Sentiment analysis is the process of extracting the opinions of customers from online reviews. In general, customers express their reviews in natural language. It becomes a complex task when applying ...sentiment analysis on those reviews. In earlier stages, word‐level features with various feature weighting methods such as Bag of Words, TF‐IDF, and Word2Vec were applied for sentiment analysis and deep learning networks are not explored much. We considered phrase level and sentence level features instead of applying word‐level features for sentiment analysis and also enhanced by applying various deep learning techniques. In this article, we have proposed a hybrid convolutional bidirectional recurrent neural network model (CBRNN) by combining two‐layer convolutional neural network (CNN) with a bidirectional gated recurrent unit (BGRU). In the proposed CBRNN model, the CNN layer extracts the rich set of phrase‐level features and BGRU captures the chronological features through long term dependency in a multi‐layered sentence. The proposed approach was evaluated on two benchmark datasets and compared with various baselines. The experimental results show that the proposed hybrid model provides better results than any other models with an F1 score of 87.62% and 77.4% on IMDB and Polarity datasets,respectively. Our CBRNN model outperforms the state of the art by 2%‐4% on these two datasets. It is also observed that, the time taken for training is slightly higher than the existing approaches with the substantial improvement in the performance.
The posting of offensive content in regional languages has increased as a result of the accessibility of low-cost internet and the widespread use of online social media. Despite the large number of ...comments available online, only a small percentage of them are offensive, resulting in an unequal distribution of offensive and non-offensive comments. Due to this class imbalance, classifiers may be biased toward the class with the most samples, i.e., the non-offensive class. To address class imbalance, a Multilingual Translation-based Data augmentation technique for Offensive content identification in Tamil text data (MTDOT) is proposed in this work. The proposed MTDOT method is applied to HASOC’21, which is the Tamil offensive content dataset. To obtain a balanced dataset, each offensive comment is augmented using multi-level back translation with English and Malayalam as intermediate languages. Another balanced dataset is generated by employing single-level back translation with Malayalam, Kannada, and Telugu as intermediate languages. While both approaches are equally effective, the proposed multi-level back-translation data augmentation approach produces more diverse data, which is evident from the BLEU score. The MTDOT technique proposed in this work achieved a promising improvement in F1-score over the widely used SMOTE class balancing method by 65%.
The class imbalance problem, in which the distribution of different classes in training data is unequal or skewed, is a prevailing problem. This can lead to classifier algorithms being biased, ...negatively impacting the performance of the minority class. In this paper, we addressed the class imbalance problem in datasets for aspect-based sentiment classification. Aspect-based Sentiment Classification (AbSC) is a type of fine-grained sentiment analysis in which sentiments about particular aspects of an entity are extracted. In this work, we addressed the issue of class imbalance by creating synthetic data. For synthetic data generation, two techniques have been proposed: paraphrasing using the PEGASUS fine-tuned model and backtranslation using the M2M100 neural machine translation model. We compared these techniques with two other class balancing techniques, such as weighted oversampling and cross-entropy loss with class weight. An extensive experimental study has been conducted on three benchmark datasets for restaurant reviews: SemEval-2014, SemEval-2015, and SemEval-2016. We applied these methods to the BERT-based deep learning model for aspect-based sentiment classification and studied the effect of balancing the data on the performance of these models. Our proposed balancing technique, using synthetic data, yielded better results than the other two existing methods for dealing with multi-class imbalance.
Colorectal Cancer is one of the most common cancers found in human beings, and polyps are the predecessor of this cancer. Accurate Computer-Aided polyp detection and segmentation system can help ...endoscopists to detect abnormal tissues and polyps during colonoscopy examination, thereby reducing the chance of polyps growing into cancer. Many of the existing techniques fail to delineate the polyps accurately and produce a noisy/broken output map if the shape and size of the polyp are irregular or small. We propose an end-to-end pixel-wise polyp segmentation model named Guided Attention Residual Network (GAR-Net) by combining the power of both residual blocks and attention mechanisms to obtain a refined continuous segmentation map. An enhanced Residual Block is proposed that suppresses the noise and captures low-level feature maps, thereby facilitating information flow for a more accurate semantic segmentation. We propose a special learning technique with a novel attention mechanism called Guided Attention Learning that can capture the refined attention maps both in earlier and deeper layers regardless of the size and shape of the polyp. To study the effectiveness of the proposed GAR-Net, various experiments were carried out on two benchmark collections viz., CVC-ClinicDB (CVC-612) and Kvasir-SEG dataset. From the experimental evaluations, it is shown that GAR-Net outperforms other previously proposed models such as FCN8, SegNet, U-Net, U-Net with Gated Attention, ResUNet, and DeepLabv3. Our proposed model achieves 91% Dice co-efficient and 83.12% mean Intersection over Union (mIoU) on the benchmark CVC-ClinicDB (CVC-612) dataset and 89.15% dice co-efficient and 81.58% mean Intersection over Union (mIoU) on the Kvasir-SEG dataset. The proposed GAR-Net model provides a robust solution for polyp segmentation from colonoscopy video frames.
Rice is one of the most important food crops in the South India. Many varieties of rice are cultivated in different regions of the India to meet the dietary needs of the ever-growing population. In ...spite of huge investment in terms of land, labour, raw materials and machinery, the farmers continuously face irrecoverable loss due to various reasons like climatic changes, drought situation and seed quality. In the current practice, the quality of the seeds is certified by the Seed Testing Laboratories (STL) and purity analysis is done manually by trained technicians. However, seed classification is not uniform across different labs, due to several factors like fatigue, eye-strain and personal circumstances of the technicians. Hence, automated rice seed variety identification becomes a crucial task for ensuring the quality and germination potential of rice crops. This research is focused on the application of Deep Neural Network (RiceSeedNet) combined with traditional image processing techniques to classify local rice seed varieties of southern Tamilnadu, India. The RiceSeed Image corpus is created for this purpose considering 13 local varieties. The captured RGB images of rice seed data consists of 13,000 images of local rice seed varieties, having 1000 images for each variety. To automate the rice seed varietal identification, vision transformer-based architecture RiceSeedNet is developed. The proposed RiceSeedNet is 97% accurate in classifying the 13 local varieties of rice seeds. The RiceSeedNet was also evaluated on a publicly available rice grain data set to study the performance of the proposed model across the different rice grain varieties. On this cross-data validation, RiceSeedNet is able to achieve 99% accuracy in classifying 8 varieties of rice grains on the public dataset.
Display omitted
•Created Rice Seed image corpus comprising 13 local rice seed varieties of Tamil Nadu, India.•Custom -designed RiceSeedNet architecture using Vision Transformer based model and obtained an accuracy of 97% for rice seed varietal identification.•Cross-data validation of RiceSeedNet was performed on the publicly available rice grain dataset and obtained 99% accuracy.
3D printing is a growing technology being incorporated into almost every industry. Although it has obvious advantages, such as precision and less fabrication time, it has many shortcomings. Although ...several attempts were made to monitor the errors, many have not been able to thoroughly address them, like stringing, over-extrusion, layer shifting, and overheating. This paper proposes a study using machine learning to identify the optimal process parameters such as infill structure and density, material (ABS, PLA, Nylon, PVA, and PETG), wall and layer thickness, count, and temperature. The result thus obtained was used to train a machine learning algorithm. Four different network architectures (CNN, Resnet152, MobileNet, and Inception V3) were used to build the algorithm. The algorithm was able to predict the parameters for a given requirement. It was also able to detect any errors. The algorithm was trained to pause the print immediately in case of a mistake. Upon comparison, it was found that the algorithm built with Inception V3 achieved the best accuracy of 97%. The applications include saving the material from being wasted due to print time errors in the manufacturing industry.
Sentiment analysis is a means of excerpting subjective information from client reviews. The existing shallow model lacks in addressing multiple relation/meaning of a word in a review. To address the ...above issue and to find an effective contextual word embedding, we have performed a thorough analysis on the existing language model, viz., Universal Language Model Fine-tuning, Embeddings from Language Models and Bidirectional Encoder Representations from Transformers (BERT). Based on the analysis, we have proposed a transfer learning-based bidirectional transformer model. We have conducted several experiments with the different transfer learning-based bidirectional transformer models to find a robust classifier for contextual embedding. In these various transfer learning approaches, the attention-extracted features are fed into different classifiers, viz., support vector machine (SVM), logistic regression (LR), long short-term memory (LSTM) and bidirectional gated recurrent unit (BGRU). BERT is a multi-head, multilayered and bidirectional transformer that finds deep contextual words existing in a review by exhibiting different patterns in different layers. In the proposed architecture, the attention-extracted deep contextual features from BERT are fed into the BGRU through transfer learning to have better contextual classification. With the proposed attention-extracted BERT-bidirectional gated recurrent unit (i.e., AeBERT-BGRU), we have obtained an F
1
score of 95.57, 86.66 and 80.29% for IMDB, Polarity and OLID dataset (weighted), which is a significant improvement when compared to state-of-the-art methods.