As social media usage proliferates, cyberbullying emerges as a severe social issue, with an increasing demand for effective detection methods. Current approaches suffer from three major limitations: (1) they often focus solely on textual or visual features while ignoring sentiment features; (2) they fail to effectively capture the attack patterns of offensive comments in cyberbullying sessions; and (3) existing methods typically use simple linear classification layers at the end of their models, making it difficult to discriminate between bullying and non-bullying sessions, especially when the differences are subtle. In this study, we first identify two primary attack patterns of cyberbullying, burst attacks that multiple offensive comments appear consecutively within short time periods, and high-density attacks where offensive comments appear throughout a comment sequence but may be interspersed with non-offensive comments. Then, to address the challenges, a novel cyberbullying detection method, Multi-Scale Sensing Network (MultiSenseNet), is proposed to capture both burst and high-density attacks through a multi-scale sliding window selector, integrate semantic and sentiment features, and replace linear layers with contrastive learning for an enhanced discrimination.

In the field of cyberbullying detection, it is crucial to understand and analyze the structural features of media sessions. However, most studies ignore the dynamic correlation between comments and temporal information as well as temporal pattern matching among sessions. To address the problem, this study proposes a Time-User Heterogeneous Neural Interaction Network (TUHIN) model. It consists of five modules: session encoding module, comment interaction module, session-time attention module, user interaction module and aggregation module. Specifically, considering comment and temporal information, we adopt a hierarchical structure to extract features from the comment and session levels of media sessions. Then, in order to analyze the interaction patterns among users, we use graph convolutional networks to model the influential situations of users participation in media sessions. Comparing with the baseline models on public Instagram and Vine datasets, our model performs better on both Recall and F1.

The rapid development and explosive usage of social media make cyberbullying detection a major concern for society. However, many existing studies only focus on textual contents and ignore multimodal social media data, including texts, images, videos, etc. Although there are a few studies on multimodal cyberbullying detection, they suffer from the following limitations: 1) fail to extract features of each modality sufficiently; 2) largely ignore the significance of multimodal correlation for cyberbullying detection; 3) consider each social media session as independent. To address these issues, we propose a novel Multimodal Correlation and Interaction-based Cyberbullying Detection method (MCICD). Specifically, we construct a post-image co-attention sub-network to capture correlations between texts and images, and design a heterogeneous user-postimage interaction sub-network to explore dependencies among social media sessions, which contain multimodal information. Experimental results on two real-world session-level social media datasets demonstrate the effectiveness of our proposed method. Additionally, it is also verified that our method performs well on the early detection of cyberbullying.

This paper proposes a new Petri-net-based Q-learning scheduling method to schedule robotic cellular manufacturing (RCM) systems efficiently. First, we use generalized and place-timed Petri nets to model RCM systems. Then, we design a reinforcement learning method with a sparse Q-table to evaluate state-transition pairs of the net’s reachability graph. It uses the negative transition firing time as a reward for an action selection and adopts a large penalty for any encountered deadlock. In addition, it balances the state space exploration and the experience exploitation by using a dynamic ϵ-greedy policy to update the state values with an accumulative reward. Three different dynamic ϵ-greedy policies are designed for different application scenarios. Some benchmark RCM systems are tested with the proposed method and several popular PN-based online dispatching rules, such as FIFO and SRPT. Simulation results demonstrate that our method schedules RCM systems as quickly as the online dispatching rules while outperforming them in terms of schedule makespan. For readers’ reference, our source code and test data are available at https://github.com/PNOptimizer/PNQL.

Transfer learning has become a key technique in deep learning, widely adopted in the industry and academia for developing customized models, especially for specific and downstream tasks solving. Despite the prevalence of transfer learning, the target model can easily inherit defects from the source model during the learning process, such as vulnerability to backdoor attacks and adversarial attacks. Thus, this work proposes a novel approach, iterative pruning learning approach (IPLA), that reduces the inheritance of potential defects during the transfer learning process. In order to reduce the vulnerability to attacks and improve the robustness of target model, IPLA evaluates the importance of weights from the source model and retains ones that are critical to the target task, then prunes the redundant weights through an iterative pruning process. Experiments are performed on 4 datasets over 2 backbone source models. Results demonstrate the satisfactory performance of our proposed method.

A rare event, such as a natural disaster, does not occur frequently, but may cause catastrophic impacts on human beings and their living environment once it happens. Social media provides people an immediate and convenient way to share their opinions. Researchers can thus use it for investigating people's actions, feelings, and attitudes during multiple rare events. By using social media data, many studies try to build the connection between rare events in a real world and people's responses, including feelings, attitudes, and behaviors, in a virtual world. In this article, we propose a feature extraction and text classification approach to Twitter text data related to a rare event, e.g., Hurricane Sandy. First, a novel feature extraction method is proposed to mine useful features from each text message. Next, a fuzzy logic-based classification method is proposed to distinguish event-related and unrelated messages. Finally, our proposed method is compared with the existing keyword search one that is widely used in analyzing the evolution of a rare event. The result reveals that the proposed approach is suitable to identify the rare event-related and unrelated text messages.

Sentiment analysis is a process of analyzing, processing, concluding, and inferencing subjective texts with the sentiment. Companies use sentiment analysis for understanding public opinion, performing market research, analyzing brand reputation, recognizing customer experiences, and studying social media influence. According to the different needs for aspect granularity, it can be divided into document, sentence, and aspect-based ones. This article summarizes the recently proposed methods to solve an aspect-based sentiment analysis problem. At present, there are three mainstream methods: lexicon-based, traditional machine learning, and deep learning methods. In this survey article, we provide a comparative review of state-of-the-art deep learning methods. Several commonly used benchmark data sets, evaluation metrics, and the performance of the existing deep learning methods are introduced. Finally, existing problems and some future research directions are presented and discussed.

Power demand side management (DSM) regulations stipulate that electric power grid enterprises of cities and provinces must complete their target of power saving through efficient power plant (EPP) programs. This investigation studies power DSM and establishes a supervision and evaluation system for the EPP program. The proposed system can be used to evaluate the performance of an EPP program such as electricity saving and investment benefit. A model of EPP is established. A cuckoo search algorithm is applied for the model optimization and its result gives an optimal decision scheme. Finally, a fuzzy comprehensive evaluation method is used to assess an EPP program. The results show that the established supervision and evaluation system and the model of an EPP program are effective and applicable for power grid enterprises.

Exploration and discovery of the relationship between social media activities and rare-event evolution have been investigated by many researchers in recent years. Their investigations have revealed the existence of such relationship. Furthermore, some researchers regard finding either a temporal or spatial pattern of social media activities as a way to evaluate the evolution of rare event. However, most of them fail to deduce an accurate time point when a rare event highly impacts social media activities. This paper concentrates on the intensity of information volume and proposes an innovative data processing method based on clustering algorithms. The proposed method can characterize the evolution of a rare event in the real world by analyzing social media activities in the virtual world. This exploration contributes to study changes of social media activities in the time domain. A case study is based on Hurricane Sandy that occurred in 2012. Social media data collected from Twitter during its arrival time span are adopted to evaluate the feasibility and effectiveness of our proposed method. First, this paper confirms that a strong correlation between a rare event and social media activities does exist. Next, it uncovers that a time difference does exist between the real and virtual worlds. In general, this paper gives a novel idea that deduces a temporal pattern of social media activities during the occurrence of rare events.

Person re-identification via cross-camera is a difficult problem in the field of target discovery and tracking. Traditional solutions depending on the characteristics of a target's appearance have low reliability and can easily lead to low matching rate because they use simple metric functions. This work proposes a more reliable measurement: correlative degree of a target's features among different camera views to do person re-identification and uses it to measure the histograms' similarity of targets. In order to obtain more discriminative features, we need to extract their appearance and space features. We use multi-layer histograms to describe them. In order to compute more accurate correlation degree, we propose to use Gaussian pyramid as alternating distance to define high-dimensional diffusion distance. Finally, we assign different weights to feature vectors so as to establish the correlative degree function based on diffusion distance. Experiments of person re-identification for different cameras show that the proposed method can achieve much better results than some known existing methods.

Recently, some researchers attempt to find a relationship between the evolution of rare events and temporal-spatial patterns of social media activities. Their studies verify that the relationship exists in both time and spatial domains. However, few of them can accurately deduce a time point when social media activities are highly affected by a rare event. Thus, it is difficult to characterize an accurate temporal pattern of social media during the evolution of a rare event. This work proposes an innovative method to characterize the evolution of a rare event by analyzing social media activities. We find that there is a time difference between the event and social media activities in a time domain. This is conducive to investigate the temporal pattern of social media activities. The proposed method focuses on the intensity of information volume by adopting a clustering algorithm. Our case study focuses on a hurricane named Sandy in 2012. Twitter data collected around it is used to verify the effectiveness of the method. The results not only verify that a rare event and social media activities have strong correlation, but also reveal that they have a time difference. This work provides an effective and reliable method to find a temporal pattern of social media when a rare event occurs.

Social media offer abundant information for studying people's behaviors, emotions and opinions during the evolution of various rare events such as natural disasters. It is useful to analyze the correlation between social media and human-affected events. This study uses Hurricane Sandy 2012 related Twitter text data to conduct information extraction and text classification. Considering that the original data contains different topics, we need to find the data related to Hurricane Sandy. A fuzzy logic-based approach is introduced to solve the problem of text classification. Inputs used in the proposed fuzzy logic-based model are multiple useful features extracted from each Twitter's message. The output is its degree of relevance for each message to Sandy. A number of fuzzy rules are designed and different defuzzification methods are combined in order to obtain desired classification results. We compare the proposed method with the well-known keyword search method in terms of correctness rate and quantity. The result shows that the proposed fuzzy logic-based approach is more suitable to classify Twitter messages than keyword word method.