Detian Zeng

Bio: Detian Zeng is an academic researcher from Central South University. The author has contributed to research in topics: Computer science & Artificial intelligence. The author has an hindex of 2, co-authored 2 publications receiving 43 citations.

Spark-based big data hybrid model mobile recommending method

Abstract: The invention puts forward a Spark-based big data hybrid model mobile recommending method comprising four steps as follows: firstly, getting a user's commodity purchase data at a mobile end; secondly, extracting user historical data from a database and importing the user historical data to an HDFS, and extracting features, such as user behavior features, brand features, user's personal consumption features and cross features; thirdly, packaging a hybrid model on a Spark platform using an RDD operator, and embedding the model interface into a big data platform for calling; and fourthly, calling the hybrid model interface to extract feature data, setting training parameters of the model, and training the hybrid model The model is estimated using a test data set and optimized, the trained hybrid model is saved, and relevant recommendation is made The method can effectively improve the efficiency of recommendation under the condition of a larger amount of data and higher data sparseness

Intrusion detection framework based on causal reasoning for DDoS

Abstract: Among network security issues, distributed denial of service (DDoS) attacks are particularly harmful to a network. Several previous machine learning (ML)-based network intrusion detection approaches have been developed to protect against DDoS attacks. However, existing ML detection approaches diagnose the causality between attacks and traffic features based mainly on purely associative features. Causal reasoning shows that this inability to disentangle correlation from causation can result in diagnostic errors. To solve this problem, this paper proposes a framework of DDoS detection based on causal reasoning to solve the problem of false associations. This framework consists of two main parts: feature selection based on “do-operations” and attack detection by counterfactual diagnosis. First, the noise features that are falsely associated with DDoS attacks are deleted during the “do-operations”. Then, the expected number of anomaly features under different DDoS attack types is calculated in the counterfactual situations. The larger the expected value that is calculated for a certain attack, the more likely it is that the anomaly features of the testing data are caused by this attack. The experiments show that the causality between DDoS attacks and the anomaly features can be fully described by our method, which, compared to other classic ML associative methods, increases the detection accuracy by approximately 5% on average.

Stgat-Mad : Spatial-Temporal Graph Attention Network For Multivariate Time Series Anomaly Detection

Abstract: Anomaly detection in multivariate time series data is challenging due to complex temporal and feature correlations. This paper proposes a novel unsupervised multi-scale stacked spatial-temporal graph attention network for multivariate time series anomaly detection (STGAT-MAD). The core of our framework is to coherently capture the feature and temporal correlations among multivariate time-series data by stackable STGAT networks. Meanwhile, a multi-scale input network is exploited to capture the temporal correlations in different time-scales. Besides, a new dataset derived from a real-world wind farm is built and released for multivariate time series anomaly detection. Experiments on the proprietary dataset and three public datasets show that our method significantly outperforms existing baseline approaches, and provides interpretability for anomaly location.

Improving the Stability of Intrusion Detection With Causal Deep Learning

Abstract: Due to factors such as differing distributions of training data and test data, false associations between features and weight associations lead to unstable detection performance and lack of generalization of network intrusion detection systems (NIDSs) based on machine learning (ML). To improve the stability and generalization of NIDSs, a detection system based on causal deep learning is proposed in this paper. First, causal weights were optimized by the propensity score through causal effects, the correlation between causal features and attack labels was increased, and the correlation between false correlation variables was weakened to improve the stability performance. Second, the approximate numerical optimization method of the Tammes problem was used to remove correlations between weights, maintain the independence of causal features, and improve the generalization of the detection system. Last, the feature distribution was disrupted by adding noise to four datasets to simulate different network environments. The results showed that our system can achieve good stability in various network environments where the training and testing datasets are not independently and identically distributed. In particular, after applying binary coding features and causal intervention (CIT) screening features, the average stability of the system improved by more than 10%.

Application recommending method and system based on user portrait behavior analysis, storage medium and computer device

Abstract: The invention provides a game recommending method and system based on user portrait behavior analysis. A feature acquirer is constructed to process user portrait data, application list data and client-reported data to obtain regular feature vectors meeting mathematical modeling requirements; various basic recommending models are used to make predictions to generate a primary user application recommendation list and corresponding download probabilities; a final application recommendation list is generated by combining the download probabilities and an actual label training fusion model. User historical behavior logs are subjected to multidimensional analysis to perform feature extraction so as to construct a user portrait data warehouse. A long and short-term memory network is introduced to the basic recommending models to learn time-series relationships of user behaviors, users' degrees of preference for objects are better depicted, and recommended game applications match well with the needs of users. Integrated learning is added to perform model fusion, learning results of the models are integrated, and accordingly, the stability and generalization ability of a recommending algorithm are improved.

Pythagorean fuzzy AHP based risk assessment methodology for hazardous material transportation: an application in Istanbul

Abstract: Environment and social life are open to hazards, because of the distribution, diffusion, and conversion processes of chemicals contained in hazardous materials. These chemicals are very dangerous. Various precautions should be taken into consideration during the displacement of hazardous materials. Therefore, it is important to identify and minimize the risks in the transportation of hazardous material. This work investigates to identify the critical risk factors and their weights for hazardous material transportation operations. The literature is reviewed, critical risk factors for hazardous material transportation are defined, and data from different experts is collected. A two-level hierarchical structure is established to evaluate risk factors. Then, the experts’ evaluations of main and sub-risk factors are consolidated using the modified Delphi method. Weights of main and sub-risk factors are obtained using the Pythagorean fuzzy analytic hierarchy process method. To show the robustness of the proposed decision-making methodology, a sensitivity analysis is conducted.

Improving Energy Efficiency of Multimedia Content Dissemination by Adaptive Clustering and D2D Multicast

Abstract: While achieving desired performance, there exist still many challenges in current cellular networks to support the multimedia content dissemination services. The conventional multimedia transmission schemes tend to serve all multicast group members with the data rate supported by the receiving user with the worst channel condition. The recent work discusses how to provide satisfactory quality of service (QoS) for all receiving users with different quality of experience (QoE) requirements, but the energy efficiency improvement of multimedia content dissemination is not its focus. In this paper, we address it based on adaptive clustering and device-to-device (D2D) multicast and propose an energy-efficient multimedia content dissemination scheme under a consistent QoE constraint. Our scheme extends the recent work with the proposed K-means-based D2D clustering method and the proposed game-based incentive mechanism, which can improve energy efficiency of multimedia content dissemination on the premise of ensuring the desired QoE for most multicast group members. In the proposed scheme, we jointly consider the cellular multicast, intracluster D2D multicast, and intercluster D2D multicast for designing the energy-efficient multimedia content dissemination scheme. In particular, we formulate the energy-efficient multicast transmission problem as a Stackelberg game model, where the macro base station (MBS) is the leader and the candidate D2D cluster heads (DCHs) are the followers. Also, the MBS acts as the buyer who buys the power from the candidate DCHs for intracluster and intercluster D2D multicast communications, and the candidate DCHs act as the sellers who earn reward by helping the MBS with D2D multicast communications. Through analyzing the above game model, we derive the Stackelberg equilibrium as the optimal allocation for cellular multicast power, intracluster D2D multicast power, and intercluster D2D multicast power, which can maximize the MBS’s utility function. Finally, the proposed scheme is verified through the simulation experiments designed in this paper.

Commodity recommendation method based on mobile electronic commerce of big data

Abstract: The invention requests to protect a commodity recommendation method based on mobile electronic commerce of big data. The method comprises the following steps that: 101: carrying out a preprocessing operation on the historical behavior data of a user; 102: according to behavior time, carrying out a data division operation on the historical data of the user; 103: marking the historical behavior dataof the user; 104: carrying out a feature engineering construction operation on the historical data of the user; 105: establishing a plurality of machine learning models, and carrying out a model fusion operation; and 106: through an established model, according to the behavior data of the user, predicting whether the user purchases a certain commodity in one future day or not. By use of the method, the historical data of the user is preprocessed and analyzed to extract features, the plurality of machine learning models are established so as to predict a probability for the user to purchase the certain commodity in one future day, and accuracy for a merchant to recommend commodities to the user is improved.