In intent recognition for mobile application reviews, traditional sentiment analysis suffers from coarse granularity, insufficient feature extraction, and excessive reliance on absolute word positions.To address these issues, this paper proposes a multi-feature fusion model.This model integrates three key types of information: text embeddings, fine-grained aspect-based sentiment features, and topic representation features.A cross-attention mechanism is employed for feature fusion.Subsequently, a Circular and Dilated Convolutional Neural Network, which incorporates exponential dilation rates and circular padding mechanisms, is utilized for intent classification.Experiments conducted on multiple public datasets demonstrate that the proposed model significantly outperforms baseline models in terms of both accuracy and Macro-F1 score.Furthermore, ablation studies validate the effectiveness of each individual module.

The latest advancements in large language models (LLMs) have enabled machine-generated text (MGT) to reach a level comparable to human-written text (HWT), making it increasingly difficult to detect MGT.This poses potential risks such as fake news, social media spam, and plagiarism.Current detection methods primarily rely on fine-tuned pre-trained language models (PLMs), but their performance may degrade when dealing with texts from different domains or various large language models, especially when the exact source is unknown.In this paper, we propose a hybrid approach that combines neural networks with feature extraction.We leverage BERT to obtain semantic representations of the global context, allowing the deep learning model to achieve a deeper understanding of the text across multiple layers, while also incorporating statistical features for enhancement.We evaluate our method on three different datasets and set up three distinct experimental configurations: in-domain, out-of-domain, and in-the-wild.The results show that our method performs comparably to other approaches in the in-domain setting and outperforms fine-tuned BERT and other detection methods in the out-of-domain and in-the-wild scenarios.

The proliferation of fake reviews on e-commerce platforms has seriously and negatively affected consumers’ purchase decisions. In recent years, some researchers have started applying graph neural networks for fake review detection and achieved better performance than other machine learning and deep learning models. However, the existing work has rarely considered the influence of edge feature information on nodes when updating node information and only used the contextual order of words in the text when constructing a graph. In this paper, we propose a new method for constructing graphs and present a multi-attention graph neural network (MAGAT-HOS) that considers both node-level attention and edge-level attention. When creating the edges of the graph, we not only consider word context order semantic information, but also high-order semantic information such as the named entity information within review text. In addition, the graph learns node representations with two different aggregation functions, which are implemented based on node-level attention and edge-level attention. At last, all the nodes’ features are aggregated using a self-attention mechanism to emphasize the important information. Evaluation results show that the model and graph structure construction method proposed in this paper achieves better performance than the state-of-the-art baseline models.

In order to enhance brand benefits or discredit competitors, some merchants hire fake reviewers to post large amounts of fake reviews on e-commerce platforms. This behavior inevitably harms consumers’ interests and causes unfair market competition for other merchants. Researches on fake review detection mainly focus on mining the content of the reviews, the behavioral features of the reviewers, or building models using deep learning. However, most existing research have not taken into the differences in motivation between fake positive and fake negative reviews, the review time distribution features of true reviewers post reviews, and how to effectively integrate multi-modal features. In this paper, we collect restaurant review datasets from Yelp.com in three different regions, and propose a fake review detection method based on a neural network model called BERT-Multi feature-TextCNN-BiGRU-Attention(BMTBA). Firstly, we use the BERT pre-training model to train a restaurant review language model. Then, we propose to use a multimodal fusion method to combine the BERT pre-trained word vector sequences with extracted multidimensional statistical features as input(including a newly proposed reviewer feature called Review weekday). Finally, considering that the motivation for fake positive and fake negative reviews is different, we construct fake positive and fake negative model separately to detect them. Multiple ablation experiments are conducted on the three datasets mentioned above, and the results show that the proposed BMTBA model outperformed the baseline model (BERT-TextCNN-BiGRU-Attention) with a higher classification detection accuracy of 94.68%.

We constructed a tripartite game model that includes consumers, sellers, and platforms to analyze the impact of fake reviews on platform revenue under different market structures. Results indicated that, in a monopoly market when the platform charges sellers transaction fees, on the one side, fake reviews yield more transaction cost for both consumers and sellers, thus generating less revenue for the platform. On the other side, if sellers choose to manufacture more fake reviews, consumers' perceived quality of product will increase in early time and the overall impact of fake reviews on platform revenue may be positive; however, in the long run, fake reviews will damage platform revenue. Furthermore, consumers' increased perceived quality will offset some of the negative impact from transaction costs if sellers choose other enhancement strategies instead of participating in review fraud. Moreover, more fake reviews will yield lower platform revenue when platforms are in a duopoly market. This study provides a theoretical basis for the platform to manage fake reviews.

With the rapid development of Internet, the issue of cyber security has increasingly gained more attention. An intrusion Detection System (IDS) is an effective technique to defend cyber-attacks and reduce security losses. However, the challenge of IDS lies in the diversity of cyber-attackers and the frequently-changing data requiring a flexible and efficient solution. To address this problem, machine learning approaches are being applied in the IDS field. In this paper, we propose an efficient scalable neural-network-based hybrid IDS framework with the combination of Host-level IDS (HIDS) and Network-level IDS (NIDS). We applied the autoencoders (AE) to NIDS and designed HIDS using word embedding and convolutional neural network. To evaluate the IDS, many experiments are performed on the public datasets NSL-KDD and ADFA. It can detect many attacks and reduce the security risk with high efficiency and excellent scalability.

Current approaches for the security of embedded software mainly focused on some specific platforms. In this paper, a security model based on code obfuscation is applied to embedded software. A control flow flattening algorithm is used to implement an automated obfuscator, which obfuscates C code first, and does source-to-source conversions to protect software on different platforms. The effectiveness of code obfuscation is evaluated by a multi-level quantitative model proposed in this paper. Related experiments are carried out on the NUC140VE3CN board and MC9S12XEPIOOMAG board, which are typical hardware platforms used in the application domain of automotive. The result of experiments shows that for one thing, the quantitative value of the effectiveness of the obfuscated program is obviously higher than that of the original program, namely the strength for software to keep it from being reversed is greater, and the overhead of time and space is acceptable; for another, the efficiency of the evaluation model is also demonstrated.

Intercity transportation infrastructures and services determine the depth and breadth of the spatial interactions among cities within an urban agglomeration, and have profound impacts on the spatial structure of the urban agglomeration. To evaluate whether the public intercity ground transportation infrastructures and services (i.e. passenger trains and long-distance buses) can support the integration and development of urban agglomerations, we propose a method for ‘transportation cluster’ detection (TCD), which has three unique features: (1) the K-shortest paths are used to quantify the proximity between cities, which is more in line with people’s travel behaviors; (2) a dendrogram is obtained through hierarchical clustering to reveal the structural hierarchies of transportation clusters; and (3) the integration of geo-modularity and hierarchical clustering assures high strength of division of transportation networks. The proposed TCD method was applied to the network of passenger trains, the network of long-distance buses, and the combined network of both in mainland China, respectively. By comparing the resultant transportation clusters with the urban agglomerations delineated by the Chinese government, cities that have weak transportation connections with other cities within an urban agglomeration were identified, and such findings could help devise transportation planning to better support the integrated development of urban agglomerations.