In order to solve the problem of low face recognition accuracy of traditional algorithms and excessive long training time in deep learning methods, a novel lightweight and short training time multiface recognition method is proposed in this paper. Firstly, an extraction model of facial feature vectors is established based on local binary mode and principal component analysis. Secondly, the beetle antennae search algorithm (BAS) is optimized using adaptive factors, and the ABAS algorithm is proposed. This paper uses the ABAS algorithm to optimize the initial threshold of the neural network and proposes the ABASNet method. Thirdly, ABASNet method is combined with the face features extracted by the face feature vector extraction model to be used for multi-face classification tasks. Finally, H-softmax (hierarchical softmax) is used to replace softmax in neural networks. By reducing the amount of calculation in the process of multi-category face classification, the training time of the ABASNet network is reduced. Through ablation experiments and comparative experiments with various methods such as IKDA + PNN algorithm and PCANet algorithm, to verify the accuracy, robustness and short training time of the ABASNet method. Among them, the maximum accuracy of the method in the ORL face dataset, ExtYaleB face dataset and FERET face dataset is 99.35%, 99.54% and 99.18% respectively. In addition, actual test results indicate that the training time and recognition time of the method in this paper are 21 seconds and 0.06 seconds respectively, which has demonstrated the lightweight and real-time performance of the proposed method. At the same time, more test results show that the proposed ABASNet method can also be implemented in embedded devices and other classification tasks.

With the popularity of the internet, short videos have become an indispensable tool to obtain health information. However, avoiding health disinformation owing to the openness of the Internet is difficult for users. Disinformation may endanger the health and lives of users. With a focus on the process of identifying short videos' health disinformation and the factors affecting the accuracy of identification, this study aimed to investigate the identification methods, coping strategies, and the impact of short videos’ health disinformation on undergraduate nursing students. The findings will provide guidance to users on obtaining high-quality and healthy information, in addition to reducing health risks. Semi-structured in-depth interviews were conducted with 22 undergraduate nursing students in October 2022, and data were collected for collation and content analyses. The techniques used to identify short videos that include health disinformation as well as how undergraduate nursing students perceived these videos' features are among the study's findings. The failure factors in identification, coping paths, and adverse impacts of short videos on health disinformation were analyzed. The platform, the material itself, and the students' individual characteristics all have an impact on their identifying behavior. Medical students continue to face many obstacles in identifying and responding to health disinformation through short videos. Preventing and stopping health disinformation not only requires individual efforts to improve health literacy and maintain rational thinking, it also requires the joint efforts of short video producers, relevant departments, and platforms.

This study focuses on the evolution of the Internet Water Army on social media, identifying a novel form known as artificial-intelligence-generated-content-enhanced social bots (AESBs), and compares their structural influence with traditional social bots in the context of public opinion guidance. Based on 3 years of real-world data from Weibo, this study develops a comprehensive framework integrating bot account detection, AESB content identification, and quantitative assessments of opinion guidance. A large-scale opinion propagation network is constructed to examine the structural roles of traditional social bots and AESB across three analytical levels: the node, community, and overall network. The results reveal substantial differences between AESB and traditional social bots. Social bots play a limited guiding role but help maintain network connectivity. In contrast, AESBs produce highly consistent and human-like content that demonstrates a significant capacity to reinforce topic focus, amplify emotional homogeneity, and deepen diffusion pathways, indicating a shift toward strategic content manipulation. These results suggest that AESBs are not merely passive generators but active agents of structural opinion control, capable of combining human mimicry with machine-level efficiency. This study advances theoretical understanding of IWA manipulation mechanisms, provides a replicable methodological approach, and offers practical implications for platform governance.

Aiming at the problem that it takes too long to manually label numerous semantic segmentation data sets of vehicle images, a fast and effective vehicle image semantic segmentation data augmentation algorithm is proposed. Firstly, to solve the problem that traditional data augmentation algorithms are difficult to generate vehicle images and corresponding labels at the same time, a vehicle image semantic segmentation data augmentation algorithm based on FCN (Fully Convolutional Network) and GCIoU (Generally Contour Intersection over Union) is proposed, which can simultaneously generate vehicle images and corresponding labels. Then, aiming at the problem that some low-quality data exist in the generated dataset, a data set quality discriminator based on LQE (Label Quality Evaluation) head is proposed. The discriminator can distinguish between low-quality and high-quality label files. Finally, aiming at the problem that the excessive weight of the label file causes the calculation speed to decrease, a lightweight algorithm for the label file based on BAS-DP (Beetle Antennae Search Douglas-Peucker Algorithm) is proposed. The lightweight algorithm can greatly decrease parameters of the label file and improve availability of data-augmented results. Experimental results show that the proposed data augmentation algorithm is better than DCGAN (Deep Convolutional Generative Adversarial Networks), WGAN (Wasserstein Generative Adversarial Networks) and other data augmentation algorithms in accuracy. The AP50 and AP75 of the proposed algorithm reach 0.924 and 0.41, respectively. In addition, the proposed data augmentation algorithm still performs well in scenes with single-object, multi-object and ultra-multi-object. Simultaneously, the proposed data augmentation algorithm has three advantages, which are higher accuracy, faster speed, and less training data required.