Mario M. Kubek
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Mario M. Kubek has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Herwig Unger
FernUniversität in Hagen
7 shared publications - Armin Robert Mikler
Georgia State University
3 shared publications - Emma Louise McDaniel
Georgia State Universrity, USA
2 shared publications - Shiraj Pokharel
Georgia State University
2 shared publications - Thomas Böhme
Ilmenau University of Technology
2 shared publications - Anoop Lashiyal
Georgia State University
1 shared publication
Latest Publications
- 2 July 2026
- Published by Springer Nature
- p. 1170-1186
Interpretable Adverse Event Detection Using a Neuro-Symbolic and Graph-Based Approach
- 2 July 2026
- Published by Springer Nature
- p. 1170-1186
Interpretable Adverse Event Detection Using a Neuro-Symbolic and Graph-Based Approach
- 14 May 2025
- Published by Springer Nature
- p. 41-50
Distinguishing AI-Generated Text from Human-Written Text Using Machine Learning
- 14 May 2025
- Published by Springer Nature
- p. 41-50
Distinguishing AI-Generated Text from Human-Written Text Using Machine Learning
- Open Access
On Self-improving Token Embeddings
This article introduces a novel and fast method for refining pre-trained static word or, more generally, token embeddings. By incorporating the embeddings of neighboring tokens in text corpora, it continuously updates the representation of each token, including those without pre-assigned embeddings. This approach effectively addresses the out-of-vocabulary problem, too. Operating independently of large language models and shallow neural networks, it enables versatile applications such as corpus exploration, conceptual search, and word sense disambiguation. The method is designed to enhance token representations within topically homogeneous corpora, where the vocabulary is restricted to a specific domain, resulting in more meaningful embeddings compared to general-purpose pre-trained vectors. As an example, the methodology is applied to explore storm events and their impacts on infrastructure and communities using narratives from a subset of the NOAA Storm Events database. The article also demonstrates how the approach improves the representation of storm-related terms over time, providing valuable insights into the evolving nature of disaster narratives.
Open AccessOn Self-improving Token Embeddings
This article introduces a novel and fast method for refining pre-trained static word or, more generally, token embeddings. By incorporating the embeddings of neighboring tokens in text corpora, it continuously updates the representation of each token, including those without pre-assigned embeddings. This approach effectively addresses the out-of-vocabulary problem, too. Operating independently of large language models and shallow neural networks, it enables versatile applications such as corpus exploration, conceptual search, and word sense disambiguation. The method is designed to enhance token representations within topically homogeneous corpora, where the vocabulary is restricted to a specific domain, resulting in more meaningful embeddings compared to general-purpose pre-trained vectors. As an example, the methodology is applied to explore storm events and their impacts on infrastructure and communities using narratives from a subset of the NOAA Storm Events database. The article also demonstrates how the approach improves the representation of storm-related terms over time, providing valuable insights into the evolving nature of disaster narratives.
- Open Access
- 13 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 374-379
Reducing Labeling Costs in Sentiment Analysis via Semi-Supervised Learning
- 13 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 374-379
Open AccessReducing Labeling Costs in Sentiment Analysis via Semi-Supervised Learning
- Open Access
Reducing Labeling Costs in Sentiment Analysis via Semi-Supervised Learning
Labeling datasets is a noteworthy challenge in machine learning, both in terms of cost and time. This research, however, leverages an efficient answer. By exploring label propagation in semi-supervised learning, we can significantly reduce the number of labels required compared to traditional methods. We employ a transductive label propagation method based on the manifold assumption for text classification. Our approach utilizes a graph-based method to generate pseudo-labels for unlabeled data for the text classification task, which are then used to train deep neural networks. By extending labels based on cosine proximity within a nearest neighbor graph from network embeddings, we combine unlabeled data into supervised learning, thereby reducing labeling costs. Based on previous successes in other domains, this study builds and evaluates this approach's effectiveness in sentiment analysis, presenting insights into semi-supervised learning.
Open AccessReducing Labeling Costs in Sentiment Analysis via Semi-Supervised Learning
Labeling datasets is a noteworthy challenge in machine learning, both in terms of cost and time. This research, however, leverages an efficient answer. By exploring label propagation in semi-supervised learning, we can significantly reduce the number of labels required compared to traditional methods. We employ a transductive label propagation method based on the manifold assumption for text classification. Our approach utilizes a graph-based method to generate pseudo-labels for unlabeled data for the text classification task, which are then used to train deep neural networks. By extending labels based on cosine proximity within a nearest neighbor graph from network embeddings, we combine unlabeled data into supervised learning, thereby reducing labeling costs. Based on previous successes in other domains, this study builds and evaluates this approach's effectiveness in sentiment analysis, presenting insights into semi-supervised learning.
- Open Access
- 1 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 80-95
Look What’s There! Utilizing the Internet’s Existing Data for Censorship Circumvention with OPPRESSION
An ongoing challenge in censorship circumvention is optimizing the stealthiness of communications, enabled by covert channels. Recently, a new variant called history covert channels has been proposed. Instead of modifying or mimicking legitimate data, such channels solely point to observed data matching secret information. This approach reduces the amount of secret data a sender explicitly must transfer and thus limits detectability. However, the only published history channel is only suitable for special scenarios due to severe limitations in terms of bandwidth. We propose a significant performance enhancement of history covert channels that allows their use in real-world scenarios through utilizing the content of online social media and online archives. Our approach, which we call OPPRESSION (Open-knowledge Compression), takes advantage of the massive amounts of textual data on the Internet that can be referenced by short pointer messages. Broadly, OPPRESSION can be considered a novel encoding strategy for censorship circumvention. We further present and evaluate our open source proof-of-concept implementation of OPPRESSION that can transfer secret data by pointing to popular online media, such as Twitter (now "X"), news websites, Wikipedia entries, and online books. The pointer itself is transmitted through existing censorship circumvention systems. Our approach minimizes the amount of traffic to be concealed in comparison to existing works, even in comparison to compression.
- 1 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 80-95
Open AccessLook What’s There! Utilizing the Internet’s Existing Data for Censorship Circumvention with OPPRESSION
An ongoing challenge in censorship circumvention is optimizing the stealthiness of communications, enabled by covert channels. Recently, a new variant called history covert channels has been proposed. Instead of modifying or mimicking legitimate data, such channels solely point to observed data matching secret information. This approach reduces the amount of secret data a sender explicitly must transfer and thus limits detectability. However, the only published history channel is only suitable for special scenarios due to severe limitations in terms of bandwidth. We propose a significant performance enhancement of history covert channels that allows their use in real-world scenarios through utilizing the content of online social media and online archives. Our approach, which we call OPPRESSION (Open-knowledge Compression), takes advantage of the massive amounts of textual data on the Internet that can be referenced by short pointer messages. Broadly, OPPRESSION can be considered a novel encoding strategy for censorship circumvention. We further present and evaluate our open source proof-of-concept implementation of OPPRESSION that can transfer secret data by pointing to popular online media, such as Twitter (now "X"), news websites, Wikipedia entries, and online books. The pointer itself is transmitted through existing censorship circumvention systems. Our approach minimizes the amount of traffic to be concealed in comparison to existing works, even in comparison to compression.
- 31 May 2024
- Published by Springer Nature
- p. 121-131
WebMap - Large Language Model-assisted Semantic Link Induction in the Web
- 31 May 2024
- Published by Springer Nature
- p. 121-131
WebMap - Large Language Model-assisted Semantic Link Induction in the Web
- Open Access
- 16 December 2022
- Published by Association for Computing Machinery (ACM)
- p. 49-56
Building term hierarchies using graph-based clustering
Classical tasks of a librarian, such as screening and categorizing new documents based on their content, are increasingly replaced by search engines or through the use of cataloging software. A first overview of a corpus topical orientation can be achieved by combining graph-based search engines and clustering methods. Existing classical clustering methods, however, often require an a priori specification of the desired number of clusters to be output and do not consider term relationships in graphs, which is deficient from a practical point of view. Therefore, fully unsupervised graph-based clustering approaches at the term level offer new possibilities that mitigate these shortcomings. Within this work, a set of novel graph-based clustering algorithms have been developed. The hierarchical clustering algorithm (HCA) forms term hierarchies by iteratively isolating nodes of a given co-occurrence graph based on the evaluation of the edge weight between the nodes. Based on the co-occurrence graph inherent relationships of terms, a new graph is built agglomerative forming individual term clusters of related terms. The feasibility of the outlined methods for text analysis is shown.
- 16 December 2022
- Published by Association for Computing Machinery (ACM)
- p. 49-56
Open AccessBuilding term hierarchies using graph-based clustering
Classical tasks of a librarian, such as screening and categorizing new documents based on their content, are increasingly replaced by search engines or through the use of cataloging software. A first overview of a corpus topical orientation can be achieved by combining graph-based search engines and clustering methods. Existing classical clustering methods, however, often require an a priori specification of the desired number of clusters to be output and do not consider term relationships in graphs, which is deficient from a practical point of view. Therefore, fully unsupervised graph-based clustering approaches at the term level offer new possibilities that mitigate these shortcomings. Within this work, a set of novel graph-based clustering algorithms have been developed. The hierarchical clustering algorithm (HCA) forms term hierarchies by iteratively isolating nodes of a given co-occurrence graph based on the evaluation of the edge weight between the nodes. Based on the co-occurrence graph inherent relationships of terms, a new graph is built agglomerative forming individual term clusters of related terms. The feasibility of the outlined methods for text analysis is shown.
- 6 June 2022
- Published by Springer Nature
- p. 6-10
Brain-Inspired Approaches to Natural Language Processing and Explainable Artificial Intelligence
- 6 June 2022
- Published by Springer Nature
- p. 6-10
Brain-Inspired Approaches to Natural Language Processing and Explainable Artificial Intelligence