Search

Diagnosing and treating abnormalities in the wrist, specifically distal radius, and ulna fractures, is a crucial concern among children, adolescents, and young adults, with a higher incidence rate during puberty. However, the scarcity of radiologists and the lack of specialized training among medical professionals pose a significant risk to patient care. This problem is further exacerbated by the rising number of imaging studies and limited access to specialist reporting in certain regions. This highlights the need for innovative solutions to improve the diagnosis and treatment of wrist abnormalities. Automated wrist fracture detection using object detection has shown potential, but current studies mainly use two-stage detection methods with limited evidence for single-stage effectiveness. This study employs state-of-the-art single-stage deep neural network-based detection models YOLOv5, YOLOv6, YOLOv7, and YOLOv8 to detect wrist abnormalities. Through extensive experimentation, we found that these YOLO models outperform the commonly used two-stage detection algorithm, Faster R-CNN, in fracture detection. Additionally, compound-scaled variants of each YOLO model were compared, with YOLOv8 m demonstrating a highest fracture detection sensitivity of 0.92 and mean average precision (mAP) of 0.95. On the other hand, YOLOv6 m achieved the highest sensitivity across all classes at 0.83. Meanwhile, YOLOv8x recorded the highest mAP of 0.77 for all classes on the GRAZPEDWRI-DX pediatric wrist dataset, highlighting the potential of single-stage models for enhancing pediatric wrist imaging.

Energy prices have gone up gradually since last year, but a drastic hike has been observedrecently in the past couple of months, affecting people’s thrift. This, coupled with the load shedding andenergy shortages in some parts of the world, led many to show anger and bitterness on the streets and on socialmedia. Despite subsidies offered by many Governments to their citizens to compensate for high energy bills,the energy price hike is a trending topic on Twitter. However, not much attention is paid to opinion mining onsocial media posts on this topic. Therefore, in this study, we propose a solution that takes advantage of botha transformer-based sentiment analysis method and topic modeling to explore public engagement on Twitterregarding energy prices rising. The former method is employed to annotate the valence of the collected tweetsas positive, neutral and negative, whereas the latter is used to discover hidden topics/themes related to energyprices for which people have expressed positive or negative sentiments. The proposed solution is tested ona dataset composed of 366,031 tweets collected from 01 January 2021 to 18 June 2022. The findings showthat people have discussed a variety of topics which directly or indirectly affect energy prices. Moreover,the findings reveal that the public sentiment towards these topics has changed over time, in particular, in2022 when negative sentiment was dominant.

In the field of text classification, researchers have repeatedly shown the value of transformer-based models such as Bidirectional Encoder Representation from Transformers (BERT) and its variants. Nonetheless, these models are expensive in terms of memory and computational power but have not been utilized to classify long documents of several domains. In addition, transformer models are also often pre-trained on generalized languages, making them less effective in language-specific domains, such as legal documents. In the natural language processing (NLP) domain, there is a growing interest in creating newer models that can handle more complex input sequences and domain-specific languages. Keeping the power of NLP in mind, this study proposes a legal documentation classifier that classifies the legal document by using the sliding window approach to increase the maximum sequence length of the model. We used the ECHR (European Court of Human Rights) publicly available dataset which to a large extent is imbalanced. Therefore, to balance the dataset we have scrapped the case articles from the web and extracted the data. Then, we employed conventional machine learning techniques such as SVM, DT, NB, AdaBoost, and transformer-based neural networks models including BERT, Legal-BERT, RoBERTa, BigBird, ELECTRA, and XLNet for the classification task. The experimental findings show that RoBERTa outperformed all the mentioned BERT versions by obtaining precision, recall, and F1-score of 89.1%, 86.2%, and 86.7%, respectively. While from conventional machine learning techniques, AdaBoost outclasses SVM, DT, and NB by achieving scores of 81.9%, 81.5%, and 81.7% for precision, recall, and F1-score, respectively.

Low-resource languages are gaining much-needed attention with the advent of deep learning models and pre-trained word embedding. Though spoken by more than 230 million people worldwide, Urdu is one such low-resource language that has recently gained popularity online and is attracting a lot of attention and support from the research community. One challenge faced by such resource-constrained languages is the scarcity of publicly available large-scale datasets for conducting any meaningful study. In this paper, we address this challenge by collecting the first-ever large-scale Urdu Tweet Dataset for sentiment analysis and emotion recognition. The dataset consists of a staggering number of 1,140,821 tweets in the Urdu language. Obviously, manual labeling of such a large number of tweets would have been tedious, error-prone, and humanly impossible; therefore, the paper also proposes a weakly supervised approach to label tweets automatically. Emoticons used within the tweets, in addition to SentiWordNet, are utilized to propose a weakly supervised labeling approach to categorize extracted tweets into positive, negative, and neutral categories. Baseline deep learning models are implemented to compute the accuracy of three labeling approaches, i.e., VADER, TextBlob, and our proposed weakly supervised approach. Unlike the weakly supervised labeling approach, the VADER and TextBlob put most tweets as neutral and show a high correlation between the two. This is largely attributed to the fact that these models do not consider emoticons for assigning polarity.

The Internet has become one of the significant sources for sharing information and expressing users' opinions about products and their interests with the associated aspects. It is essential to learn about product reviews; however, to react to such reviews, extracting aspects of the entity to which these reviews belong is equally important. Aspect-based Sentiment Analysis (ABSA) refers to aspects extracted from an opinionated text. The literature proposes different approaches for ABSA; however, most research is focused on supervised approaches, which require labeled datasets with manual sentiment polarity labeling and aspect tagging. This study proposes a semi-supervised approach with minimal human supervision to extract aspect terms by detecting the aspect categories. Hence, the study deals with two main sub-tasks in ABSA, named Aspect Category Detection (ACD) and Aspect Term Extraction (ATE). In the first sub-task, aspects categories are extracted using topic modeling and filtered by an oracle further, and it is fed to zero-shot learning as the prompts and the augmented text. The predicted categories are the input to find similar phrases curated with extracting meaningful phrases (e.g., Nouns, Proper Nouns, NER (Named Entity Recognition) entities) to detect the aspect terms. The study sets a baseline accuracy for two main sub-tasks in ABSA on the Multi-Aspect Multi-Sentiment (MAMS) dataset along with SemEval-2014 Task 4 sub-task 1 to show that the proposed approach helps detect aspect terms via aspect categories.

Personality is a set of traits and unique characteristics that give both consistency and individuality to a person’s behavior. As personality is accepted as an indicator of job performance, recruiters aim to retrieve these behavior traits in the screening process. One issue is that using personality questionnaires is less favored by applicants and negatively affects the pace of the recruitment process. Therefore, it is desired to infer one’s personality traits during an interview. Nowadays, companies tend to use Asynchronous Video Interviews (AVIs), an interview setting where applicants record their answers to pre-defined questions and an interviewer is not present. These recordings are valuable to recruiters since emotions expressed in the AVIs reveal personality insights. With the increasing interest in understanding human behavior, Multimodal Sentiment Analysis (MSA) has emerged as a popular research field. MSA aims to recognize the expressed emotions and sentiment in multimodal data. In this paper, we investigate how MSA can assist personality prediction in AVIs. We propose a novel emotion-to-personality prediction method where dimensions of the Big Five model of personality are predicted based on the emotion distribution in an interview. We collect an AVI dataset in order to test our method. The use of self-assessment i.e. personality questionnaire shows a unique and novel way of validating the personality prediction model on interviews from the AVI dataset. Our findings show that the emotion-to-personality prediction method achieves promising results when predicting the three strongest personality traits, based on the Big Five personality traits, exhibited by an individual in the AVI.

Labeling of trademark images with Vienna codes from the Vienna classification is a manual process carried out by domain experts by searching trademark image databases using specific keywords. Manual labeling is both a time-consuming and error-prone process. Therefore, in this paper, we investigate how deep learning techniques can improve and automate labeling of new unlabeled trademark images. Three different deep learning models, namely CNN, LSTM and GRU, are trained and tested on a collected dataset composed of 14,500 unique logos extracted from the European Union Intellectual Property Office Open Data Portal. A set of controlled experiments establishing baseline results on the dataset showed that CNN outperforms the other two models in terms of both accuracy and training time. The experimental results also suggest that deep learning models are an important tool that can be applied by Intellectual Property Offices in real-world applications to automate the trademark image classification task.

Recent advancements in big data, algorithms, and computing power have triggered significant enhancements in artificial intelligence (AI). Almost every aspect of travel and tourism is currently impacted by AI, which can be evidenced in a variety of applications including robots, conversational systems, smart travel agents, prediction and forecasting systems, voice recognition, and natural language processing. In this article, we examine how AI has altered and continues to alter the key operations and processes in the tourism industry, with special emphasis on sustainability. After applying the PRISMA framework to guide our search process, the study identified 69 relevant articles published between January 1, 2018, and November 1, 2022. The mapping results revealed that the field is expanding quickly, despite the noted obstacles and challenges. We identified several factors and challenges that should be considered in order to advance the level of research and development in this area. Among these factors, we emphasize the importance and the need for standardized and multimodal datasets, transformer-based and more advanced representation techniques, and standardized performance evaluation metrics for AI models. Also, based on these challenges, some recommendations are provided, and future research directions are identified.

Understanding complex machine learning and artificial intelligence models have always been challenging because these models are black-box, and often we don't know what information models rely upon to infer. Explainable Artificial Intelligence (XAI) has emerged as a new exciting field to explain and understand these machine learning models as humans can understand and improve them. In the past few years, there have been numerous research articles on explainable artificial intelligence for medical and healthcare. 1687 documents are being studied and analysed using bibliometric methods in this work. There are certain systematic reviews on the same topic, but this study is the first of its kind to use a quantitative method to analyze a large number of publications. The results of this study show that the research in this field took place in 2011, and there have been quite many publications in the following years. We have also identified top-cited journals and articles. Through thematic analysis, we have found some important thematic areas of research in the field of XAI for medical and healthcare. The findings showed that the USA is the global leader in XAI research, followed by China and Canada at second and third place, respectively.