Deep learning has advanced medical imaging, but limited interpretability hinders clinical adoption. Class activation maps (CAM) provide visual explanations, yet methods such as Score-CAM are computationally expensive, requiring a forward pass for each activation map and limiting real-time applicability despite their high fidelity. To overcome this limitation, LowRank-CAM is proposed, which aggregates activation maps into a global matrix and applies singular value decomposition (SVD) to extract dominant spatial modes. The resulting top-r low-rank attention masks, with r

Major depressive disorder (MDD) is a global mental health problem that significantly affects individuals’ daily activities. The diagnosis of MDD is a challenging issue due to people’s stigma and less interest in reaching the clinic for healthcare assistance. People prefer to share their thoughts and feelings through text posts on social media platforms. The main aim of this article is to bridge the gap between medical experts and depressed individuals in identifying the symptoms of MDD early to provide effective treatment before it reaches a critical stage. This article creates a “hybrid model of DistilBERT with a convolutional neural network (CNN)—(HDC),” combining the power of two different deep learning architectures, DistilBERT and CNN, along with advances in natural language processing (NLP) to detect symptoms of MDD in alignment with DSM-5 through analyzing content from social networks. Experiments are conducted using standard online tweet data. The data augmentation technique solves data imbalance problems and avoids model-biased predictions. Precision, recall, f1-score, and accuracy are metrics used to evaluate the current technique with other baseline models. Experimental results show that the “HDC” model achieved 94.13% accuracy and outperformed cutting-edge methodologies for detecting depression symptoms.

The class activation map (CAM) is useful in identifying significant image features that the convolutional neural network (CNN) model is considering while making the prediction. This is critical especially in medical diagnosis like scenarios. However, existing gradient-based methods like Grad-CAM often produce low-quality visualization results due to gradient errors despite their computational efficiency. On the other hand, non-gradient methods like Score-CAM produce quality visualization that comes with high computational costs. The proposed method SVD filters Grad-CAM (SVD-Grad-CAM), which leverages singular value decomposition (SVD) to overcome the limitations of Grad-CAM. SVD-Grad-CAM filters gradients within the gradient matrix to compute the weight of the feature map for a specific class. This filtering process is achieved by selecting the top k principal components from the SVD decomposition, which discards less important patterns and potential error data. Consequently, SVD-Grad-CAM enhances the quality of Grad-CAM by reducing the clutter of multiple region highlights. The MURA dataset, focusing on elbow study type, is utilized to assess CAM visualization quality, with a DenseNet-169 CNN model fine-tuned via transfer learning. A total of 564 validation radiographs are used in empirical comparison, showing that SVD-Grad-CAM improves average drop, average increase, maximum coherency, and Average DCC by 30%, 21.67%, 19.91% and 22.56% respectively, in comparison to Grad-CAM. Code:: https://github.com/ramaiahthota02/SVD-Grad-CAM-v1.git

Depression is a kind of emotion that negatively impacts people's daily lives. The number of people suffering from long-term feelings is increasing every year across the globe. Depressed patients may engage in self-harm behaviors, which occasionally result in suicide. Many psychiatrists struggle to identify the presence of mental illness or negative emotion early to provide a better course of treatment before they reach a critical stage. One of the most challenging problems is detecting depression in people at the earliest possible stage. Researchers are using Natural Language Processing (NLP) techniques to analyze text content uploaded on social media, which helps to design approaches for detecting depression. This work analyses numerous prior studies that used learning techniques to identify depression. The existing methods suffer from better model representation problems to detect depression from the text with high accuracy. The present work addresses a solution to these problems by creating a new hybrid deep learning neural network design with better text representations called “Fasttext Convolution Neural Network with Long Short-Term Memory (FCL).” In addition, this work utilizes the advantage of NLP to simplify the text analysis during the model development. The FCL model comprises fasttext embedding for better text representation considering out-of-vocabulary (OOV) with semantic information, a convolution neural network (CNN) architecture to extract global information, and Long Short-Term Memory (LSTM) architecture to extract local features with dependencies. The present work was implemented on real-world datasets utilized in the literature. The proposed technique provides better results than the state-of-the-art to detect depression with high accuracy.

Abstract Facial expression recognition has been a long‐standing problem in the field of computer vision. This paper proposes a new simple scheme for effective recognition of facial expressions based on a hybrid feature descriptor and an improved classifier. Inspired by the success of stationary wavelet transform in many computer vision tasks, stationary wavelet transform is first employed on the pre‐processed face image. The pyramid of histograms of orientation gradient features is then computed from the low‐frequency stationary wavelet transform coefficients to capture more prominent details from facial images. The key idea of this hybrid feature descriptor is to exploit both spatial and frequency domain features which at the same time are robust against illumination and noise. The relevant features are subsequently determined using linear discriminant analysis. A new least squares support vector machine parameter tuning strategy is proposed using a contemporary optimisation technique called Jaya optimisation for classification of facial expressions. Experimental evaluations are performed on Japanese female facial expression and the Extended Cohn–Kanade (CK+) datasets, and the results based on 5‐fold stratified cross‐validation test confirm the superiority of the proposed method over state‐of‐the‐art approaches.

Recommender system is a popular tool that aims to provide personalized suggestions to user about items, products, services, and so on. Recommender system has effectively been used in online social networks, especially the location-based social networks for providing suggestions for interesting places known as POIs (points-of-interest). Popular recommender systems explore historical data to learn users’ preferences and, subsequently, they recommend locations to an active user. This strategy faces a major problem when a new POI or business evolves in a city. New business has no historical user experience data. Thus, a recommender system fails to gather enough knowledge about the new businesses, resulting in ignoring them during recommendations. This scenario is popularly known as a cold-start POI problem. Users never get recommendations of the new businesses in a city even though they can be relevant to a user. Also, from a business owner’s perspective, such a recommendation strategy does not help its reachability among users. Therefore, it is important for a recommender system to remain updated with new businesses in a city and ensure that all relevant POIs are recommended to a user irrespective of their lifetime. A POI recommendation approach is proposed in this work that can effectively handle the new businesses, or the cold-start POI problem, in a city. We crowdsource descriptions of cold-start POIs from various online social networks. The reviews of users are exploited here to learn the inherent features at the existing POIs and the new crowdsourced POIs. Finally, the proposed approach recommends top- K POIs consisting of the existing and new POIs. We perform experiments on the real-world Yelp dataset, which is one of the largest available data resources containing details on a wide range of businesses, users, and reviews. The proposed approach is compared with four existing POI recommendation approaches. The obtained results show that our approach outperforms others in handling cold-start POIs.

Recommender System (RS) is an essential component of many businesses, especially in e-commerce domain. RS exploits the preference history (rating, purchase, review, etc.) of users in order to provide the recommendations. A user in traditional RS can provide only one rating value about an item. Deep Neural Networks have been used in this single rating system to improve recommendation accuracy in the recent times. However, the single rating systems are inadequate to understand the usersfi preferences about an item. On the other hand, business enterprises such as tourism, e-learning, etc. facilitate users to provide multiple criteria ratings about an item, thus it becomes easier to understand users' preference over single rating system. In this paper, we propose an extended Stacked Autoencoders (a Deep Neural Network technique) to utilize the multi-criteria ratings. The proposed network is designed to learn the relationship between each user's criteria and overall rating efficiently. Experimental results on real world datasets (Yahoo! Movies and TripAdvisor) demonstrate that the proposed approach outperforms state-of-the-art single rating systems and multi-criteria approaches on various performance metrics.

In Natural Language Processing (NLP), sarcasm analysis in the text is considered as the most challenging task. It has been broadly researched in recent years. The property of sarcasm that makes it harder to detect is the gap between the literal and its intended meaning. It is a particular kind of sentiment which is capable of flipping the entire sense of a text. Sarcasm is often expressed verbally through the use of high pitch with heavy tonal stress. The other clues of sarcasm are the usage of various gestures such as gently sloping of eyes, hands movements, shaking heads, etc. However, the appearances of these clues for sarcasm are absent in textual data which makes the detection of sarcasm dependent upon several other factors. In this article, six algorithms were proposed to analyze the sarcasm in tweets of Twitter. These algorithms are based on the possible occurrences of sarcasm in tweets. Finally, the experimental results of the proposed algorithms were compared with some of the existing state-of-the-art.

Sentiment Analysis is a technique to identify people's opinion, attitude, sentiment, and emotion towards any specific target such as individuals, events, topics, product, organizations, services etc. Sarcasm is a special kind of sentiment that comprise of words which mean the opposite of what you really want to say (especially in order to insult or wit someone, to show irritation, or to be funny). People often expressed it verbally through the use of heavy tonal stress and certain gestural clues like rolling of the eyes. These tonal and gestural clues are obviously not available for expressing sarcasm in text, making its detection reliant upon other factors. In this paper, two approaches to detect sarcasm in the text of Twitter data were proposed. The first is a parsing-based lexicon generation algorithm (PBLGA) and the second was to detect sarcasm based on the occurrence of the interjection word. The combination of two approaches is also shown and compared with the existing state-of-the-art approach to detect sarcasm. First approach attains a 0.89, 0.81 and 0.84 precision, recall and f -- score respectively. Second approach attains 0.85, 0.96 and 0.90 precision, recall and f -- score respectively in tweets with sarcastic hashtag.