Vol. 41 · No. 3

The Internet of Things (IoT) industry is growing with the high-quality collaboration with Cloud Computing. The data generated by the IoT devices is quite large which can be efficiently stored and processed by the cloud. Further, the scenario like COVID-19 led to an unexpected flood of IoT devices on enabling networks to facilitate online services, which increases the potential threats to the companies fighting to remain operational during the crises. Still, the problem with the IoT devices is their weak security implications because vendors prioritize other factors like energy-saving and efficiency at the cost of security. The Attacker can send malicious requests through the vulnerable IoT device to the network and exploit the cloud in various ways. So, to address this issue, a Game Theoretic Approach to enhance IDS detection (GTA-IDS) in Cloud Environment has been devised that helps the Defender system to be more efficient, accurate in decision-making and save energy. The algorithm based on relative information entropy has been developed to defend against such attacks. The Bayesian Nash Equilibrium (BNE) has been used to make the Defender's strategies and perform actions to maximize its payoffs. The model has been tested on the NSL-KDD dataset and the results have been compared to the existing techniques. The results show that despite efforts made by the Attacker, the Defender always gets a better gain and ultimately eliminates the attack.

This paper proposes a method based on LDA model and Word2Vec for analyzing Microblog users' insurance demands. First of all, we use LDA model to analyze the text data of Microblog user to get their candidate topic. Secondly, we use CBOW model to implement topic word vectorization and use word similarity calculation to expand it. Then we use K-means model to cluster the expanded words and redefine the topic category. Then we use the LDA model to extract the keywords of various insurance information on the ``Pingan Insurance'' website and analyze the possibility of users with different demands to purchase various types of insurance with the help of word vector similarity. Finally, the validity of the method in this paper is verified against Microblog user information. The experimental results show that the accuracy, recall rate and F1 value of the LDA-CBOW extending method have been proposed compared with that of the traditional LDA model, respectively, which proves the feasibility of this method. The results of this paper will help insurance companies to accurately grasp the preferences of Microblog users, understand the potential insurance needs of users timely, and lay a foundation for personalized recommendation of insurance products.

Along with the development of sonar technology, the detection accuracy and stability of sonar have been improved. A large amount of seabed sediment information can be obtained through sonar detection. However, this information is often accompanied by noise interference, resulting in poor quality of the generated images. Moreover, sonar images are different from conventional images. There are single-channel images. The model needs to classify the images according to the texture features in the images. Coupled with the scarcity of sonar data, this makes it difficult to accurately classify the seabed sediment. According to the characteristics of sonar images, we propose the ShuffleNet-DSE which is a classification model based on deep learning. The ShuffleNet-DSE network is based on ShuffleNet-V2, while ensuring the lightweight of the network, it incorporates feature dense connection and Squeeze-and-Excitation (SE) structure channel self-attention. And combined with the sonar image's characteristics, the partial activation function of the model is changed to the Swish. The experimental results show that compared with the traditional machine learning classification method, ShuffleNet-DSE has greatly improved the classification accuracy and the computational cost. Compared with excellent neural network models such as AlexNet, MobileNet-V3, GoogLeNet and ResNet, it is more suitable for sonar image processing.

In the field of image processing, due to the need of expertise and skills in deep-sea biology and the disadvantages of high labor cost and long time consuming, it has always been a difficult task to mark the images of deep-sea benthic organisms. To solve this problem, this paper proposes a new image intelligent labeling algorithm LACP AL (Localization-Aware-Choice and Pseudo Label Active Learning) which is based on Localization-Aware Active Learning. LACP AL is an active learning framework based on Faster R-CNN, it finds the ``valuable'' samples from unlabeled samples by clustering algorithm for every training; it selects hard-to-identify samples for manual annotation and further optimizes the model; and it proposes an improved pseudo-labeling mechanism to expand the training set and improve the model accuracy. According to the publicly available dataset provided by 2020 China Underwater Robot Professional Contest, a series of experiments has been done to verify that our algorithm can achieve higher recognition accuracy with fewer training samples compared with the existing algorithms for Marine benthic image recognition.

Collaborative filtering is a popular recommendation algorithm. It predicts user's interests according to the ratings or behaviour of other users in the system. However, the collaborative filtering recommender system suffers from several major limitations including scalability, sparsity, and cold start. In this paper, a collaborative filtering recommendation approach using radial basis function (RBF) network and power method is proposed. The proposed system has offline and online phases. In the offline phase, the sparse user-item rating matrix is completed by using RBF network based on Cover's theorem on the separability of patterns. RBF network learning is done by unsupervised kernel-based fuzzy c-means clustering algorithm for selecting RBF centers, and supervised gradient descend method for selecting RBF weights. In the offline phase, we predict non-rated items of a user. Then the full rating matrix is used to rank all the users. The ranking is done by solving an eigenvalue problem. This paper overcomes the scalability problem by clustering the users, the sparsity problem by completing the sparse rating matrix, and the new user cold start problem by recommending the top rated items of the high-ranked user. The results of the experiments, on the benchmark data sets, show that the proposed system produces high quality recommendation, in terms of accuracy and quality.

The danmaku text clustering is a hot topic in online video reviews. Given the problem of unsatisfactory clustering accuracy caused by short text and many new words, the danmaku text clustering algorithm based on feature extension and word-pair filtering OBTM is proposed. First, a new-word discovery algorithm based on weight optimization is proposed to retain the features of new words in the danmaku text. Then, the internal information and external knowledge of new words are used to expand the features of the danmaku text for reduced feature sparsity. Furthermore, the OBTM topic model based on word-pair filtering is designed to eliminate noise features. Finally, the Single-Pass algorithm based on cluster center iteration is proposed to obtain the clustering results of topic feature words. Experimental results show that the algorithm proposed in this paper is 13.33

The adoption of Big Data systems by the companies is relatively new, although the data modeling and system design are ages old. Despite the fact that traditional databases are built on solid foundations, they cannot handle the swift and massive flow of data coming from multiple different sources. Herein, NoSQL databases are an inevitable alternative. However, these systems are schemaless compared to traditional databases. It is important to emphasize that schemaless does not mean no-schema which would mean that NoSQL databases do not need modeling. Hence, there is a need for conceptual models to define the data structure in these databases. This paper sheds a light on the importance of the UML in showing how to store Big Data described through meta-models within NoSQL databases. We propose a novel Big Data modeling methodology for NoSQL databases called UML4NoSQL, which is independent of the target system, and taking into account the four Big Data characteristics: Variety, Volume, Velocity, and Veracity (4 V's). The approach relies on the UML blocks with a data-up technique; it starts with a use-case and the class diagram resulting from the understanding of the data at hand and the definition of the developer's strategies while focusing on the user's needs. To illustrate our approach, we take a case study from health care domain. We show that our approach produces designs that can be implemented on NoSQL document-oriented system with respect to Big Data 4 V's.

Day by day, music streaming services grow the volume of data on the internet. To help the users to find songs that fit their interests, music recommender systems can be used to filter a large number of songs according to the preference of the user. However, the context in which the users listen to songs must be taken into account, which justifies the usage of context-aware recommender systems. Although there are some works about context-aware music recommender systems, there is a lack of automatic techniques for extracting contextual information for these systems. Thus, the goal of this work is to propose two methods to acquire contextual information (represented by embeddings) for each song, given the sequence of songs that each user has listened to. The first method, called Metadata-Aware, uses tags and genres to enrich the embeddings with additional information. The second method, called Dual Recurrent Neural Network, uses such a network to improve the embeddings generated from long sequences of songs. The embeddings generated by both methods were evaluated with four context-aware music recommender systems in two datasets. The results showed that the embeddings, obtained by our proposals, present better results than the state-of-the-art method proposed in the literature (in some cases with gains of more than 100

Reliability, longevity, and maintenance costs of electric power industry installations and equipment depend strongly on the extent to which their design reflects relevant environmental factors, such as expected levels of local environmental pollution. These factors guide the choice of specific types of components – insulators, towers, conductors, etc. – and are often estimated through complex and tedious long-term field measurements of pollution deposits. In Slovakia, such field measurements were mandated by the national standard STN 33 0405. This standard was retired in 2015 without replacement, which opened the way for developing alternative and less cumbersome methods. One such alternative is to apply artificial intelligence techniques to atmospheric pollution and other relevant data, which is already routinely monitored and collected in many countries. In this paper, we explore the strength of the relationships between the field measurements performed in various regions of Slovakia according to STN 33 0405 and atmospheric pollution data monitored and collected by the Slovak Hydrometeorological Institute (SHMÚ). The paper is focused on input attributes significance, in relation to output attributes. It represents the first phase of our long-term research aiming at the creation of reliable regression models of local pollution in order to replace the cumbersome field measurements mandated by STN 33 0405.

Wave forecasting approaches based on deep learning techniques have recently made a great progress. In this study, we developed a deep learning model based on Gated Recurrent Unit (GRU) and sequence-to-sequence neural networks (GRUS), to improve the forecasting accuracy of significant wave heights for the Taiwan Strait, where ocean waves and winds own their unique characteristics. The performances of our proposed GRUS model and the other deep learning models based on WaveNet and Long Short-Term Memory (LSTM) were compared by means of wind and wave observations at three buoys in the study area. Model parameters were optimized by means of various model experiments. Performance comparison illustrates that our proposed GRUS model outperforms the other models in 24-hour s forecasting, while the GRUS has extraordinary ability for short-term prediction (prediction horizon is less than 6 h). Moreover, for high wave states prediction (e.g., wave height over 4 m), the GRUS has the strongest prediction ability among the models, in which forecasted wave heights are mostly lower than the corresponding observations.