Earticle

최근 지능화된 보안 패러다임의 변화에 따라, 다양한 정보보안 시스템에서 발생하는 각종 정보를 인공지능 기반 이상탐지에 적용하기 위한 연구가 증가하고 있다. 따라서 본 연구는 로그와 같은 시계열 데이터를 수치형 특성인 벡터로 변환하기 위하여 딥러닝 기반 Word2Vec 모델의 CBOW와 Skip-gram 추론 방식과 동시발생 빈도 기반 통계 방식을 사용하여 공개된 ADFA 시스템콜 데이터에 대하여, 벡터의 차원, 시퀀스 길이 및 윈도우 사이즈를 고려한 다양한 임베딩 벡터로의 변환에 대한 실험을 진행하였다. 또한 임베딩 모델로 생성된 벡터를 입력으로 하는 GRU 기반 이상 탐지 모델 을 통해 탐지 성능뿐만 아니라 사용된 임베딩 방법들의 성능을 비교 평가하였다. 통계 모델에 비해 추론 기반 모델인 Skip-gram이 특정 윈도우 사이즈나 시퀀스 길이에 치우침 없이 좀 더 안정되게(stable) 성능을 유지하여, 시퀀스 데이 터의 각 이벤트들을 임베딩 벡터로 만드는데 더 효과적임을 확인하였다.

Recently, with the change of the intelligent security paradigm, study to apply various information generated from various information security systems to AI-based anomaly detection is increasing. Therefore, in this study, in order to convert log-like time series data into a vector, which is a numerical feature, the CBOW and Skip-gram inference methods of deep learning-based Word2Vec model and statistical method based on the coincidence frequency were used to transform the published ADFA system call data. In relation to this, an experiment was carried out through conversion into various embedding vectors considering the dimension of vector, the length of sequence, and the window size. In addition, the performance of the embedding methods used as well as the detection performance were compared and evaluated through GRU-based anomaly detection model using vectors generated by the embedding model as an input. Compared to the statistical model, it was confirmed that the Skip-gram maintains more stable performance without biasing a specific window size or sequence length, and is more effective in making each event of sequence data into an embedding vector.

본 논문에서는 온라인 뉴스 기사에서 자동으로 추출된 키워드 집합을 활용하여 특정 시점에서의 세부 주제별 토픽을 추출하고 정형화하는 새로운 방법론을 제시한다. 이를 위해서, 우선 다량의 텍스트 집합에 존재하는 개별 단어들의 중요도를 측정할 수 있는 복수의 통계적 가중치 모델들에 대한 비교 실험을 통해 TF-IDF 모델을 선정하였고 이를 활용하여 주요 키워드 집합을 추출하였다. 또한 추출된 키워드들 간의 의미적 연관성을 효과적으로 계산하기 위해서 별도로 수집된 약 1,000,000건 규모의 뉴스 기사를 활용하여 단어 임베딩 벡터 집합을 구성하였다. 추출된 개별 키워드들은 임베딩 벡터 형태로 수치화되고 K-평균 알고리즘을 통해 클러스터링 된다. 최종적으로 도출된 각각의 키워드 군집에 대한 정성적인 심층 분석 결과, 대부분의 군집들이 레이블을 쉽게 부여할 수 있을 정도로 충분한 의미적 집중성을 가진 토픽들로 평가되었다.

In this paper, we propose a new methodology for extracting and formalizing subjective topics at a specific time using a set of keywords extracted automatically from online news articles. To do this, we first extracted a set of keywords by applying TF-IDF methods selected by a series of comparative experiments on various statistical weighting schemes that can measure the importance of individual words in a large set of texts. In order to effectively calculate the semantic relation between extracted keywords, a set of word embedding vectors was constructed by using about 1,000,000 news articles collected separately. Individual keywords extracted were quantified in the form of numerical vectors and clustered by K-means algorithm. As a result of qualitative in-depth analysis of each keyword cluster finally obtained, we witnessed that most of the clusters were evaluated as appropriate topics with sufficient semantic concentration for us to easily assign labels to them.

The enhanced pyramid graph was recently proposed as an interconnection network model in parallel processing for maximizing regularity in pyramid networks. We prove that there are two edge-disjoint Hamiltonian cycles in the enhanced pyramid networks. This investigation demonstrates its superior property in edge fault tolerance. This result is optimal in the sense that the minimum degree of the graph is only four.

Tokenization is the process of segmenting the input text into smaller units of text, and it is a preprocessing task that is mainly performed to improve the efficiency of the machine learning process. Various tokenization methods have been proposed for application in the field of natural language processing, but studies have primarily focused on efficiently segmenting text. Few studies have been conducted on the Korean language to explore what tokenization methods are suitable for document classification task. In this paper, an exploratory study was performed to find the most suitable tokenization method to improve the performance of a representative spatio-temporal document classifier in Korean. For the experiment, a convolutional neural network model was used, and for the final performance comparison, tasks were selected for document classification where performance largely depends on the tokenization method. As a tokenization method for comparative experiments, commonly used Jamo, Character, and Word units were adopted. As a result of the experiment, it was confirmed that the tokenization of word units showed excellent performance in the case of representative spatio-temporal document classification task where the semantic embedding ability of the token itself is important.

This study proposes a novel deep neural network model that can accurately detect objects and their relationships in an image and represent them as a scene graph. The proposed model utilizes several multimodal features, including linguistic features and visual context features, to accurately detect objects and relationships. In addition, in the proposed model, context features are embedded using graph neural networks to depict the dependencies between two related objects in the context feature vector. This study demonstrates the effectiveness of the proposed model through comparative experiments using the Visual Genome benchmark dataset.

A graph is a data structure consisting of nodes and edges between these nodes. Graph embedding is to generate a low dimensional vector for a given graph that best represents the characteristics of the graph. Recently, there have been studies on graph embedding, especially using deep learning techniques. However, until now, most deep learning-based graph embedding techniques have focused on unweighted graphs. Therefore, in this paper, we propose a graph embedding technique for weighted graphs based on long short-term memory (LSTM) autoencoders. Given weighted graphs, we traverse each graph to extract node-weight sequences from the graph. Each node-weight sequence represents a path in the graph consisting of nodes and the weights between these nodes. We then train an LSTM autoencoder on the extracted node-weight sequences and encode each nodeweight sequence into a fixed-length vector using the trained LSTM autoencoder. Finally, for each graph, we collect the encoding vectors obtained from the graph and combine them to generate the final embedding vector for the graph. These embedding vectors can be used to classify weighted graphs or to search for similar weighted graphs. The experiments on synthetic and real datasets show that the proposed method is effective in measuring the similarity between weighted graphs.

As the cyberattack subject has expanded from individual to group, attack patterns have become a complicated form of cyber campaigns. Although detecting the attack groups that operated the cyber campaigns is an important issue, complex methods such as deep learning are difficult to use due to the lack of campaign data. This paper proposes Camp2Vec, a lightweight statistics-based embedding for cyber campaigns, enabling attack group detection. The proposed method models a relationship between a campaign and techniques in the ATT&CK® framework as a document and words. Experimental results with expert-labeled datasets prove that Camp2Vec identifies representative attack groups successfully.

As the major source of information, digital images play an indispensable role in our lives. However, with the development of image processing techniques, people can optionally retouch or even forge an image by using image processing software. Therefore, the authenticity and integrity of digital images are facing severe challenge. To resolve this issue, the fragile watermarking schemes for image authentication have been proposed. According to different purposes, the fragile watermarking can be divided into two categories: fragile watermarking for tamper localization and fragile watermarking with recovery ability. The fragile watermarking for image tamper localization can only identify and locate the tampered regions, but it cannot further restore the modified regions. In some cases, image recovery for tampered regions is very essential. Generally, the fragile watermarking for image authentication and recovery includes three procedures: watermark generation and embedding, tamper localization, and image self-recovery. In this article, we make a review on self-embedding fragile watermarking methods. The basic model and the evaluation indexes of this watermarking scheme are presented in this paper. Some related works proposed in recent years and their advantages and disadvantages are described in detail to help the future research in this field. Based on the analysis, we give the future research prospects and suggestions in the end.

As one of the most significant challenges in the virtual data center, the virtual data center embedding has attracted extensive attention from researchers. The existing research works mainly focus on how to design algorithms to increase operating revenue. However, they ignore the energy consumption issue of the physical data center in virtual data center embedding. In this paper, we focus on studying the energy-aware virtual data center embedding problem. Specifically, we first propose an energy consumption model. It includes the energy consumption models of the virtual machine node and the virtual switch node, aiming to quantitatively measure the energy consumption in virtual data center embedding. Based on such a model, we propose two algorithms regarding virtual data center embedding: one is heuristic, and the other is based on particle swarm optimization. The second algorithm provides a better solution to virtual data center embedding by leveraging the evolution process of particle swarm optimization. Finally, experiment results show that our proposed algorithms can effectively save energy while guaranteeing the embedding success rate.

Virtual network embedding plays a vital role in network virtualization, as it determines the deployment and connection of virtual networks to the physical network in the 5G and beyond. An efficient virtual network embedding algorithm is essential to ensure that virtual networks are embedded in a way that meets the performance, security, and resource requirements of the virtual networks and their users. The integration of reinforcement learning with virtual network embedding can lead to more intelligent and efficient network management, which can enhance the performance of large-scale networked systems. Reinforcement learning has the potential to improve and overcome some limitations of traditional algorithms, such as the need for prior knowledge of network conditions and the difficulty in dealing with non-linear and dynamic network environments. Therefore, we conducted this survey to provide a comprehensive overview and examine potential future directions for the optimal reinforcement learning-based virtual network embedding solutions. However, applying reinforcement learning directly to virtual network embedding is a challenging task that requires further research and study. Additionally, it encourages researchers to examine the potential of reinforcement learning in virtual network embedding, identify the challenges for its application, and cover various factors related to the reinforcement learning application in virtual network embedding, including motivations, performance metrics, and challenges.

Aspect-based sentiment analysis is to discover the sentiment polarity towards an aspect from user-generated natural language. So far, most of the methods only use the implicit position information of the aspect in the context, instead of directly utilizing the position relationship between the aspect and the sentiment terms. In fact, neighboring words of the aspect terms should be given more attention than other words in the context. This paper studies the influence of different position embedding methods on the sentimental polarities of given aspects, and proposes a position embedding interactive attention network based on a long short-term memory network. Firstly, it uses the position information of the context simultaneously in the input layer and the attention layer. Secondly, it mines the importance of different context words for the aspect with the interactive attention mechanism. Finally, it generates a valid representation of the aspect and the context for sentiment classification. The model which has been posed was evaluated on the datasets of the Semantic Evaluation 2014. Compared with other baseline models, the accuracy of our model increases by about 2% on the restaurant dataset and 1% on the laptop dataset.

High-utility itemset mining (HUIM) is a dominant technology that enables enterprises to make real-time decisions, including supply chain management, customer segmentation, and business analytics. However, classical support value-driven Apriori solutions are confined and unable to meet real-time enterprise demands, especially for large amounts of input data. This study introduces a groundbreaking model for top-N high utility itemset mining in real-time enterprise applications. Unlike traditional Apriori-based solutions, the proposed convolutional sequential embedding metrics-driven cosine-similarity-based multilayer perception learning model leverages global and contextual features, including semantic attributes, for enhanced top-N recommendations over sequential transactions. The MATLAB-based simulations of the model on diverse datasets, demonstrated an impressive precision (0.5632), mean absolute error (MAE) (0.7610), hit rate (HR)@K (0.5720), and normalized discounted cumulative gain (NDCG)@K (0.4268). The average MAE across different datasets and latent dimensions was 0.608. Additionally, the model achieved remarkable cumulative accuracy and precision of 97.94% and 97.04% in performance, respectively, surpassing existing state-of-the-art models. This affirms the robustness and effectiveness of the proposed model in real-time enterprise scenarios.

Owing to the hyper-connectivity of convergence environments, the Internet of Blended Environments (IoBE) has emerged As a result, the environments and architectures in which cyber-security threats can occur have steadily diversified leading to an increase in security incidents. However, existing detection systems lack correlation analysis and thus cannot detect the corresponding diverse attack paths and attack chains effectively. In this paper, we propose a data prediction technique in which knowledge graph embedding technology is applied to predict blended threats in complex environments such as the IoBE. We also verify the performance of the proposed technique.

This paper proposes an intelligent sensing framework for Internet-of-Things platforms, where sensor measurements stem from multiple causes. Sensors are selectively chosen for data collection to identify the cause with partial measurements. We employ variational deep embedding, a generative model capable of clustering and generation, to identify causes, cluster measurements accordingly, and determine causes for estimating complete measurements from partial data. These estimates aid in efficient sensor selection for data collection. Results demonstrate early and reliable cause sensing and complete measurement estimation using the proposed framework.

Implementation of faster pairing calculation is the basis of efficient pairing-based cryptographic protocol implementation. Generally, pairing is a costly operation carried out over the extension field of degree $k{\geq}12$. But the twist property of the pairing friendly curve allows us to calculate pairing over the sub-field twisted curve, where the extension degree becomes k/d and twist degree d = 2, 3, 4, 6. The calculation cost is reduced substantially by twisting but it makes the discrete logarithm problem easier if the curve parameters are not carefully chosen. Therefore, this paper considers the most recent parameters setting presented by Barbulescu and Duquesne [1] for pairing-based cryptography; that are secure enough for 128-bit security level; to explicitly show the quartic twist (d = 4) and sextic twist (d = 6) mapping between the isomorphic rational point groups for KSS (Kachisa-Schaefer-Scott) curve of embedding degree k = 16 and k = 18, receptively. This paper also evaluates the performance enhancement of the obtained twisted mapping by comparing the elliptic curve scalar multiplications.

Anomaly detection systems for Industrial Control System (ICS) cybersecurity are designed to identify irregularities in network packets or operational data. However, they cannot detect attacks like Stuxnet, which physically injects malicious control logic. While existing studies on control logic modulation address this issue, they rely on separate storage and produce false positives. To overcome these limitations, this paper proposes an anomaly detection method that embeds PLC control logic, preserving its structure. By training the model on this embedded control logic, it learns to detect anomalies effectively. Experiments using the PLC control logic from a power plant's water treatment system confirmed that the proposed method successfully detects anomalous control logic.

This study compares two embedding-based natural language processing techniques—Sentence-BERT (SBERT) combined with HDBSCAN clustering and BERTopic modeling—for detecting complex emotions in short Korean online comments. Using 33,531 comments collected from a YouTube relationship counseling channel, we examined how each method captures nuanced and overlapping sentiments such as affection, avoidance, and conflict. Both models used identical SBERT embeddings and UMAP-based dimensionality reduction, and their clustering performance was quantitatively evaluated using Silhouette Score, Davies–Bouldin Index (DBI), and Calinski–Harabasz Index (CHI). The results show that BERTopic achieved higher coherence and clearer topic boundaries (Silhouette = 0.40, DBI = 0.85, CHI = 15,157) compared to SBERT–HDBSCAN (Silhouette = –0.23, DBI = 1.49, CHI = 1,230). Although both methods yielded high noise ratios due to the leaf-based density clustering, BERTopic effectively reclassified semantically relevant comments through its ClassTF-IDF weighting, improving topic stability and interpretability. These findings suggest that BERTopic provides superior performance for analyzing short, emotion-rich Korean text and offers methodological insight for future sentiment analysis research. This electronic document is a “live” template and already defines the components of your paper [title, text, heads, etc.] in its style sheet.

Against the backdrop of global Go development and China’s sports governance transformation, traditional elite-oriented models (national system and Go dojos道场) fail to meet local demands for popularization, cultural inheritance, and industrial linkage. In China，as Go promotion shifts to local cities, constructing a comprehensive system becomes critical. Luoyang, a “Hundred-Dan Go City,” offers rich empirical data with its 43 professional players and 152 total dan ranks by 2023. Existing studies focus on competition, cultural value, or single cases, lacking analysis of multi-factor synergy and the interaction between cultural embedding and market drive. This study addresses three questions: how cultural embedding supports market drive; whether Luoyang’s multi-source funding model is sustainable; and how to balance popularization and elite cultivation. Using case study, in-depth interview, bibliometrics, and cost-benefit analysis, this study employs cultural embedding, market drive, and sports ecosystem theories, proposing a “Culture-Market-Talent” triangular synergy model. Findings show that cultural embedding and market drive synergize to form a positive cycle; Luoyang’s multi-stakeholder collaboration achieves an ecological closed-loop; the model is universally applicable. The study enriches sports ecosystem theory, provides practical solutions for local cities, and offers policy references for integrating sports, culture, and tourism. Limitations include single-case bias and historical data gaps; future research could expand case scope and explore digital integration.

Automatic question-answering is a classical problem in natural language processing, which aims at designing systems that can automatically answer a question, in the same way as human does. The need to query information content available in various formats including structured and unstructured data has become increasingly important. Thus, Question Answering Systems (QAS) are essential to satisfy this need. QAS aim at satisfying users who are looking to answer a specific question in natural language.Moreover, it is a representative of open domain QA systems, where the answer selection process leans on syntactic and semantic similarities between the question and the answering text snippets. Such approach is specifically oriented to languages with fine grained syntactic and morphologic features that help to guide the correct QA match. Furthermore, word and sentence embedding have become an essential part of any Deep-Learning-based natural language processing systems as they encode words and sentences in fixed-length dense vectors to drastically improve the processing of textual data. The paper will concentrate on incorporating the sentence embedding with its various techniques like Infersent, ElMo and BERT in the construction of Question Answering systems, and also it can be used in game play.