Most existing Chinese spam detection models suffer such problems as inaccurate representation, unsatisfactory detection effect and poor practicality. To address these problems, a capsule network model combining the long-short attention mechanism is proposed here to achieve efficient Chinese spam detection. For text representation, the proposed model uses a multi-channel structure based on the long-short attention mechanism, which can capture complex text features in spam and generate contextual word vectors with more semantic information. For feature mining and classification, the model improves the structure of the traditional capsule network without compromising the classification performance and optimizes the dynamic routing algorithm, so that the model has a high accuracy without reducing the running speed. Experimental results show that the model outperformed the current mainstream methods such as TextCNN, LSTM and even BERT in characterization and detection; and it achieved an accuracy as high as 98.72% on an unbalanced dataset and 99.30% on a balanced dataset.

To address the shortcomings of existing deep learning models and the characteristics of microblog speech, we propose the DCCMM model to improve the effectiveness of microblog sentiment analysis. The model employs WOBERT Plus and ALBERT to dynamically encode character-level text and word-level text, respectively. Then, a convolution operation is used to extract local key features, while cross-channel feature fusion and multi-head self-attention pooling operations are used to extract global semantic information and filter out key data, before using the multi-granularity feature interaction fusion operation to effectively fuse character-level and word-level semantic information. Finally, the Softmax function is used to output the results. On the weibo_senti_100k dataset, the accuracy and F1 values of the DCCMM model improve by 0.84% and 1.01%, respectively, compared to the best-performing comparison model. On the SMP2020-EWECT dataset, the accuracy and F1 values of the DCCMM model improve by 1.22% and 1.80%, respectively, compared with the experimental results of the best-performing comparison model. The results showed that DCCMM outperforms existing advanced sentiment analysis models.

Human activities are usually collective, so clustering has become an important feature of human behavior. This paper studied the evolution of the community in the process of public opinion propagation so as to put forward a public opinion evolution model for the network community number. This study proposed the community number evolution model of public opinion based on stochastic competitive learning, and the proposed model consists of an increase in the number of communities and a decrease in the number of communities. The highlight of this model is that on the one hand, it realizes the research on the evolution of public opinions on the dynamic network; on the other hand, unlike other public opinion evolution models, this model pays attention to the community number increase and decrease rules in the evolution of public opinions. Then, as an extension of the community number evolution model of public opinion, the community number prediction model had been proposed. Based on Twitter data from the 2017 London Bridge attack, the proposed models were validated by experiments. In the verification section of this paper, two methods had been introduced as a comparison. The experimental results show that the community number evolution model of public opinion is correct.

In recent years, the number of malicious web pages has increased dramatically, posing a great challenge to network security. While current machine learning-based detection methods have emerged as a promising alternative to traditional detection techniques. However, these methods are commonly based on single-modal features or simple stacking of classifiers built on various features. As a result, these techniques are not capable of effectively fusing features from different modalities, ultimately limiting the detection effectiveness. To address this limitation, we propose a malicious web page detection method based on multi-modal learning and pre-trained models. First, in the input stage, the raw URL and HTML tag sequences of web pages are used as input features. To help the subsequent model learn the relationship between the two modalities and avoid information confusion, modal-type encoding, and positional encoding are introduced. Next, a single-stream neural network based on the ConvBERT pre-trained model is used as the backbone classifier, and it learns the representation of multi-modal features through fine-tuning. For the output part of the model, a linear layer based on large margin softmax is applied to the decision-making. This activation function effectively increases the classification boundary and improves the robustness. In addition, a coarse-grained modal matching loss is added to the model optimization objective to assist the models in learning the cross-modal association features. Experimental results on synthetic datasets show that our proposed method outperforms traditional single-modal detection methods in general, and has advantages over baseline models in terms of accuracy and reliability.

With the widespread use of SMS (Short Message Service), the proliferation of malicious SMS has emerged as a pressing societal issue. While deep learning-based text classifiers offer promise, they often exhibit suboptimal performance in fine-grained detection tasks, primarily due to imbalanced datasets and insufficient model representation capabilities. To address this challenge, this paper proposes an LLMs-enhanced graph fusion dual-stream Transformer model for fine-grained Chinese malicious SMS detection. During the data processing stage, Large Language Models (LLMs) are employed for data augmentation, mitigating dataset imbalance. In the data input stage, both word-level and character-level features are utilized as model inputs, enhancing the richness of features and preventing information loss. A dual-stream Transformer serves as the backbone network in the learning representation stage, complemented by a graph-based feature fusion mechanism. At the output stage, both supervised classification cross-entropy loss and supervised contrastive learning loss are used as multi-task optimization objectives, further enhancing the model’s feature representation. Experimental results demonstrate that the proposed method significantly outperforms baselines on a publicly available Chinese malicious SMS dataset.