Search

In e-commerce, online reviews play a vital role in understanding customer requirements. Companies can perform analysis of these reviews to make right business decisions. However, manually analyzing the content generated by each user on social media is challenging due to its high volume and frequency. Sentiment analysis (SA) is a Natural Language Processing (NLP) technique which provides an automated solution to this problem. It is a challenging subtask of text classification. Over a long time, researchers have proposed different approaches to solve it. This paper proposes an unsupervised learning approach using transformer architecture to perform SA on women's clothing e-commerce dataset which is imbalanced in nature. In this study, we have fine-tuned DistilBERT, a cutting-edge transformer based pre-trained model and developed models for two subtasks of SA named as Sentiment Classification (SC) and Product Recommendation (PR). The proposed models have achieved the highest F1 scores (0.79 for SC and 0.85 for PR), AUC scores (0.98 for SC and 0.96 for PR) along with the highest accuracy of 0.96 for SC and 0.91 for PR. We found that the performance of our models is least affected by imbalanced dataset issues. The results show that the proposed models have significantly outperformed the traditional supervised approaches and various other existing state-of-the-art (SOTA) models. This study can contribute to a better understanding of consumer sentiment and consumer psychology in the e-commerce transaction business. [ABSTRACT FROM AUTHOR]

Ophthalmic disorders represent a major global health problem leading to visual impairment and even blindness if not detected and treated early. Deep learning is a popular approach where pre-trained models are now often used to diagnose a variety of diseases from different medical images. In order to provide an accurate and reliable categorization of ophthalmic illnesses, this study suggests a fusion or ensemble technique that harnesses the power of pre-trained models like ResNet50, DenseNet and EfficientNet. The proposed ensemble model enhances the performance by incorporating the benefits of three architectural paradigms, guiding the convergence of each one's distinct strengths to achieve improved overall classification result. The ensemble architecture has undertaken inclusive validation on multiple fundus image datasets, showing improved performance over lone models. An accuracy of 92% demonstrates the potential of ensemble model in the categorization of eye diseases like cataract, diabetic retinopathy, glaucoma and normal eyes. In addition to achieving cutting-edge accuracy, the ensemble also offers an AUC-ROC score of 1.00. This technique presents a viable alternative for boosting the accuracy and reliability in identifying various eye illnesses by combining diverse model capabilities. The empirical data and conclusions show that the ensemble model has the potential to make a substantial contribution to the field of ocular medical imaging and patient care by enabling precise and timely treatments that may significantly improve the lives of millions of people worldwide who are impacted by ophthalmic diseases. [ABSTRACT FROM AUTHOR]

User identification across multiple online social networks is beneficial for building knowledge graphs. Under privacy protection considerations, researchers have shown increasing interest in user identification based on username similarity. However, existing solutions rely on manual features extracted by domain experts and do not exploit the deep semantic features of usernames. Moreover, existing solutions are limited to monolingual user names such as English or Chinese, ignoring other multilingual usernames. This paper proposes a multilingual pre-trained model-based username similarity method for user identification across multiple online social networks. First, we use many multilingual corpora to enable the model to learn more semantic information and extract deep semantic features of usernames. Then, fine-tuning is performed on our constructed dataset of multilingual usernames across multiple online social networks. Ultimately assess the similarity of user identities across multiple online social networks. Our method facilitates user identification with limited data. Finally, the efficiency of our model is verified on three constructed real-world multilingual username datasets across multiple online social networks and compared with existing state-of-the-art methods. Experimental results show that the proposed algorithm outperforms the compared algorithms. [ABSTRACT FROM AUTHOR]

Pneumonia is a respiratory illness that is highly infectious and affecting one or both lungs. It can lead to fatalities if left undiagnosed and untreated in time. While a radiologist can diagnose pneumonia just by looking at a chest X-ray, there are certain factors that may increase the chances of a misdiagnosis such as fatigue, multi-presenting symptoms, or inadequate overall experience in assessment of patients with low-prevalence presentations. A publicly available chest X-ray dataset was used in this study to create a portable executable clinical decision support tool to help determine the existence of pneumonia on a given chest X-ray image. ResNet50, DenseNet201, Xception, MobileNetV2, and ResNet101 were retrained to confirm pneumonia in chest X-rays both on the original and augmented dataset. An ensemble network was also constructed to improve the results by combining the strengths of the included pre-trained models using average probability. In this study, a total of 37 experiments were performed. Since there are instances when no single model excels on all metrics, the model with the highest Matthews Correlation Coefficient (MCC) was selected as the best model. MCC outputs an optimal score only if the classifier was able to get a high percentage for both positive and negative samples. To increase further the performance of the ensemble model, a novel method was introduced that determines which pre-trained model will give the greatest increase in the ensemble model performance when removed from the ensemble. The result is the identification of the best ensemble model variant consisting of only ResNet50, DenseNet201, Xception, and ResNet101 with Accuracy=93%, Precision=91%, Recall=98%, F1-Score=94%, Specificity=83%, MCC=85%. A stand-alone no-install application was created for this purpose to enable those with limited access or no access to internet to run the tool on their Window-based computer. Our tool can be used as portable stand-alone clinical decision support tool by radiologists when evaluating possible pneumonia case. It can also be used as a teaching tool for general practitioners and medical students. [ABSTRACT FROM AUTHOR]

Brain imaging plays a crucial role in the diagnosis and treatment of various neurological disorders, providing valuable insights into the structure and function of the brain. Techniques such as magnetic resonance imaging (MRI) and computed tomography (CT) enable non-invasive visualization of the brain, aiding in the understanding of brain anatomy, abnormalities, and functional connectivity. However, cost and radiation dose may limit the acquisition of specific image modalities, so medical image synthesis can be used to generate required medical images without actual addition. CycleGAN and other GANs are valuable tools for generating synthetic images across various fields. In the medical domain, where obtaining labeled medical images is labor-intensive and expensive, addressing data scarcity is a major challenge. Recent studies propose using transfer learning to overcome this issue. This involves adapting pre-trained CycleGAN models, initially trained on non-medical data, to generate realistic medical images. In this work, transfer learning was applied to the task of MR-CT image translation and vice versa using 18 pre-trained non-medical models, and the models were fine-tuned to have the best result. The models' performance was evaluated using four widely used image quality metrics: Peak-signal-to-noise-ratio, Structural Similarity Index, Universal Quality Index, and Visual Information Fidelity. Quantitative evaluation and qualitative perceptual analysis by radiologists demonstrate the potential of transfer learning in medical imaging and the effectiveness of the generic pre-trained model. The results provide compelling evidence of the model's exceptional performance, which can be attributed to the high quality and similarity of the training images to actual human brain images. These results underscore the significance of carefully selecting appropriate and representative training images to optimize performance in brain image analysis tasks. • Pre-trained CycleGANs can effectively deal with limited labeled medical images. • Models performance were evaluated comprehensively with Peak SNR, Structural Similarity, UQI, and Visual Fidelity. • Models latent space was illustratively analyzed to explain class separability for paired image translation. • Improved results achieved when carefully selecting non-medical representative training datasets related to underlying physiology. [ABSTRACT FROM AUTHOR]

Nowadays, many attendance applications utilise biometric techniques such as the face, fingerprint, and iris recognition. Biometrics has become ubiquitous in many sectors. Due to the advancement of deep learning algorithms, the accuracy rate of biometric techniques has been improved tremendously. This paper proposes a web-based attendance system that adopts facial recognition using open-source deep learning pre-trained models. Face recognition procedural steps using web technology and database were explained. The methodology used the required pre-trained weight files embedded in the procedure of face recognition. The face recognition method includes two important processes: registration of face datasets and face matching. The extracted feature vectors were implemented and stored in an online database to create a more dynamic face recognition process. Finally, user testing was conducted, whereby users were asked to perform a series of biometric verification. The testing consists of facial scans from the front, right (30 – 45 degrees) and left (30 – 45 degrees). Reported face recognition results showed an accuracy of 92% with a precision of 100% and recall of 90%. [ABSTRACT FROM AUTHOR]

Sequential data is widely used in several fields, such as power payload prediction, traffic flow prediction, and stock trend prediction. Driven by the urgent needs, sequential forecasting based on deep learning methods has received lots of attention in recent years. However, the potential of deep learning methods in sequential data forecasting has not yet fully been exploited in terms of model architecture. In this study, a pre-trained nodal model with multiple fusion layers architecture (Sequence Prediction via Node Fusion, SPNF) was proposed, the model considered both connections of nodes and the temporal components of nodes to predict value of next node. Multiple fusion layers were adopted to capture spatial features and temporal dependencies from historical data. The proposed model also conpensated missing data via a masking mechanism. To validate the proposed model, experiments were also carried out using field-captured traffic data, the performance of proposed model were compared with classical and state-of-theart models, such as, ARIMA, SVR, LSTM, TGC-LSTM neural networks. The results showed that proposed model yields higher accuracy and robustness than others, especially in the case of large sequence changes occurred. [ABSTRACT FROM AUTHOR]

Sequential data is widely used in several fields, such as power payload prediction, traffic flow prediction, and stock trend prediction. Driven by the urgent needs, sequential forecasting based on deep learning methods has received lots of attention in recent years. However, the potential of deep learning methods in sequential data forecasting has not yet fully been exploited in terms of model architecture. In this study, a pre-trained nodal model with multiple fusion layers architecture (Sequence Prediction via Node Fusion, SPNF) was proposed, the model considered both connections of nodes and the temporal components of nodes to predict value of next node. Multiple fusion layers were adopted to capture spatial features and temporal dependencies from historical data. The proposed model also conpensated missing data via a masking mechanism. To validate the proposed model, experiments were also carried out using field-captured traffic data, the performance of proposed model were compared with classical and state-of-theart models, such as, ARIMA, SVR, LSTM, TGC-LSTM neural networks. The results showed that proposed model yields higher accuracy and robustness than others, especially in the case of large sequence changes occurred. [ABSTRACT FROM AUTHOR]

Smart contracts are programmable protocols that run on Ethereum and require gas to be deployed and used. Gas-expensive operations in some smart contracts can cause users to consume extra gas in transactions. There are already several methods to detect gas-expensive patterns in smart contracts. However, some problems still need to be solved: most static analysis methods are based on specialist knowledge, and the patterns used to detect gas-expensive patterns must be manually summarized before the detection methods can be applied. Furthermore, due to the explosive growth of smart contracts, which generate a large amount of data, it is challenging to reuse these methods across different patterns. To address these issues, this work first proposes a new learning-based method, ExpenGas, based on the idea of evolutionary computation-based machine learning to detect Expensive Operation patterns of smart contracts through pre-trained techniques and multi-crucial data flow graphs. The low complexity of the multi-crucial data flow graph enables the model to focus on key features. Finally, by testing on 21981 smart contract files, ExpenGas has 83.05% accuracy and 91.96% recall in detecting the Expensive Operation patterns of gas-expensive patterns, which is significantly more optimal than the current state-of-the-art methods. • Detection of waste gas problems in smart contract loops. • Introducing pre-trained techniques into detection. • Using multi-crucial data flow graphs to reduce the interference of useless information. • The experimental results are more optimal than existing detection methods. • This method can be used in other waste gas problem detection. [ABSTRACT FROM AUTHOR]

Colorectal cancer has a high mortality rate that continuously affects human life globally. Early detection of it extends human life and helps in preventing disease. Histopathological inspection is a frequently used approach to diagnose and detect colorectal cancer. Visual inspection of histopathological diagnosis requires more inspection time and the decision depends on the subjective perception of clinicians. This work proposed lightweight, less complex convolutional neural network-based architecture for automated classification of multi-class colorectal tissue histopathological images using two publicly available datasets, colorectal histology, and NCT-CRC-HE-100K, respectively. Histopathological images are provided as input to pre-trained models Xception, InceptionResNetV2, DenseNet121, VGG16, and the proposed network colorectal cancer classification convolutional neural network. This is the first study that compares the computational time of different deep learning architectures for the classification of colorectal tissue. The developed network requires less computational time for training compared to other pre-trained models. Accuracy, sensitivity, precision, false-positive rate, false-negative rate, specificity, F-1 score, and area under the curve have been used to evaluate the performance of the proposed architecture. The proposed network attained an accuracy of 93.50%, and 96.26% on the colorectal histology dataset, and NCT-CRC-HE-100K dataset, respectively. On the merged dataset, an accuracy of 99.21% is achieved by the newly developed network. The comparative analysis shows that the proposed framework outperformed existing state-of-the-art approaches. Clinicians may install the presented CRCCN-Net to confirm the diagnosis in the hospitals. [Display omitted] • Colorectal histology, NCT-CRC-HE-100K dataset, and merged datasets are used. • A new CRCCN-Net architecture is used to classify colorectal tissue in multi classes. • Developed CRCCN-Net obtained the highest classification accuracy of 99.21%. • The proposed CRCCN-Net has lesser computational time complexity. [ABSTRACT FROM AUTHOR]

Colorectal cancer has a high mortality rate that continuously affects human life globally. Early detection of it extends human life and helps in preventing disease. Histopathological inspection is a frequently used approach to diagnose and detect colorectal cancer. Visual inspection of histopathological diagnosis requires more inspection time and the decision depends on the subjective perception of clinicians. This work proposed lightweight, less complex convolutional neural network-based architecture for automated classification of multi-class colorectal tissue histopathological images using two publicly available datasets, colorectal histology, and NCT-CRC-HE-100K, respectively. Histopathological images are provided as input to pre-trained models Xception, InceptionResNetV2, DenseNet121, VGG16, and the proposed network colorectal cancer classification convolutional neural network. This is the first study that compares the computational time of different deep learning architectures for the classification of colorectal tissue. The developed network requires less computational time for training compared to other pre-trained models. Accuracy, sensitivity, precision, false-positive rate, false-negative rate, specificity, F-1 score, and area under the curve have been used to evaluate the performance of the proposed architecture. The proposed network attained an accuracy of 93.50%, and 96.26% on the colorectal histology dataset, and NCT-CRC-HE-100K dataset, respectively. On the merged dataset, an accuracy of 99.21% is achieved by the newly developed network. The comparative analysis shows that the proposed framework outperformed existing state-of-the-art approaches. Clinicians may install the presented CRCCN-Net to confirm the diagnosis in the hospitals. [Display omitted] • Colorectal histology, NCT-CRC-HE-100K dataset, and merged datasets are used. • A new CRCCN-Net architecture is used to classify colorectal tissue in multi classes. • Developed CRCCN-Net obtained the highest classification accuracy of 99.21%. • The proposed CRCCN-Net has lesser computational time complexity. [ABSTRACT FROM AUTHOR]