ABSTRACT With advancements in technology, the number of people using various social networks to share information daily has increased exponentially. To spread information to maximum users, there is a need to effectively identify highly influential nodes in social networks. Still, existing centrality measures and methods have drawbacks relating to computational complexity and accuracy. The proposed method considers the Game of Thieves algorithm to rank influential nodes in social networks effectively with lower computational time that outperforms existing closeness centrality, eigenvector, Game Theory, and VoteRank++ algorithms. To evaluate the performance of the proposed work, experiments were conducted on synthetic and real‐world network datasets. The results were evaluated with the help of the Susceptible‐Infected‐Recovered (SIR) model and Node Removal Procedure (NRP) evaluation metrics. These results showed that GOT outperforms the existing measures by 12% on average while giving more accurate results than the standard algorithms for large‐scale networks.

The increasing prevalence of stress-related issues highlights the urgent need for effective stress detection and intervention strategies in counseling. Stress, a common precursor to various psychological and physical health problems, arises from multiple sources and manifests through diverse life events. Understanding these sources or stressors is essential for providing targeted and effective counseling. Stressors and life events often influence and compound each other, creating a network of interrelated factors contributing to an individual’s overall stress level. This study proposes a contextualized MentalRoBERTa and Graph Transformer based model to identify these stressors and significant stressful life events. Graph transformer allow nodes to selectively attend to messages received from neighbors using multihead attention, enhancing their capacity to understand intricate relationships. The proposed model was evaluated on the stress annotated dataset (SAD), and a synthetically generated dataset. The results show that the proposed model outperforms existing methods, achieving F1-scores of 67.93 on the SAD dataset, and 97.26 on the SMOTE dataset. These findings demonstrate the effectiveness of our approach in accurately identifying and classifying stressors, making it a valuable tool for stress analysis and intervention.

Abstract Interaction via social media involves frequent code‐mixed text, spelling errors and noisy elements, which creates a bottleneck in the performance of natural language processing applications. This proposed work is the first approach for code‐mixed Hindi‐English social media text that comprises language identification, detection and correction of non‐word (Out of Vocabulary) errors as well as real‐word errors occurring simultaneously. Each identified language (Devanagari Hindi, Roman Hindi, and English) has its own complexities and challenges. Errors are detected individually for each language and a suggestive list of the erroneous words is created. After this, a fuzzy graph between different words of the suggestive lists is generated using various semantic relations in Hindi WordNet. Word embeddings and Fuzzy graph‐based centrality measures are used to find the correct word. Several experiments are performed on different social media datasets taken from Instagram, Twitter, YouTube comments, Blogs, and WhatsApp. The experimental results demonstrate that the proposed system corrects out‐of‐vocabulary words as well as real‐word errors with a maximum recall of 0.90 and 0.67, respectively for Dev_Hindi and 0.87 and 0.66, respectively for Rom_Hindi. The proposed method is also applied for state‐of‐art sentiment analysis approaches where the F1‐score has been visibly improved.

Social networking sites, blogs, and online articles are instant sources of news for internet users globally. However, in the absence of strict regulations mandating the genuineness of every text on social media, it is probable that some of these texts are fake news or rumors. Their deceptive nature and ability to propagate instantly can have an adverse effect on society. This necessitates the need for more effective detection of fake news and rumors on the web. In this work, we annotate four fake news detection (EFN) and rumor detection datasets with their emotion class labels using transfer learning. We show the correlation between the legitimacy of a text with its intrinsic emotion for fake news and rumor detection, and prove that even within the same emotion class, fake and real news are often represented differently, which can be used for improved feature extraction. Based on this, we propose a multitask framework for fake news and rumor detection, predicting both the emotion and legitimacy of the text. We train a variety of deep learning models in single-task (STL) and multitask settings for a more comprehensive comparison. We further analyze the performance of our multitask approach for EFN in cross-domain settings to verify its efficacy for better generalization across datasets, and to verify that emotions act as a domain-independent feature. Experimental results verify that our multitask models consistently outperform their STL counterparts in terms of accuracy, precision, recall, and F1 score, both for in-domain and cross-domain settings. We also qualitatively analyze the difference in performance in STL and multitask learning (MTL) models.

Depression, a prevalent mental health concern, requires timely identification and intervention. Automating the early stage identification of depression cues within social media text has become critically important. Existing methods for depression identification through text obtained from social media have shown promising results; however, these methods do not address the graphlike nature of social media data. To date, there has been little work addressing this problem by appropriately modeling social media data. This article aims to effectively harness the power of multiembedding techniques and graph attention networks (GATs) for depression identification. To overcome these limitations, the current study presents a novel methodology that combines multiple token-level embeddings, including BERT, RoBERTa, and DeBERTa, with GATs to leverage the graphlike nature of social media data and detect depression cues from such text. A dataset comprising 100 000 tweets has been curated using data from publicly available annotated tweet datasets. This dataset maintains a balanced distribution between depressive and nondepressive text samples. This was followed by a rigorous cleaning and preprocessing pipeline. Next, these data were transformed into a numeric feature matrix using multiple-word embeddings, which enabled the treatment of tweets as nodes in a graphlike structure. This graph was used for training a GAT model with multiple self-attentional layers, culminating in a linear layer for mapping traits to binary classes. The model presented in the current study achieves a precision of 97.2%, a recall of 96.4%, and an F1 score of 96.7% on a benchmark dataset.

Social media, blogs, and online articles are instant sources of news for internet users globally. But due to their unmoderated nature, a significant percentage of these texts are fake news or rumors. Their deceptive nature and ability to propagate instantly can have an adverse effect on society. In this work, we hypothesize that legitimacy of news has a correlation with its emotion, and propose a multi-task framework predicting both the emotion and legitimacy of news. Experimental results verify that our multi-task models outperform their single-task counterparts in terms of accuracy.

Consumer reviews online may contain suggestions useful for improving the target products and services. Mining suggestions is challenging because the field lacks large labelled and balanced datasets. Furthermore, most prior studies have only focused on mining suggestions in a single domain. In this work, we introduce a novel up-sampling technique to address the problem of class imbalance, and propose a multi-task deep learning approach for mining suggestions from multiple domains. Experimental results on a publicly available dataset show that our up-sampling technique coupled with the multi-task framework outperforms state-of-the-art open domain suggestion mining models in terms of the F-1 measure and AUC.

In this article, we propose a system called “UTTAM,” for correcting spelling errors in Hindi language text using supervised learning. Unlike other languages, Hindi contains a large set of characters, words with inflections and complex characters, phonetically similar sets of characters, and so on. The complexity increases the possibility of confusion and occasionally leads to entering a wrong character in a word. The existence of spelling errors in text significantly decreases the accuracy of the available resources, like search engine, text editor, and so on. The proposed work is the first approach to correct non-word (Out of Vocabulary) errors as well as real-word errors simultaneously in a sentence of Hindi language. The proposed method investigates the human behavior, i.e., the type and frequency of spelling errors done by humans in Hindi text. Based on the type and frequency of spelling errors, the heterogeneous data is collected in matrices. This data in matrices is used to generate the suitable candidate words for an input word. After generating candidate words, the Viterbi algorithm is applied to perform the word correction. The Viterbi algorithm finds the best sequence of candidate words to correct the input sentence. For Hindi, this work is the first attempt for real-word error correction. For non-word errors, the experiments show that “UTTAM” performs better than the existing systems SpellGuru and Saksham.