MG-SIN: Multigraph Sparse Interaction Network for Multitask Stance Detection.

Stance detection on social media aims to identify if an individual is in support of or against a specific target. Most existing stance detection approaches primarily rely on modeling the contextual semantic information in sentences and neglect to explore the pragmatics dependency information of words, thus degrading performance. Although several single-task learning methods have been proposed to capture richer semantic representation information, they still suffer from semantic sparsity problems caused by short texts on social media. This article proposes a novel multigraph sparse interaction network (MG-SIN) by using multitask learning (MTL) to identify the stances and classify the sentiment polarities of tweets simultaneously. Our basic idea is to explore the pragmatics dependency relationship between tasks at the word level by constructing two types of heterogeneous graphs, including task-specific and task-related graphs (tr-graphs), to boost the learning of task-specific representations. A graph-aware module is proposed to adaptively facilitate information sharing between tasks via a novel sparse interaction mechanism among heterogeneous graphs. Through experiments on two real-world datasets, compared with the state-of-the-art baselines, the extensive results exhibit that MG-SIN achieves competitive improvements of up to 2.1% and 2.42% for the stance detection task, and 5.26% and 3.93% for the sentiment analysis task, respectively.

A Model-Agnostic Approach to Mitigate Gradient Interference for Multi-Task Learning.

Multitask learning (MTL) is a powerful technique for jointly learning multiple tasks. However, it is difficult to achieve a tradeoff between tasks during iterative training, as some tasks may compete with each other. Existing methods manually design specific network models to mitigate task conflicts, but they require considerable manual effort and prior knowledge about task relationships to tune the model so as to obtain the best performance for each task. Moreover, few works have offered formal descriptions of task conflicts and theoretical explanations for the cause of task conflict problems. In this article, we provide a formal description of task conflicts that are caused by the gradient interference problem of tasks. To alleviate this issue, we propose a novel model-agnostic approach to mitigate gradient interference (MAMG) by designing a gradient clipping rule that directly modifies the interfering components on the gradient interfering direction. Specifically, MAMG is model-agnostic and thus it can be applied to a large number of multitask models. We also theoretically prove the convergence of MAMG and its superiority to existing MTL methods. We evaluate our method on a variety of real-world large datasets, and extensive experimental results confirm that MAMG can outperform some state-of-the-art algorithms on different types of tasks and can be easily applied to various methods.

Perceiving Social-Emotional Volatility and Triggered Causes of COVID-19.

Health support has been sought by the public from online social media after the outbreak of novel coronavirus disease 2019 (COVID-19). In addition to the physical symptoms caused by the virus, there are adverse impacts on psychological responses. Therefore, precisely capturing the public emotions becomes crucial to providing adequate support. By constructing a domain-specific COVID-19 public health emergency discrete emotion lexicon, we utilized one million COVID-19 theme texts from the Chinese online social platform Weibo to analyze social-emotional volatility. Based on computed emotional valence, we proposed a public emotional perception model that achieves: (1) targeting of public emotion abrupt time points using an LSTM-based attention encoder-decoder (LAED) mechanism for emotional time-series, and (2) backtracking of specific triggered causes of abnormal volatility in a cognitive emotional arousal path. Experimental results prove that our model provides a solid research basis for enhancing social-emotional security outcomes.

Automated Vulnerability Discovery and Exploitation in the Internet of Things.

Recently, automated software vulnerability detection and exploitation in Internet of Things (IoT) has attracted more and more attention, due to IoT's fast adoption and high social impact. However, the task is challenging and the solutions are non-trivial: the existing methods have limited effectiveness at discovering vulnerabilities capable of compromising IoT systems. To address this, we propose an Automated Vulnerability Discovery and Exploitation framework with a Scheduling strategy, AutoDES that aims to improve the efficiency and effectiveness of vulnerability discovery and exploitation. In the vulnerability discovery stage, we use our Anti-Driller technique to mitigate the "path explosion" problem. This approach first generates a specific input proceeding from symbolic execution based on a Control Flow Graph (CFG). It then leverages a mutation-based fuzzer to find vulnerabilities while avoiding invalid mutations. In the vulnerability exploitation stage, we analyze the characteristics of vulnerabilities and then propose to generate exploits, via the use of several proposed attack techniques that can produce a shell based on the detected vulnerabilities. We also propose a genetic algorithm (GA)-based scheduling strategy (AutoS) that helps with assigning the computing resources dynamically and efficiently. The extensive experimental results on the RHG 2018 challenge dataset and the BCTF-RHG 2019 challenge dataset clearly demonstrate the effectiveness and efficiency of the proposed framework.

A large scale code resolution service network in the Internet of Things.

In the Internet of Things a code resolution service provides a discovery mechanism for a requester to obtain the information resources associated with a particular product code immediately. In large scale application scenarios a code resolution service faces some serious issues involving heterogeneity, big data and data ownership. A code resolution service network is required to address these issues. Firstly, a list of requirements for the network architecture and code resolution services is proposed. Secondly, in order to eliminate code resolution conflicts and code resolution overloads, a code structure is presented to create a uniform namespace for code resolution records. Thirdly, we propose a loosely coupled distributed network consisting of heterogeneous, independent; collaborating code resolution services and a SkipNet based code resolution service named SkipNet-OCRS, which not only inherits DHT’s advantages, but also supports administrative control and autonomy. For the external behaviors of SkipNet-OCRS, a novel external behavior mode named QRRA mode is proposed to enhance security and reduce requester complexity. For the internal behaviors of SkipNet-OCRS, an improved query algorithm is proposed to increase query efficiency. It is analyzed that integrating SkipNet-OCRS into our resolution service network can meet our proposed requirements. Finally, simulation experiments verify the excellent performance of SkipNet-OCRS.

Covariance, correlation matrix, and the multiscale community structure of networks.

Empirical studies show that real world networks often exhibit multiple scales of topological descriptions. However, it is still an open problem how to identify the intrinsic multiple scales of networks. In this paper, we consider detecting the multiscale community structure of network from the perspective of dimension reduction. According to this perspective, a covariance matrix of network is defined to uncover the multiscale community structure through the translation and rotation transformations. It is proved that the covariance matrix is the unbiased version of the well-known modularity matrix. We then point out that the translation and rotation transformations fail to deal with the heterogeneous network, which is very common in nature and society. To address this problem, a correlation matrix is proposed through introducing the rescaling transformation into the covariance matrix. Extensive tests on real world and artificial networks demonstrate that the correlation matrix significantly outperforms the covariance matrix, identically the modularity matrix, as regards identifying the multiscale community structure of network. This work provides a novel perspective to the identification of community structure and thus various dimension reduction methods might be used for the identification of community structure. Through introducing the correlation matrix, we further conclude that the rescaling transformation is crucial to identify the multiscale community structure of network, as well as the translation and rotation transformations.