Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217514
D. O. Orina, R. Rimiru, W. Mwangi
The increasing competition in the banking industry has made customer churn analysis and prediction a crucial concern. Banks must now adopt customer retention strategies while also working towards acquiring new customers to expand their market share. In today’s business environment, machine learning techniques and algorithms are crucial for banks because acquiring new customers is more expensive than retaining existing ones. This proposed research aims to compare many supervised machine learning algorithms, and based on the experimental results, suggest the best-suited model for predicting customer churn. The process involves cross-validation, balancing data using the SMOTE algorithm, and utilizing both simple machine algorithms and ensemble methods for modeling. The experiments conducted in this study revealed that the random forest model performed the best, achieving an accuracy of 88%, an area under the curve (AUC) of 0.85, and an f1-score of 0.85 when using balanced data. This result is consistent with related research considered in this paper, which has demonstrated random forest as one of the most effective algorithms for customer predictive classification issues. Feature importance analysis from the optimization models indicated that the difference between depositing and withdrawing was the most significant attribute, while the maximum deposit per product had the least significance. The data mining techniques proposed to be used in this research include Decision Tree, Neural Networks, Support Vector Machine, Logistic Regression, Random Forest, XG-Boost, Ada-Boost, and K-Nearest Neighbor.
{"title":"A Comparative Study of Predictive Data Mining Techniques for Customer Churn in the Banking Industry","authors":"D. O. Orina, R. Rimiru, W. Mwangi","doi":"10.1109/IMSA58542.2023.10217514","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217514","url":null,"abstract":"The increasing competition in the banking industry has made customer churn analysis and prediction a crucial concern. Banks must now adopt customer retention strategies while also working towards acquiring new customers to expand their market share. In today’s business environment, machine learning techniques and algorithms are crucial for banks because acquiring new customers is more expensive than retaining existing ones. This proposed research aims to compare many supervised machine learning algorithms, and based on the experimental results, suggest the best-suited model for predicting customer churn. The process involves cross-validation, balancing data using the SMOTE algorithm, and utilizing both simple machine algorithms and ensemble methods for modeling. The experiments conducted in this study revealed that the random forest model performed the best, achieving an accuracy of 88%, an area under the curve (AUC) of 0.85, and an f1-score of 0.85 when using balanced data. This result is consistent with related research considered in this paper, which has demonstrated random forest as one of the most effective algorithms for customer predictive classification issues. Feature importance analysis from the optimization models indicated that the difference between depositing and withdrawing was the most significant attribute, while the maximum deposit per product had the least significance. The data mining techniques proposed to be used in this research include Decision Tree, Neural Networks, Support Vector Machine, Logistic Regression, Random Forest, XG-Boost, Ada-Boost, and K-Nearest Neighbor.","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114179686","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217655
Abdelrahman Ezzeldin Nagib, M. Saeed, Shereen Fathy El-Feky, Ali Khater Mohamed
The rapid spread of the COVID-19 pandemic has created a pressing urgent need for accurate and efficient diagnostic tools. Recently, Convolutional neural networks (CNN) have shown great potential in classifying COVID-19 infected cases from X-ray images, but the choice of weight initialization technique plays a crucial role in their performance of the Convolutional neural networks. In this research Paper, comparative study of different weight initialization techniques COVID-19 in the context of COVID-19 classification, performance evaluation techniques have been implemented such as Glorot, Orthogonal, and Random Uniform and results shows that that the Random Uniform initialization technique outperforms other weight initialization techniques in terms of overall classification accuracy. Keywords: COVID-19, Convolutional Neural Networks, X-ray images, Weight initialization, Classification
{"title":"A Comparative Study of Weight Initialization Techniques for Convolutional Neural Networks in COVID-19 Classification from X-ray Images.","authors":"Abdelrahman Ezzeldin Nagib, M. Saeed, Shereen Fathy El-Feky, Ali Khater Mohamed","doi":"10.1109/IMSA58542.2023.10217655","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217655","url":null,"abstract":"The rapid spread of the COVID-19 pandemic has created a pressing urgent need for accurate and efficient diagnostic tools. Recently, Convolutional neural networks (CNN) have shown great potential in classifying COVID-19 infected cases from X-ray images, but the choice of weight initialization technique plays a crucial role in their performance of the Convolutional neural networks. In this research Paper, comparative study of different weight initialization techniques COVID-19 in the context of COVID-19 classification, performance evaluation techniques have been implemented such as Glorot, Orthogonal, and Random Uniform and results shows that that the Random Uniform initialization technique outperforms other weight initialization techniques in terms of overall classification accuracy. Keywords: COVID-19, Convolutional Neural Networks, X-ray images, Weight initialization, Classification","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114388312","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217428
Maroua Louail, Chafia Kara-Mohamed alias Hamdi-Cherif
Text classification (TC) is the process by which the computer has the ability to provide a label to a given text based on its content. Term Frequency–Inverse Document Frequency (TF-IDF) is one of the popular methods used for Arabic text representation. The high number of dimensions and sparseness are among the main issues faced by the TF-IDF method, requiring large space storage, high computational costs and the risk of overfitting. In this paper, we focus on four distance-based meta-features: CosKNN, L2KNN, CosCent and L2Cent derived from the TF-IDF representations, as a dimensionality reduction method for Arabic text classification. Four well-known classifiers are used in the present work: K-Nearest Neighbors, Logistic Regression, Support Vector Machines and Random Forest to evaluate the impact of these distance-based meta-features on the classification performance. The obtained results prove that the proposed dimensionality reduction method improves the classification accuracy in 50% of the cases and speed up the training phase (between 8x and 1764x faster) when compared to the original TF-IDF. As far as we know, distance-based meta-features are used for Arabic text classification for the first time.
{"title":"Distance-Based Meta-Features for Arabic Text Classification","authors":"Maroua Louail, Chafia Kara-Mohamed alias Hamdi-Cherif","doi":"10.1109/IMSA58542.2023.10217428","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217428","url":null,"abstract":"Text classification (TC) is the process by which the computer has the ability to provide a label to a given text based on its content. Term Frequency–Inverse Document Frequency (TF-IDF) is one of the popular methods used for Arabic text representation. The high number of dimensions and sparseness are among the main issues faced by the TF-IDF method, requiring large space storage, high computational costs and the risk of overfitting. In this paper, we focus on four distance-based meta-features: CosKNN, L2KNN, CosCent and L2Cent derived from the TF-IDF representations, as a dimensionality reduction method for Arabic text classification. Four well-known classifiers are used in the present work: K-Nearest Neighbors, Logistic Regression, Support Vector Machines and Random Forest to evaluate the impact of these distance-based meta-features on the classification performance. The obtained results prove that the proposed dimensionality reduction method improves the classification accuracy in 50% of the cases and speed up the training phase (between 8x and 1764x faster) when compared to the original TF-IDF. As far as we know, distance-based meta-features are used for Arabic text classification for the first time.","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117073884","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217619
Hager Saleh, Nora El-Rashidy, Eman Mohamed, Ahmad M Sultan, Ayman El Nakeeb, Shaker El-Sappagh
Pancreaticoduodenectomy (PD) is a complex surgery used mainly to treat tumors and other pancreas disorders. PD is considered one of the most challenging surgeries because it may have several complications, including bleeding and infections in the surgical area, temporary or permanent diabetes, and pancreatic leakage (PL), which may lead to morbidity and mortality. In this study, we build an accurate and medically oriented machine learning model that predicts PL after PD based on patient markers collected only before the PD operation. The study is made using a real-world dataset for 397 Egyptian patients. The proposed machine learning pipeline starts with a data preprocessing step that handles the missing data values by the median values. In the next step, diverse interpretable classifiers, including logistic regression, random forest, decision tree, support vector machine, XGBoost, and AdaBoost, are utilized to predict the PL. Hyperparameter optimization is done using grid search with k-fold cross-validation. The results indicate that XGBoost achieves the highest marks, outperforming the state-of-the-art techniques in several evaluation metrics (i.e., accuracy= 91%, precision= 90.96%, recall= 91.0%, F1-score= 90.97%, and AUC=89.78%. The resulting model is accurate enough to be medically relevant for PL prediction in real healthcare settings.
{"title":"Machine learning model for predicting pancreatic fistula after pancreatoduodenectomy","authors":"Hager Saleh, Nora El-Rashidy, Eman Mohamed, Ahmad M Sultan, Ayman El Nakeeb, Shaker El-Sappagh","doi":"10.1109/IMSA58542.2023.10217619","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217619","url":null,"abstract":"Pancreaticoduodenectomy (PD) is a complex surgery used mainly to treat tumors and other pancreas disorders. PD is considered one of the most challenging surgeries because it may have several complications, including bleeding and infections in the surgical area, temporary or permanent diabetes, and pancreatic leakage (PL), which may lead to morbidity and mortality. In this study, we build an accurate and medically oriented machine learning model that predicts PL after PD based on patient markers collected only before the PD operation. The study is made using a real-world dataset for 397 Egyptian patients. The proposed machine learning pipeline starts with a data preprocessing step that handles the missing data values by the median values. In the next step, diverse interpretable classifiers, including logistic regression, random forest, decision tree, support vector machine, XGBoost, and AdaBoost, are utilized to predict the PL. Hyperparameter optimization is done using grid search with k-fold cross-validation. The results indicate that XGBoost achieves the highest marks, outperforming the state-of-the-art techniques in several evaluation metrics (i.e., accuracy= 91%, precision= 90.96%, recall= 91.0%, F1-score= 90.97%, and AUC=89.78%. The resulting model is accurate enough to be medically relevant for PL prediction in real healthcare settings.","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129414101","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217767
Maha AbdElwareth, Mariem Abdou, Michael Adel, Alaa Hatem, Login Darwish, Remon Mamdouh, Sahar Selim
A brain tumor is an extremely hazardous illness that can affect people of any age. Less than 50% of individuals with brain cancer have a chance of surviving. As a result, precise segmentation of brain tumors is crucial for the diagnosis, planning of the course of treatment, and tracking of the tumor growth. Deep Learning (DL) models can increase the precision and speed of brain tumor diagnosis by precisely segmenting and identifying tumor locations in medical pictures. In this study, we compare four DL models for segmenting brain tumors, the 3D U-Net, the Attention Res U-Net, the U-Net++, and the U-Net Transformer (UNETR). We used 485 MRI (Magnetic Resonance Imaging) scans from the BraTS 2018 dataset, which include annotated ground truth tumor segmentations. We carried out preprocessing operations such as label merging, cropping, and z-score normalization. We evaluated the performance of two models using the dice coefficient metric. Our findings demonstrated that the Attention Res U-Net has a higher segmentation accuracy than the other three U-Net models, with a testing dice coefficient of 0.79 against 0.78, 0.77, 0.72 for the 3D U-net, UNETR, and U-net++ respectively. The results point to the Attention Res U-Net as a potentially useful method for brain tumor segmentation tasks.
脑肿瘤是一种极其危险的疾病，可以影响任何年龄的人。只有不到50%的脑癌患者有机会存活。因此，脑肿瘤的精确分割对于诊断、规划治疗过程和跟踪肿瘤生长至关重要。深度学习(DL)模型可以通过对医学图像中肿瘤位置的精确分割和识别，提高脑肿瘤诊断的精度和速度。在这项研究中，我们比较了四种用于脑肿瘤分割的深度学习模型:3D U-Net、注意力Res U-Net、U-Net++和U-Net Transformer (UNETR)。我们使用了来自BraTS 2018数据集的485个MRI(磁共振成像)扫描，其中包括带注释的ground truth肿瘤分割。我们进行了预处理操作，如标签合并、裁剪和z-score归一化。我们使用骰子系数度量来评估两个模型的性能。我们的研究结果表明，Attention Res U-Net的分割精度高于其他三种U-Net模型，其测试骰子系数为0.79，而3D U-Net、UNETR和U-Net ++的测试骰子系数分别为0.78、0.77和0.72。研究结果表明，Attention Res U-Net是一种潜在的有用的脑肿瘤分割方法。
{"title":"A Comparative Analysis of Deep Learning Models for Brain Tumor Segmentation","authors":"Maha AbdElwareth, Mariem Abdou, Michael Adel, Alaa Hatem, Login Darwish, Remon Mamdouh, Sahar Selim","doi":"10.1109/IMSA58542.2023.10217767","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217767","url":null,"abstract":"A brain tumor is an extremely hazardous illness that can affect people of any age. Less than 50% of individuals with brain cancer have a chance of surviving. As a result, precise segmentation of brain tumors is crucial for the diagnosis, planning of the course of treatment, and tracking of the tumor growth. Deep Learning (DL) models can increase the precision and speed of brain tumor diagnosis by precisely segmenting and identifying tumor locations in medical pictures. In this study, we compare four DL models for segmenting brain tumors, the 3D U-Net, the Attention Res U-Net, the U-Net++, and the U-Net Transformer (UNETR). We used 485 MRI (Magnetic Resonance Imaging) scans from the BraTS 2018 dataset, which include annotated ground truth tumor segmentations. We carried out preprocessing operations such as label merging, cropping, and z-score normalization. We evaluated the performance of two models using the dice coefficient metric. Our findings demonstrated that the Attention Res U-Net has a higher segmentation accuracy than the other three U-Net models, with a testing dice coefficient of 0.79 against 0.78, 0.77, 0.72 for the 3D U-net, UNETR, and U-net++ respectively. The results point to the Attention Res U-Net as a potentially useful method for brain tumor segmentation tasks.","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127640555","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217658
Hamada Nayel, Nourhan Marzouk, Ahmed N. Elsawy
Named Entity Recognition (NER) plays a vital role in extracting meaningful information from textual data in the medical domain. This paper focuses on NER for Arabic medical texts, specifically targeting the recognition of disease entities. The study presents a comparative analysis of deep learning techniques, including Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), LSTM-CRF, and Bidirectional LSTM (BiLSTM), applied to a dataset comprising Arabic medical texts related to diseases. The dataset is meticulously annotated, ensuring accurate labelling of disease entities for training and evaluation purposes. The models are trained and evaluated using appropriate loss functions and evaluation metrics, such as precision, recall, and F1-score. Comparative experiments are conducted to assess the performance of each model on the disease dataset. The results demonstrate the effectiveness of deep learning techniques for NER in Arabic medical texts, with the LSTM-CRF and BiLSTM-CRF models outperforming the standalone CRF and LSTM models. LSTM-CRF and BiLSTM-CRF models reported F1-score of 0.97 and 0.94. These hybrid models achieve higher precision, recall, and F1-score, showcasing their ability to accurately identify disease entities in Arabic medical texts. The findings of this study contribute to the advancement of NER techniques for Arabic medical texts, focusing on disease entities. The comparative analysis of CRF, LSTM, LSTM-CRF, and BiLSTM models provides valuable insights into their respective strengths and limitations of NER for Arabic medical texts. These insights can guide the selection and implementation of appropriate models for disease entity recognition in Arabic medical texts, facilitating accurate information extraction and analysis in the medical domain.
{"title":"Named Entity Recognition for Arabic Medical Texts Using Deep Learning Models","authors":"Hamada Nayel, Nourhan Marzouk, Ahmed N. Elsawy","doi":"10.1109/IMSA58542.2023.10217658","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217658","url":null,"abstract":"Named Entity Recognition (NER) plays a vital role in extracting meaningful information from textual data in the medical domain. This paper focuses on NER for Arabic medical texts, specifically targeting the recognition of disease entities. The study presents a comparative analysis of deep learning techniques, including Conditional Random Fields (CRF), Long Short-Term Memory (LSTM), LSTM-CRF, and Bidirectional LSTM (BiLSTM), applied to a dataset comprising Arabic medical texts related to diseases. The dataset is meticulously annotated, ensuring accurate labelling of disease entities for training and evaluation purposes. The models are trained and evaluated using appropriate loss functions and evaluation metrics, such as precision, recall, and F1-score. Comparative experiments are conducted to assess the performance of each model on the disease dataset. The results demonstrate the effectiveness of deep learning techniques for NER in Arabic medical texts, with the LSTM-CRF and BiLSTM-CRF models outperforming the standalone CRF and LSTM models. LSTM-CRF and BiLSTM-CRF models reported F1-score of 0.97 and 0.94. These hybrid models achieve higher precision, recall, and F1-score, showcasing their ability to accurately identify disease entities in Arabic medical texts. The findings of this study contribute to the advancement of NER techniques for Arabic medical texts, focusing on disease entities. The comparative analysis of CRF, LSTM, LSTM-CRF, and BiLSTM models provides valuable insights into their respective strengths and limitations of NER for Arabic medical texts. These insights can guide the selection and implementation of appropriate models for disease entity recognition in Arabic medical texts, facilitating accurate information extraction and analysis in the medical domain.","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127691492","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-07-15DOI: 10.1109/IMSA58542.2023.10217752
A. Sulavko, I. Panfilova, A. Samotuga, Samal Zhumazanova
A method of biometric authentication based on the thermogram of the subject's face was proposed. This method allows you to associate a biometric image of a person with a cryptographic key or password, as well as protect the biometric image and key (password) from being compromised during storage and transmission over communication channels. This effect was achieved through the use of a fuzzy neural extractor trained according to the GOST R 52633.5 standard. The solution also uses a deep convolutional neural network for face detection and an Inception-Resnet network for feature embedding. RetinaFace, ResNet50 and VGG-Face were tested as alternatives to these neural network models. The best result achieved was EER = 4.91
提出了一种基于人脸热像图的生物特征认证方法。此方法允许您将人的生物特征图像与加密密钥或密码关联，并保护生物特征图像和密钥(密码)在存储和通过通信通道传输期间不被泄露。这种效果是通过使用根据GOST R 52633.5标准训练的模糊神经提取器来实现的。该解决方案还使用深度卷积神经网络进行人脸检测，并使用Inception-Resnet网络进行特征嵌入。我们测试了RetinaFace、ResNet50和VGG-Face作为这些神经网络模型的替代品。最佳结果为EER = 4.91
{"title":"Biometric Authentication Using Face Thermal Images Based on Neural Fuzzy Extractor","authors":"A. Sulavko, I. Panfilova, A. Samotuga, Samal Zhumazanova","doi":"10.1109/IMSA58542.2023.10217752","DOIUrl":"https://doi.org/10.1109/IMSA58542.2023.10217752","url":null,"abstract":"A method of biometric authentication based on the thermogram of the subject's face was proposed. This method allows you to associate a biometric image of a person with a cryptographic key or password, as well as protect the biometric image and key (password) from being compromised during storage and transmission over communication channels. This effect was achieved through the use of a fuzzy neural extractor trained according to the GOST R 52633.5 standard. The solution also uses a deep convolutional neural network for face detection and an Inception-Resnet network for feature embedding. RetinaFace, ResNet50 and VGG-Face were tested as alternatives to these neural network models. The best result achieved was EER = 4.91","PeriodicalId":110239,"journal":{"name":"2023 Intelligent Methods, Systems, and Applications (IMSA)","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132144700","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：