterative strategies to select features according to the indicators. Experimental results show that the proposed algorithm has good effectiveness and stability.

Knowledge reduction in decision set-valued systems

FANG Fengqi, WU Weizhi

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  82-89.  doi:10.6040/j.issn.1671-9352.7.2023.384

Abstract ( 180 )   HTML ( 0 )   PDF (4231KB) ( 212 )   Save

References | Related Articles | Metrics

To solve the problem of knowledge reduction in data sets with a set-valued decision, several types of decision systems such as decision set-valued systems, certainty decision set-valued systems and propensity decision set-valued systems are first defined. A comparative study is then discussed on decision set-valued systems and several relevant types of information systems, and characteristics of decision set-valued systems are clarified. Finally, combined with the three-way decision method, notions of single-valued reducts and multi-valued reducts in decision set-valued systems are proposed and a method for the computation of reducts in decision set-valued systems is explored. The results show that the method can effectively extract information on certainty decision set-valued systems.

Multi-label online stream feature selection based on high-dimensional correlation

ZHU Liquan, LIN Yaojin, MAO Yu, CHENG Yuxuan

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  90-99.  doi:10.6040/j.issn.1671-9352.7.2023.148

Abstract ( 170 )   HTML ( 1 )   PDF (6305KB) ( 211 )   Save

References | Related Articles | Metrics

This paper proposes a multi-label online stream feature selection algorithm based on high-dimensional correlation. The algorithm employs an equivalent mapping of the label space and constructs a weighted undirected graph based on the high-dimensional label space. It utilizes graph information and Jaccard index to measure the high-dimensional weights between labels. The significance of newly arrived features is calculated based on the high-dimensional correlation of the labels, and the significance level of new features is determined through iterative mean significance. Furthermore, a balanced global and local online feature selection algorithm is designed to dynamically optimize the selected feature subset by considering the global correlation between the selected features and the label space, thereby filtering out irrelevant features. Redundant features are eliminated by analyzing the local correlation among the selected features. The testing results validate the effectiveness of the proposed algorithm through comparative tests with six other multi-label feature selection methods.

Feature selection for partial label learning based on neighborhood rough sets

GAO Hefei, LI Yan, WANG Shuo

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  100-113.  doi:10.6040/j.issn.1671-9352.7.2023.4204

Abstract ( 173 )   HTML ( 0 )   PDF (7771KB) ( 104 )   Save

References | Related Articles | Metrics

A feature selection method for partial label learning based on neighborhood rough sets is proposed. A partial label neighborhood decision system is constructed, and the concepts of lower approximation and dependency of neighborhood rough sets are then defined in partial label learning. On this basis, a feature selection algorithm suitable to partial label classification is developed. This method can measure the similarity between labels in the set of candidate labels while granulating the feature space in the neighborhood, and select a subset of features with strong relevance to the label information. Two generation mechanisms for false positive candidate labels are used which are different from the most often used random method, and their impact on the results are compared and analyzed in the experiments. Finally, extensive experimental results on six real-world and six controlled synthetic partial label data sets are presented to demonstrate the effectiveness of the proposed feature selection method.

Covering rough fuzzy sets and optimal scale selection in multi-scale decision systems

SHI Hongyi, MA Zhouming

JOURNAL OF SHANDONG UNIVERSITY(NATURAL SCIENCE). 2024, 59(5):  114-130.  doi:10.6040/j.issn.1671-9352.7.2023.380

Abstract ( 204 )   HTML ( 0 )   PDF (12546KB) ( 300 )   Save

References | Related Articles | Metrics

This paper extends the covering rough fuzzy set model by considering the membership degree of objects in the decision attribute in the neighborhood of the objects minimum description. Two different covering rough fuzzy set are proposed. Based on this, four covering rough set models in multi-scale decision systems are constructed by combining covering rough fuzzy sets with multi-scale decision systems. The corresponding positive region and attribute importance are defined, and the optimal scale selection algorithm is designed. Finally, comparative experiments compare the difference in regression prediction performance between the optimal scale selected by the four covering rough fuzzy set models in such multi-scale decision systems and the original scale. The experimental results indicate that the optimal scale combination selected by model four of covering rough fuzzy sets in multi-scale decision systems can effectively improve the predictive ability of the regression model.